Last Sunday, I spent the afternoon at Shopify’s Ottawa office alongside a room full of founders and builders. By the end of the day, 8 of them had taken the stage to demo what they’d been working on. That’s the whole idea behind Builder Sundays, and it’s why I think every early-stage founder within driving distance should experience one at least once.

What Shopify Builder Sundays Actually Is

Builder Sundays is a free, drop-in deep-work session that Shopify runs every Sunday from 12pm to 5pm in Toronto, Montreal, and Ottawa. The format is simple: you bring a laptop and something you want to make progress on, work alongside other builders, and let the energy of the room do the rest. Deep work runs until around 3:30pm, followed by open demos and an hour of networking. Spots for the following Sunday are released Wednesdays at noon ET, and you register individually, including for any guests you bring.

Why Builders Should Give It a Shot

Apart from connecting with other builders in the room, Builder Sundays offer 3 things founders absolutely need to grow.

1. Practice Your Pitch

You don’t sign up to pitch ahead of time. If you want to demo, you put your name on the board that day on a first-come, first-served basis, then take the stage and show what you managed to build.

There’s no polish requirement, no committee, and no gatekeeping. You simply get up and pitch. And that’s what makes it so useful: you’re practicing in a low-stakes environment where you can experiment, get comfortable on stage, and learn what resonates.

2. Build the Habit of Shipping

A tight deadline forces you to cut scope, make decisions quickly, and actually finish something. That’s a muscle many founders lose while endlessly refining roadmaps, tweaking ideas, and planning for the “right” time to launch.

3. Get Real Feedback

You also get something that’s genuinely hard to replicate: a room full of people willing to give you immediate, honest feedback. They’ll ask questions, poke holes in your idea, test your assumptions, and point out what’s missing, all before you’ve spent weeks or months building in isolation.

8 Pitches From Builder Sundays

So you get the vibe, here’s a quick tour of the eight pitches that took the stage that afternoon.

Worth It: The Anti-Diet App Built in 2 Hours

Brandon and Brian, two students heading into their first year of university, set themselves a simple challenge: build something end-to-end in two hours. The idea, presentation, and product all started around 1pm that afternoon.

Their creation, Worth It, takes the opposite approach to typical health-food apps. Instead of finding the lowest-calorie option nearby, it pulls the menus of surrounding restaurants, runs them through AI, and surfaces the highest-calorie, most worth-it foods for your cheat day.

They pointed at an $800 billion diet and wellness industry and 88 million people tracking their macros (protein, carbohydrates, and fat), then argued the disciplined crowd is exactly who wants a proper break when they take one.

The demo found their location on O’Connor Street and served up a double bacon burger with cheddar and fries within 5km.

Built with React and TypeScript and powered by Anthropic, the app cost them just 32 cents to test. They were clear that the goal wasn’t to build a serious business, but to challenge themselves to go from idea to working product, get comfortable presenting, and see how far they could push 2 hours of focused work.

The room’s reaction summed it up: the amount of fidelity they reached in two hours was pretty crazy.

Tandem Robotics: Farm Robots Built to Work Alongside Farmers

Ming and his team at Tandem Robotics started with the farmer, not the robot. After talking to growers at the Lansdowne Farmers Market, he heard the same problem over and over: they need more labor, but it’s getting harder and more expensive to find.

According to the numbers he shared, 44% of farmers can’t find the workers they need, while labor accounts for 38% of farm operating revenue. The physical toll is significant too, with 58% of farm workers experiencing lower back pain within their first 12 months.

His answer is a collaborative robot designed to work alongside farmers, not replace them. The robot handles the heavy lifting and repetitive work, reducing the amount of walking, carrying, bending, and physical strain while allowing farmers to focus on what they know best. It can carry up to 300 pounds, and in early trials, the team found it could transplant 2.5 times faster and de-weed roughly four times faster than traditional methods.

The robot has already been tested on more than 5 farms, many within 10km of Ottawa. One detail that really stood out: farmers reported zero of their usual pain while working with the robot. And when the team takes a robot away for maintenance or to move it to another farm, their biggest concern is whether they can manage without it.

The pitch wasn’t about replacing farmers. It was about giving them the tools to keep doing what they love, for longer, with less physical strain.

Keating: AI That Teaches You Without Melting Your Brain

Keith opened with a simple question: who here worries about losing skills by relying too heavily on AI? A lot of hands went up.

That concern is exactly what led him to build Keating, an AI-powered teaching system designed to help people learn with AI without becoming dependent on it.

Keith has been building Keating for 3 months, and the project has already passed 2,000 npm downloads, with more than 500 weekly downloads.

Instead of giving you the usual wall of chatbot text, Keating turns AI into a more structured learning experience, with flashcards, visualizations, quizzes, and courses that users can build and share. The system also adapts to how you learn and can evolve its own code to improve the experience.

Keating is open source and runs in the browser, with your data staying under your control. You can bring your own API key, connect it to OpenAI, Anthropic, or Ollama, or even run certain models locally in the browser using WebGPU. Keith said his own AI usage for learning cost him only about $20 over three months.

The product has come a long way since Keith first demoed an early version at Builder Sundays 3 months ago. What started as an experiment has turned into a genuinely interesting take on one of AI’s biggest use cases: not just getting answers, but actually learning.

From $1,000 in 3D-Printed Signs to a Finance Student Hub

Frank came up to show 2 things he’s been building. The first is a 3D printing business he runs on Facebook Marketplace and Etsy, documented on TikTok, that cleared over $1,000 in revenue in its first month. His edge was leaning into what small shops do best: custom letters and signs where customers pick the exact thickness, color, and size, plus catching a couple of viral product trends and printing risers for trading-card displays. With just one printer and a handful of materials, he was typically shipping orders in under a week.

The second is The Finance Student, a platform he built after struggling to find a single place for finance students to discover corporate networking events, job opportunities, and other resources. And, in true Builder Sundays fashion, Frank had a little help from Claude, which he called “a GOAT.”

Ottawa is live first, with a hiring section that tracks live application deadlines across banking, wealth management, accounting, and internships.

Frank ended by asking the room for help with 2 challenges: automating the process of finding and posting relevant events, and figuring out how to market the platform.

This was a perfect example of what Builder Sundays are about: build something around a problem you’ve personally experienced, then bring it to a room full of people who can help you figure out what comes next.

Colive: A Better Way to Find a Roommate

Osman came up with a simple observation: finding a roommate can be a nightmare, especially for students and newcomers. Listings are scattered across websites, Facebook Marketplace, word of mouth, and community groups, and even then, you’re never quite sure who you’re dealing with.

His solution is Colive, a roommate-matching platform he described as “the Tinder of roommates.” The idea is to match people based on lifestyle and behavioral preferences, so you can find someone who is not only looking for a room, but actually compatible with you. Users go through a verification process, and the platform also includes a community feed where newcomers and students can share advice, experiences, and local information.

Trust is a major part of the product. Users are verified through their email and Canadian phone number, with additional identity checks being tested as the platform develops. Osman emphasized that Colive is currently in beta and that he’s actively using feedback from early users to shape the product.

Opero: One Workspace to Run Your Entire Business

As an accessibility consultant and software developer, this founder knows firsthand how fragmented running a small business can become. A growing business can easily end up juggling 11 different jobs, 11 different tools, 11 logins, and countless data exports just to keep everything moving. Marketing, sales, invoicing, collections, project management, and operations all live in separate systems that rarely communicate with each other.

His solution is Opero, a unified workplace platform designed to bring those core business functions into one workspace, with one login and one bill. Instead of simply putting different tools side by side, Opero focuses on connecting the underlying data, so things like clients, quotes, invoices, projects, and communications can work together in context.

The platform has been in development for 2 years and already has 79 active workspaces, serving solo operators, independent workers, and small teams. The founder built the platform himself using React, TypeScript, PHP, and Python, initially coding it before incorporating AI agents into the development process.

His pitch was less about showing off another piece of software and more about challenging other founders in the room: if you’re building a business, how many tools are you going to need to run it?

The end goal is simple: give Canadian businesses one place to run their entire operation.

An AI Gym Coach That Roasts You Into Shape

Sarbjeet came to Builder Sundays with a product he had actually built years ago. His original workout app had attracted more than 3,000 users, but he eventually stopped maintaining it because keeping the codebase going manually was too much work. This year, he decided to bring it back with a new twist: AI-powered coaching.

The app tracks workouts, progressive overload, strength, body weight, and training streaks, while remembering what you lifted in previous sessions so you don’t have to. It can also swap exercises when a machine is occupied, help design workouts, and give you feedback after each set. And if you need a little extra motivation, you can turn on “Roastmaster,” an AI coach that will call you out when you’re slacking.

The app uses a bring-your-own-key model, with support for models including OpenAI, Anthropic, and DeepSeek. It also keeps workout data and AI conversations on the user’s phone rather than storing them on a central server.

What started as a workout tracker is now becoming something closer to a personal AI training partner: one that remembers your progress, adapts to your workouts, and, if you choose Roastmaster, isn’t afraid to roast you when you start slacking.

18, Ambitious, and Just Getting Started

David’s pitch was a little different from the others. Instead of demoing a product, the 18-year-old got on stage to share his story and talk about where he’s headed next.

After moving from Seoul to Canada as a child, David struggled with the adjustment, and later found himself dealing with low confidence and isolation during high school. Working out, meeting new people, and pushing himself outside his comfort zone helped him turn things around. When document complications prevented him from attending his dream university just after high school, he decided to make the most of the unexpected gap year.

View this post on Instagram

A post shared by David (@__yodavids)

He took on part-time jobs, worked as a tutor, trained in Muay Thai and powerlifting, and got involved with 4 different startups. He also helped generate more than 100,000 views from a campaign at Carleton University and co-founded Pitch Tank Ottawa, where he helped sell out 4 events.

Now, he’s heading to Western University to study Management and Organizational Studies, with an advanced entry opportunity to the Ivey Business School. Rather than trying to map out the next five years, David said he wants to stay open to new experiences and opportunities. But he does know one thing for sure: “I’m going to be a millionaire.”

The Point Isn’t to Have It All Figured Out

If there was one thing I took away from Shopify Builder Sundays, it’s that none of these founders waited for the perfect idea, the perfect product, or the perfect moment to start.

They came with different ideas, different levels of experience, and very different reasons for building. But they all had one thing in common: they were willing to put something into the world before they had all the answers.

And that’s what makes Builder Sundays worth experiencing.

Some of these ideas will become companies. Some will pivot. And some will change completely after a few more Sundays. But each founder now has momentum, feedback, and a room full of people who’ve seen what they’re building.

For anyone building from Ottawa, that’s a pretty powerful place to start. You don’t need a pitch deck, a team, a five-year plan, or a billion-dollar idea. You just need something you’re willing to build, and of course, the courage to show it to others.

About the author

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post What Founders Ship at Shopify Builder Sundays in Ottawa appeared first on BYVI.

I flew to San Francisco for YC Startup School 2026 with something I didn’t have a year ago: a product real people actually use. Having a real product changed everything about this trip.

Here’s what the sessions, the hallway run-ins, and 6,000 founders in one place did to the way I think about building, and why I came home to Ottawa readier to go bigger on my own terms.

Why I live on Twitter/X, not LinkedIn

Two lines from the stage stuck with me before I’d even left the building. Boris Cherny joked, “Maybe don’t listen to the LinkedIn influencers.”

Alexandr Wang’s session put it plainer: “Top advice: ignore LinkedIn.” My takeaway landed fast: more Twitter/X, less LinkedIn.

Strange thing to write in a piece you might be reading on LinkedIn, I know. But it’s the honest version.

The reason I’ve been able to track AI closely and pick up new tools quickly is Twitter/X. It’s the platform I’ve used longest and most consistently, and it’s where people share what they’re trying before it gets sanded into polished content. I spot a tool early, then I find out if it’s any good by throwing it at a problem I already have.

The room didn’t make me a builder

The bigger takeaway from the trip, though, was confidence. And that’s really what I want to hand to anyone reading this who doesn’t have the “right” background and is wondering whether to build anyway.

Startup School didn’t turn me into a builder. Building had already changed who I was by the time I showed up. Over the past year, AI stretched what I could do alone. The people I met stretched what I thought was worth doing.

What my YC Startup School experience actually gave me

Last year I applied to Startup School and got rejected. I flew to the Bay Area anyway, went to the events around it, and found my way into the community.

That trip forced me to face something: I had years of product experience and pages of ideas, but nothing I could point at and say, ‘I made this.’

One year later, I came back with Joe Speaking. It started as a fix for my own English-speaking problem. I was user number one, and it got me the language score I needed. Since then it’s reached more than 1,700 users through purely organic growth, 130 of them paying, and it became my first iOS app.

At Startup School I met people from top universities, research labs, and big tech. Plenty were younger than me with resumes that read better than mine on paper. A year earlier, that would’ve had me quietly questioning whether I belonged. This time I just felt proud. I didn’t need a famous school, company, title, or investor to explain who I was, because I’d put something real into the world and people were using it.

Building became my credential.

Startup School didn’t give me that. It helped me see I’d already earned it. While I was drafting this, Paul Graham posted the line that sums it up: confidence comes from evidence.

Here’s my evidence from one year. I went from product person to someone who can turn a personal problem into a shipped product, then debug it, distribute it, support it, and improve it.

Even while wandering around San Francisco, I shipped 1 iOS release and 3 web releases remotely, acted on feedback from 3 users, and burned through roughly 4.4 billion tokens. That pace still feels a little unreal to me.

There’s no magic prompt

Boris Cherny’s session explained why some people get so much more out of tools like Claude Code. There’s no magic prompt. His advice was to treat the model empirically: give it hard work, give it a way to check its own output, watch where it breaks, then adjust the prompt, context, tools, or workflow. Then run it again. The loop is the skill, not any single line you paste in.

That’s exactly how I got better. Not by hoarding prompt templates, but by testing tools against problems I genuinely needed to solve and letting the results grade me. A template promises a shortcut, but the model, the task, and the tools keep shifting, so what worked last week quietly stops working.

Running that loop against my own real problems taught me what to actually change when the output missed, and the feedback was honest because it was mine.

AI didn’t take my job off my plate. I still had to decide what to build, understand users, judge the output, and make the tradeoffs. What it did was amplify my agency.

One person still isn’t enough

The trip reminded me that one person still isn’t enough for everything. Tools made me faster. The people I met changed what I thought was possible, and brought the judgment, trust, and perspective no tool can provide.

Permission to build what you actually want

Most conversations gravitated to giant markets, hockey-stick growth, fundraising, and whatever’s trending. That energy is part of what makes the place worth it. My path just looks different.

Joe Speaking is a product, an experiment, and a way to learn all at once. I built it because I needed it, not because the problem was fashionable or easy to pitch. I literally put my own name on it.

Because AI has dropped the cost of execution, I have more room to follow that instinct. But more freedom raises the stakes on choosing a direction.

AI can make me faster. It can’t tell me what to care about.

What I took from Peter Steinberger’s talk was permission:

Afterward I caught a few minutes with a YC partner, walked through my background, why I built Joe Speaking, and what I want to explore next. The product has grown further than I expected and become genuinely useful to a group of people. But I’m not going to jam it into a story built for YC or fundraising before I understand what the bigger company should even be. I might apply later, when the product, the opportunity, and my ambition actually line up. For now I’m fine saying: not yet.

That’s probably the cleanest summary of my trip. Call it a refusal to perform ambition for other people. I’m plenty ambitious. I just won’t fake a version of it to fit someone’s slide.

Building for myself means starting from a problem I understand deeply, listening to users, and seeing how far it goes before I call it something bigger. Feeling a little out of place might’ve been part of finally finding my own lane.

> be joe.

> always build for myself and be my own first user. even the product name has “joe” in it.

> build to solve my own problems, not to optimize for yc or fundraising.

> twitter is still on my phone dock.

> on my own track.

Thank you @steipete ! pic.twitter.com/octq3iwkQq

— Joe Hu (@hubeiqiao) July 26, 2026

Staying on your track doesn’t mean thinking small

Authenticity can turn into an excuse, though. “I’m staying on my own track” should never quietly mean “I’m scared to think bigger.” Startup School pushed me here, hard.

Alexandr Wang framed this as a once-in-a-civilization moment. The question I walked out with: what kind of world do I want my work to help build? I’m good at spotting a concrete problem and shipping a practical fix. I’m still learning to connect those fixes to a much larger vision.

Sam Altman argued that coding agents have raised the ceiling for small teams. They let a tiny team chase work that used to be out of reach, not just do the familiar stuff faster.

Blake Scholl made that tangible in the physical world. He came from software, not aerospace, and still built toward an economically viable supersonic airliner. His story was a reminder that better technology doesn’t hand us a better future on its own. Someone has to build the products that carry it forward. I’ve long been drawn to products that bridge the physical and digital worlds. I don’t know the exact shape yet, but Startup School pushed me to look past whatever’s directly in front of me.

Later, at a fireside with Vinod Khosla, it got even blunter: dream bigger, take more risks, accept that you might fail. For me that means pushing my own path further than I first thought possible, instead of copying someone else’s company or someone else’s definition of winning.

“How hard can it be?”

Jensen Huang’s session gave me another handle on confidence. NVIDIA didn’t start with the right technical approach. When the team finally admitted that, they bought textbooks, learned what they had to, and rebuilt. His point was simple:

Great founders don’t already know everything. They’re willing to face reality and go learn.

He described the attitude he still brings to unfamiliar problems: “How hard can it be?” The honest answer is usually “harder than you think.” But picturing every future ordeal in advance can rattle you enough to never start. So start. Learn what today asks of you. Then build toward tomorrow.

And that felt personal. Last year I didn’t know how to build the product I now run. This year, users occasionally wake me up with support messages. New problems, sure, but problems I’m grateful to have. Resilience isn’t repeating the same idea forever or pretending everything works. It’s facing what’s real, learning fast, adjusting, and continuing. Not knowing something isn’t a permanent identity. You can learn your way there.

The best of the YC Startup School experience happened between sessions

The talks were great. Some of the meaningful stuff happened in the gaps. At a Google DeepMind after-party I met Logan Kilpatrick and showed him what I’d built with Gemini Live. Telling the builder what I thought, face to face, beat any feedback form I’ve ever filled out.

I first met Nolan Lwin at Founders, Inc. We crossed paths again at Startup School, then at a coffee shop he recommended, where nearly everyone around me was grinding on a YC application. Everyone but me.

Coming back from the restroom, someone stopped me: Mark Nakatani. We’d tried to meet at Startup School and missed each other, only to collide there by accident. I also finally met Nelson Lee in person after knowing him through X for years.

There were real conversations, new friendships, after-parties, and even builder hotpot. Some of these meetings probably felt ordinary to the other person. To me they carried years of context.

When you’ve followed someone’s work, used what they built, or traded posts for a long time, meeting in person changes the whole thing.

Help first, the network builds itself

Sam Altman also talked about how the best networks grow from genuine curiosity and helping people, not from treating every chat as a transaction. That matched my whole trip. I didn’t walk into rooms hunting a target. I shared what I was building, listened to other people’s stories, asked questions, and let the unexpected conversations happen.

It’s how I think about a personal brand too. To me, a brand is the memory people build from what you do over and over, not a polished profile or a job title.

Sometimes helping is sharing what you know or making an intro. Sometimes it’s honest feedback, or just caring about what someone’s trying to build. Writing this is one of those ways.

I’m early, and I haven’t figured it all out. But maybe the honest version of my journey helps someone start sooner, feel less intimidated, or trust their own path a little more.

I want what I publish to make me more useful and more honest, not just more visible.

What San Francisco gives me, and what Ottawa does

In San Francisco, especially at a YC event, ambitious ideas feel normal. Conversations open with agents, models, infrastructure, distribution, and products that were impossible a year ago. Startup thinking leaks into ordinary places. That density is powerful, and it’s still a bubble. Step outside it and plenty of people have never meaningfully used an AI agent, and plenty of businesses are only starting to grasp what today’s tools can do.

There’s an insane amount of alpha in:
– People in SF leaving SF to see how people and companies around the world are using AI (hint: most are still using it primarily as search)
– People outside SF spending even two weeks in SF to see how people and companies are operating here…

— Akshay Kothari (@akothari) July 21, 2026

So the right rhythm for me might be to visit San Francisco now and then, soak up the ambition, meet people, then go home to communities where the problems and the level of AI adoption look completely different.

I sometimes wonder what would’ve happened if I’d moved straight to the Bay instead of staying in Ottawa. Maybe I’d be more connected, or building faster. Maybe I’d have optimized too quickly for the conversations right around me. Ottawa gave me distance, and time to explore freely and find my own direction. That distance cost me access. It also gave me room to become different. San Francisco shows me what’s possible. Ottawa reminds me how much of the world hasn’t reached that future yet.

I also met a lot of Canadians on this trip, a surprising number of them from Waterloo. It left me chewing on Canada’s talent story. How do we create more places where ambitious builders keep bumping into each other? How do we give them better reasons to take risks and stay long enough to build the next wave of communities?

I don’t have a tidy policy answer. But as an immigrant building here, I want Canada to be more than a place that grows and attracts talent right before that talent leaves. I dig into that more here: Canada helped Joe become a builder. Does Canada know how to keep builders?

What I’m taking home from the YC Startup School experience

Last year I flew to the Bay with ideas and uncertainty. This year I flew home with a product, real users, more confidence, and friends from around the world. Startup School felt like a reward for the previous year’s work, and a reminder that a personal win doesn’t have to follow anyone else’s template.

You don’t have to chase every trend or reshape yourself for investors and algorithms. Start with something you actually care about, keep learning, help people, and build something you’re proud to share.

I still don’t know exactly what I’ll build next, or who I’ll be when I’m back in San Francisco. But I know how I want to keep going:

• Be real.
• Be yourself.
• Help others genuinely.
• Think bigger.
• Stay on your own track.

Thank you, Y Combinator, for putting this many ambitious, thoughtful, and generous people in one place.

About the Author

Joe Hu is the founder of Just Joe Technologies, a one-person, AI-powered studio based in Ottawa, Canada, where he builds products with AI and shares the journey publicly. His first iOS app, Joe Speaking, started as a fix for his own English-speaking problem. He writes about building as an explorer, builder, and founder at hubeiqiao.com and posts on X @hubeiqiao.

The post How YC Startup School Rewired the Way I Build appeared first on BYVI.

The hardest part of building something new is never the part everyone assumes will be hard. I keep coming back to this, and Travis Kalanick’s new venture is the cleanest example I’ve seen in a while.

After roughly 8 years of near-total stealth following his 2017 exit from Uber, Kalanick’s company Atoms just raised $1.7 billion in a round led by Andreessen Horowitz, with Uber (the company whose investors forced him out) joining as an investor and Ben Horowitz taking a board seat.

The pitch is enormous: use software, sensors, and robotics to automate the physical economy across food, mining, and logistics

Travis Kalanick is back.

After 8 years of building in stealth, he joins Ben Horowitz and Erik Torenberg to tell the story of his latest company, Atoms, and its vision to digitize the physical world one industry at a time:

01:32 The a16z & Uber backstory
15:36 Different… https://t.co/CmHLuzWYBR pic.twitter.com/hjzHk1vkBf

— a16z (@a16z) July 22, 2026

Ask most people what the hard part of an automated kitchen is, and you’ll hear the same list:

• The computer vision.
• The robotic arms.
• The supply chain.
• The capital.

Every one of these sounds hard because it is hard, and because it’s what makes it into the pitch deck. So that’s where the money and the engineering attention go.

The hard part turned out to be everything nobody puts on a slide.

The machine that slows down when it has to bag your food

Atoms’ flagship food system, the Bowl Builder, takes an order straight off the point-of-sale, dispenses and weighs ingredients, adds sauce, lids the bowl, labels it, drops in utensils, and closes the bag. On paper it can assemble up to 300 bowls per hour.

Except it can’t, not when it also has to bag them. The moment bagging enters the workflow, throughput drops to about 200 bowls per hour. Kalanick’s own word for the bagging problem, on the TBPN podcast, was that it’s “surprisingly hard.”

Look at what that means. A company that just pulled $1.7 billion in a single round, run by one of the most operationally aggressive founders of his generation, loses a third of its output to the most unexpected step in the whole process. And it’s not the sensing. Not the dispensing. Not even the sauce. It’s the bag.

That’s the “fries in the bag” problem: the step so ordinary nobody budgets it any respect, right up until it’s the exact thing standing between you and shipping.

Why the boring problems are the hard ones

There’s a reason this pattern repeats. The glamorous problems attract attention, talent, and money precisely because everyone can see them coming. They get scoped, staffed, and de-risked. The boring problems don’t get that treatment, so they arrive undefended, and they arrive late, usually after you’ve already committed to a timeline and a number.

A bowl is a rigid, predictable object. A plastic bag is not. It’s floppy, it clings to itself, it doesn’t hold its shape, and no two are quite alike. Everything a robot is good at (precision, repetition, hard geometry) is exactly what the bag refuses to give it.

This is what quietly reshaped how I think about leadership, and about finance specifically.

The leadership lesson hiding in the bag

It’s tempting to treat “hard problems” as the big strategic ones, the market, the model, the fundraise, and hand the messy operational stuff to someone else. That instinct feels like leverage. It’s usually the opposite.

The leaders who actually ship are the ones who stayed close enough to the work to know where the real friction is.

Not because they insist on doing everything themselves; that doesn’t scale and it isn’t the point. It’s that you cannot fix what you’ve never gotten close enough to see. Kalanick knows his throughput drops on the bagging step because someone in that organization got close enough to the machine to watch it happen and put a number on it.

Delegation doesn’t work against this. Good delegation is how you build a company. But hand off a problem before you understand it and all you’ve really done is arrange for someone else to discover it for you, later, and usually at far more cost to the business than if you’d caught it early.

I sit in finance, which makes this personal in a specific way. Every model I’ve ever built has a “fries in the bag” line item: the assumption everyone waved through because it seemed too small to matter. The onboarding step that quietly determines whether a customer ever activates. The reconciliation process that works fine at 50 transactions and breaks at 5,000. The manual handoff between two teams that no dashboard captures because it lives in the gap between them.

On a spreadsheet these look like rounding errors. In the business they’re the difference between a plan that holds and one that doesn’t.

The costliest surprises I’ve seen were never the risks we named and argued about in the boardroom. We had those covered; that’s what boardrooms are for. The expensive ones were the steps so ordinary nobody thought to ask whether they’d actually work at scale.

What Atoms actually confirms

It would be easy to read the Atoms story as a victory lap: stealth genius resurfaces, raises a fortune, takes on the physical world. And maybe it turns into that. Horowitz is betting that reshaping the old, heavy parts of the economy needs a founder with range across software and mechanical engineering, and that Kalanick is that founder.

But the part of the story I keep pointing people to is the honesty about the bag, not the $1.7 billion. A founder with that much capital and conviction telling you, plainly, that the humble step is the one giving him trouble. Read it the right way and that’s a good sign: someone got close enough to the friction to name it before it showed up in the numbers on its own.

The strategic bets are real, and they’ve earned the attention they get. But the thing most likely to cost you a third of your output is usually sitting one step downstream of where everyone is looking, dressed up as a task too simple to fail.

Before I sign off on any plan now, I ask myself one thing, and I’d pass it to you:

What’s your “fries in the bag” problem, the boring step you’ve already handed off, assuming it would sort itself out?

Better to go find it yourself than to read about it later in the P&L.

About the Author

Daniel Zhao is Head of Finance at Landbase and a multiple-time founder, most recently co-founder of LavaReach (acquired). He started his career as a CPA, building the financial rigor that still shapes how he operates, then moved from the ledger into the parts of a company where the numbers are actually made or lost: revenue, operations, and go-to-market. That path, from accounting fundamentals to B2B revenue leadership, gives him a rare vantage point on how startups really grow, and where they quietly break.

The post Every Startup Has a “Fries in the Bag” Problem Most Founders Miss appeared first on BYVI.

On July 30 LinkedIn started letting anyone flag your post as “AI slop” with two taps, then feed that flag straight into a model that decides who sees you next. For founders who treat LinkedIn as their main distribution channel, that little button quietly rewrote the rules of what earns reach.

Here’s what actually shipped, why it lands on us harder than on anyone else, and how to post so the button never gets pointed at you.

What LinkedIn actually shipped

Open LinkedIn, scroll for 30 seconds, and you’ll spot them by their tics. “This isn’t just a product update. It’s a masterclass in resilience.” “What a 4-hour flight delay taught me about scaling a Series A.” The “Let that sink in.” And the “Agree?” at the end of each post. And of course one of my favorites: the “Unpopular opinion:” that is, in fact, the most popular opinion on the platform.

That specific flavor of nothing now has a button attached to it. Tap the three-dots menu on any post or comment and you’ll find a “Seems like AI slop” option, as Chief Product Officer Hari Srinivasan showed in his own post announcing the change.

Hari Srinivasan, Chief Product Officer, LinkedIn Ecosystem

The button isn’t just a mute switch. Srinivasan said the flags feed LinkedIn’s classifiers directly: “We are ramping the ability for members to tell us if they believe a post or comment seems like AI slop. Slop is hard to define and the definition changes; this lets us tune our models and make better feeds.”

In plainer terms, every report is a training signal, and posts the crowd tags as slop get less algorithmic reach, especially in the suggested content LinkedIn pushes to people outside your network.

Three other moves came in the same announcement, and each one matters for how you show up.

Why the sudden urgency? Because the numbers are genuinely embarrassing.

AI detection firm Pangram scanned more than 1 million social posts and found LinkedIn was the most AI-saturated platform of all: over 40% of its long-form posts flagged as fully AI-generated, and while LinkedIn made up only a third of everything Pangram scanned, it accounted for 62% of all the AI content the firm flagged.

Why this lands on founders harder than anyone

LinkedIn is where a lot of us go to raise, hire, and sell before we have the logos to do it the easy way. For an early-stage founder, our feed is the pitch deck that runs 24/7.

Let’s start with the obvious risk. If you’ve been scaling your “thought leadership” by feeding bullet points into a model and posting whatever came out, you were already living in the 40%.

Now the readers you’re trying to win, the ones who were going to remember your name at fundraise time, can label that output slop in two taps, and the algorithm will believe them.

The channel you leaned on to look credible can now make you look generic. And unlike a paid campaign you can pause, a reputation for slop compounds quietly: the classifier remembers, and your reach decreases before you notice the flag in your dashboard.

The deeper shift is what LinkedIn is protecting. Srinivasan said the goal is to let “members get feedback from real humans on what sounds authentic.” That’s a trust economy, and trust is the one asset a pre-revenue founder actually has to trade.

And no, this isn’t a LinkedIn-only tantrum. Substack just shipped a tool with Pangram that tells readers when a newsletter was written by AI.

Pinterest added toggles to cut AI content from feeds, and Snapchat said it will stop recommending AI slop outright. Pangram itself raised $9 million to chase this problem across the web.

The LinkedIn posts I’d publish now

The slop button punishes exactly the content that was never going to convert anyone anyway. So your authentic, specific, founder-voiced posts just caught a tailwind.

Write the post only you could write.

The fastest tell of slop is that literally anyone could have posted it. Your defense is proprietary: a real churn number, a deal that face-planted and why, a metric nobody else can touch.

Look at how Tom Hunt (I’m a huge fan) writes about his company Fame: they ship around 1,000 client assets a week, the reviews that used to need people now run on AI, and a job that once meant a high-paying hire is “now a $7 API call.”

You can’t prompt that post into existence, because the numbers only live inside his business. A model can fake your tone all day. But it can’t come up with specific figures you pulled from your own Stripe account.

Kill the AI polish pass.

LinkedIn just deleted its own “enhance your post” button because the polish was the problem, so take the hint. Draft in your own words, then use AI the way LinkedIn’s replacement tool now does: to proofread, not to rewrite.

The second a model buffs your sentences into that smooth, generic filler, you’ve traded your voice (the thing that survives the new ranking) for the thing it now punishes.

If you don’t want to use LinkedIn’s proofreader, try this prompt below with any AI model. It edits like a line editor, not a ghostwriter, protects your numbers, and strips every tell up top without flattening your voice.

ROLE. You are my line editor, not my ghostwriter. Make my draft cleaner and clearer while keeping it unmistakably mine. The test: if someone who knows my writing couldn't tell I wrote it, you failed. When in doubt, do less.

VOICE PRIMING (optional, recommended). Before editing, study 2 to 3 of my past posts (pasted below) for my sentence length, rhythm, the words I reach for and the ones I never use. Match that fingerprint. Do not average me toward "good LinkedIn writing."

YOU MAY: fix grammar, spelling, punctuation, and typos; tighten only genuinely redundant or confusing sentences (cut filler words, never ideas); point out (don't fix) any claim that reads as vague or unsupported.

YOU MAY NOT: change my word choices, rhythm, or structure to sound "more professional"; add adjectives, transitions, hype, hooks, or summary lines I didn't write; smooth out my fragments, asides, or short punchy sentences; touch any number, name, quote, or date (treat them as load-bearing); use em dashes or en dashes (use commas, colons, periods, or parentheses).

BANNED PATTERNS (remove if present, never add): "This isn't just X, it's Y" contrast framing; throat-clearers like "Here's the thing" or "In today's fast-paced world"; engagement bait like "Let that sink in," "Agree?," "Unpopular opinion:"; single sentences stacked one per line for drama; corporate gloss like "leverage," "unlock," "game-changer," "masterclass in," "humbled to announce"; emoji used as punctuation.

PROCESS: read the whole draft first to learn my voice; make the lightest edit that does the job; if a change might alter my meaning or voice, leave it and flag it.

OUTPUT: (1) the edited draft; (2) "Changes," each fix in one line with why; (3) "Flags," anything vague, unsupported, or that sounds like me on autopilot rather than at my sharpest.

MY DRAFT: [paste it here]

Put your face on it and show up with proof.

With 100 million verified profiles, LinkedIn is all-in on identity. Post from your verified personal profile and bolt on proof a machine can’t hallucinate.

• The actual customer message
• The real graph
• The awkward group shot from the offsite
• The team photo where nobody’s looking at the camera
• Imperfect behind the scene moment.
  

Trade cadence for evidence.

The old growth hack was volume, and volume is precisely what the automation crackdown is built to shut down. LinkedIn is already cracking down on engagement pods and automated comments, blocking hundreds of thousands of comment attempts a day. So post 3 times a week with proof instead of daily with filler. One post carrying a number you earned for sure beats 5 that sound like everyone else on LinkedIn.

Treat comments like the product.

LinkedIn is spending real engineering effort removing bot comments because fake engagement damages trust faster than fake posts do.

So do the opposite of automation: read the actual post, reply to people by name, reference the actual thing they said, argue in good faith. Real conversation in your comments is now a trust signal the platform is actively protecting, and it costs you nothing but attention.

When in doubt, use a meme.

A model can fake a paragraph. But it can’t fake taste, timing, or a joke that actually lands. A well-chosen meme is proof there’s a human with judgment behind the account, and it’s native to how people already talk on LinkedIn.

Evan Lee does this well: he pairs a meme with a sharp point about matching your ads to your funnel instead of another 400-word “here’s what I learned” essay, and it lands harder.

You don’t need to be funny every time. You need to sound like a person, and nothing says “not a bot” like a meme you actually found funny.

The bottom line on AI slop button

The slop button reads like a threat, and for the founders who outsourced their voice to a model, it is one. But I think LinkedIn just did something useful: it turned “sounds like a real person with something to say” into the biggest distribution advantage.

The same tools that made it easy to flood the LinkedIn feed just made flooding it pointless. What’s left is something founders have plenty of and big companies can’t copy: a clear point of view, backed by numbers only you can see, in a voice that’s clearly yours.

About the author

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post LinkedIn Just Gave Every Reader a AI Slop Button to Kill Your Reach appeared first on BYVI.

Snap now blocks wholly AI-generated videos from Spotlight recommendations, and the number of people posting original content is climbing fast. For founders fighting for reach, that turns “made by a real person” into a distribution edge you can actually build a growth loop around. Here’s how to move on it before everyone else catches up.

What Snap actually changed

On July 31, 2026, Snap updated its Spotlight recommendation system so that, as the company put it, “wholly AI-generated videos will no longer be eligible for recommendation on Spotlight.” The reasoning is blunt: as low-quality, repetitive AI content floods the internet, Snap wants Spotlight to stay a place where people discover “original perspectives, personal storytelling, and the moments that people choose to create and share themselves.”

Two details matter for anyone building a company.

1. This isn’t Snap turning its back on AI.

Content that’s been enhanced or edited with Snapchat’s own AI creative tools stays eligible for recommendation, and it carries transparency indicators so viewers know what they’re looking at. The line Snap drew is between content that is wholly synthetic and content where a human is doing the creating, with AI as a tool in their hands.

2. The audience is already voting with their cameras.

Snap reports that the number of unique Spotlight contributors globally is up more than 120% compared to last year. Back in April, the company shared that daily US Spotlight creators grew more than 70% year over year in Q1 2026, and that creators who post daily grow their followers more than 10x faster than those who don’t. This is happening on a platform that reached 474 million daily active users in Q4 2025.

Put those together and you get a rare thing in distribution: a large, growing audience where the algorithm is now actively tilted toward exactly the kind of content a scrappy founder can make with a phone (if, of course, your buyers are actually on Snapchat).

One honest caveat before you commit: this audience skews young. Snapchat reaches 90% of 13- to 24-year-olds across 25+ countries, 18- to 24-year-olds are its single largest age group at 35.4%, and roughly 79% of the base is under 35.

If you sell to enterprise buyers, treat Spotlight as a brand and recruiting play rather than a demand channel. If your customers are Gen Z or younger Millennials, this is your front row.

This is a pattern, not a one-off

If Snap were alone here, I’d tell you to note it and move on. It isn’t.

YouTube got there first. In July 2025 it renamed its “repetitious content” rule to “inauthentic content” and made mass-produced, template-driven videos ineligible for monetization. YouTube was careful to say AI-assisted work still earns money, but content that’s “easily replicable at scale” with “little to no variation” doesn’t. Same line as Snap, drawn in a different place.

Pinterest went the transparency route. In April 2025 it rolled out “AI modified” labels globally and gave users a “see fewer” control for AI-generated Pins in categories like beauty and home decor. When a platform hands users a dial to turn down synthetic content, it’s telling you where demand actually sits.

TikTok built the plumbing for all of it. In 2024 it became the first video sharing platform to implement C2PA Content Credentials, the open standard that attaches provenance metadata so AI-generated uploads get labeled automatically.

That standard is backed by Adobe, Google, Microsoft, OpenAI, and Sony, which tells you provenance labeling is becoming infrastructure, not a feature.

Why this is an opening for founders

Most coverage of these changes frames them as a crackdown. I’d flip it. Every one of these moves lowers the value of the thing your best-funded competitors can buy at scale (synthetic volume) and raises the value of the thing you already have (a real person with a real story and a real product).

That’s a redistribution of reach toward the small and the specific. And the buyers are primed for it: a Snap study with MAGNA found people respond most positively to creators who prioritize “keeping it real” over polished perfection, and 49% say they’re interested in buying from authentic creator content.

Here’s where I’d focus.

Founder-led content just got a bigger multiplier.

The platforms are rewarding a face, a voice, and a point of view, which is the one asset a solo founder can produce that a content farm can’t fake at scale. Scrub Daddy turned a founder and a smiling sponge into hundreds of millions of TikTok views by leaning entirely on personality and behind-the-scenes reality. You don’t need a studio. You need to show up as yourself, consistently, on a surface that’s now weighted in your favor.

“Boring” behind-the-scenes footage is now premium inventory.

The candid, unpolished, shot-on-a-phone moment is precisely what these algorithms are being tuned to surface. Duolingo built a following in the millions on unscripted, chaotic clips of its mascot filmed around the office, not on rendered ad spots. For a startup, the build itself (packing orders, debugging live, talking to your first customers) is the content, and it reads as authentic because it is.

Speed and reaction beat production value.

Authentic content wins on timeliness, and small teams are structurally faster than big ones. Ryanair runs one of the most-watched brand accounts in short-form by posting same-day, self-deprecating reactions that a large brand’s approval chain could never ship in time. A two-person startup can react to a trend before lunch. That’s a genuine edge now that the feed favors native, of-the-moment posts.

Distribution is getting cheaper exactly where paid is getting more expensive.

As synthetic volume gets throttled, organic reach for real creators becomes more valuable, and Snap’s own numbers (daily posters growing followers 10x faster) show the compounding is real. Gymshark built a nine-figure brand largely on athlete and customer content rather than ad spend, and that playbook gets stronger every time a platform tilts toward human-made posts. If your CAC is climbing, an authenticity-weighted feed is a discount you can go claim.

@charles__johnson OG Gymshark Addict #gymshark #fyp #davidlaid #OGGymshark #gym ♬ original sound – c0nd1nh0

Provenance becomes a trust signal you can market with.

With C2PA labels spreading, “made by a human” is turning into a verifiable badge rather than a vibe. Liquid Death built a billion-dollar brand on real people and a real point of view, and in a labeled-content world that kind of provenance stops being a nice-to-have and starts being a filter customers use to decide who to trust. Being demonstrably human is becoming a moat.

Where this goes

The last decade of social taught founders that reach was something you rented from an algorithm you couldn’t influence and mostly had to pay. What’s happening now is quieter and better: the algorithms are re-pricing reach toward the one thing money can’t manufacture, which is a real person making something real. And that’s a market correction in your favor.

About the Author

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post Snapchat Spotlight Banned AI Video. Here’s the Founder Play appeared first on BYVI.

Pull up your own landing page and read the headline out loud. If the pitch is some version of “the fastest way to [do the thing],” “a beautiful, intuitive interface for your workflow,” “automate your expense reports,” “all your tools in one place,” or “powerful, no-code, and easy to learn”, stop on it for a second. What those lines have in common is that every one of them describes a task a person repeats inside your software. And doing a repeatable task inside software is exactly what an AI just learned to copy by watching.

On July 21, 2026, Anthropic shipped a feature in Claude Cowork called Record a skill.

When you stop, Claude studies the recording and turns it into a reusable skill it can run again on command. No prompt, no script, no instructions to hand-author. You show it once and it does it forever.

OpenAI shipped a near-identical feature, Record & Replay, in Codex 33 days earlier. When two rival labs converge on the same idea inside a single product cycle, it’s not just a feature anymore. It’s a whole new direction.

And this direction points straight at a whole category of software: products whose value was mostly a nice front-end, a library of templates, or a bundle of integrations wrapped around a task a person repeats.

That is the most replicable thing you can own. And copying it just went from “a rival team and a 6-month head start” to “one afternoon with the recorder running.”

This is already happening, not coming

When Anthropic shipped its Claude Cowork plugins in late January 2026, roughly $285 billion was wiped off software, financial-services and asset-management stocks in a single session, a rout Jefferies’ trading desk nicknamed the “SaaSpocalypse.” By spring, analysts were tallying the damage across enterprise software in the trillions.

So investors were doing blunt math: agents that handle work at the task level mean companies need fewer seats, and per-seat licensing is exactly how most software makes its money.

Then it got specific. 3 days before Anthropic launched Claude Design in April, its chief product officer, Mike Krieger, quietly resigned from Figma’s board.

Anthropic is running the oldest predatory playbook in Big Tech (Save this).

Here is what actually happened.

Anthropic's own Chief Product Officer, Mike Krieger, was sitting on Figma's board and he resigned on April 14, 2026.

Three days later, Anthropic launched Claude Design,… pic.twitter.com/UjBFf16nCK

— Milk Road AI (@MilkRoadAI) July 5, 2026

Claude Design went live on April 17. Figma’s stock fell about 7% the same day. If you use Claude Design, you know that the tool takes a plain-language brief and produces prototypes, slide decks, and landing pages, then hands a website design straight to Claude Code as a build-ready bundle.

For a designer whose job was execution, that’s a large chunk of the work, gone. And investors didn’t read it as a one-company story: Adobe, Wix, and GoDaddy all slid in the same window. The market repriced the entire design-to-publish stack.

To be fair to the other side of the trade: Figma still owns an estimated 80–90% of professional UI/UX design, Claude Design is a token-hungry research preview, and LLMs remain shaky the moment you edit individual visual elements.

Wedbush called the sell-off an “Armageddon scenario” that’s “far from reality,” and enterprises won’t rip out billions in existing software overnight. Anthropic itself is hedging – partnering with Salesforce, Intuit, and DocuSign rather than only replacing them. The point isn’t that software dies. It’s that the interface stopped being the thing worth defending.

Figma made the point clearest of all: it shook hands with the company supposedly eating its lunch. In February 2026, the two shipped a feature called Code to Canvas that pulls interfaces built in Claude Code straight into Figma as fully editable design layers. You can now type “Send this to Figma,” and running AI-generated code becomes something a designer can drag, tweak, and refine.

It’s a shrewd hedge: rather than pretend agentic coding isn’t happening, Figma made itself the place that work lands. But it also captures the tightrope every incumbent is walking.

As CNBC put it, the risk is that Figma is building a better on-ramp to a highway it no longer controls.

Why a UI is no longer a moat

For 15 years, a great interface was a legitimate edge. Being the tool that made a painful task feel easy could carry a company to a real business, because the alternative (a competitor rebuilding your UX and re-earning your users’ habits) was slow and expensive.

Whatever your users do over and over inside your product is precisely what these recorders capture best. If the only thing standing between a user and the outcome is your screens, an agent can increasingly deliver the outcome and skip your screens.

So the strategic question flips from “do we have the best interface?” to “do we have anything a recording can’t capture?”

That reframing is the whole game. Below is where I’d look.

What actually defends you now

Six things a screen recording can’t capture. Build for these.

1. Proprietary data that compounds.

Not “we have data”, everyone says that. Data you uniquely generate and that gets better the more your product is used: repayment outcomes for a lender, real-world delivery times for a logistics tool, conversion results tied to specific decisions.

An agent can watch your screen and copy the workflow but it cannot reconstruct a dataset built from a hundred thousand real outcomes it never saw. The moat that keeps a Bloomberg Terminal or Stripe’s Radar fraud models hard to touch.

2. Be the system of record.

Agents are getting very good at reading data and taking actions, but someone still has to be the authoritative ledger: the place data actually lives, where changes are committed with permissions, audit trails, and integrity guarantees.

If you’re a thin view on top of someone else’s source of truth, you’re the most replaceable layer in the stack. If you’re where the data is born and held, you’re infrastructure, and the agents end up routing through you.

You must own the write layer, not just the read.

3. Real network effects.

Value that grows with each user in a way no one can clone by copying your front-end: marketplaces, collaboration where the graph itself is the product, data network effects where one customer’s usage improves the product for the next.

Slack is the clearest case. Anyone can rebuild the chat interface in a weekend, but no one can rebuild the years of messages, channels, and integrations your whole company has wired into it.

4. Be the accountable party in high-stakes work.

In regulated or consequential domains (health, finance, legal, safety) someone has to be licensed, insured, auditable, and liable. An agent can draft the filing, but it can’t be the certified entity that signs it and carries the risk. If your business sits where trust and accountability are structurally required, you own a position agents can’t take by watching a screen.

It’s the moat under a Gusto, a Carta, or a Vanta, where someone has to be the accountable party of record when payroll, a cap table, or a compliance audit is wrong.

5. Own distribution and brand.

When capability commoditizes, distribution becomes the scarce asset. The company users think of first, trust with their data, and already have a relationship with wins even when the underlying task gets automated.

Founders under-invest here because it feels less defensible than tech, but right now it’s the opposite. Owning the customer relationship and the channel is one of the few edges an agent can’t reproduce from a demo. HubSpot and Canva didn’t win on features alone, they both won on being the name people already reach for.

6. Be the tool the agent calls

Stop standing in the agent’s path, become the thing it calls. Expose your product as a tool, skill, or MCP server that Claude and Codex route to for a specialized capability they can’t do well themselves.

Be the best node in an agentic workflow rather than a UI the agent bypasses. So become the Twilio of your niche, the thing agents route through, not around.

And move off per-seat pricing, which is the exact model the market is punishing. Price the outcome or the usage, so your revenue grows with value delivered rather than with headcount that agents are about to shrink.

The bottom line

Prompting isn’t dead, and neither is software. But the ground shifted, and it shifted under a specific kind of company: the one whose edge was a lovely interface for a task people repeat. If that’s you, this is the year to build something durable underneath it.

The defensible positions are the ones an agent can’t cheaply reproduce by watching: proprietary data that compounds, being the system of record, real network effects, structural accountability, distribution, and being the capability the agents depend on rather than the layer they route around. Those are the real product, and no recording can reach them.

So build that. Build the company agents route through, lean on, and depend on. The most capable tools the industry has ever shipped are now aimed squarely at the grunt work, which frees you to pour your energy into the things that were always going to decide whether you win.

The founders who spend the next few years building something a recording can’t capture are the ones this era belongs to.

ABOUT THE AUTHOR

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post Claude Now Learns Any Task by Watching and It’s Coming for Your Software appeared first on BYVI.

A shopper opens ChatGPT and types “best standing desk under $400 that ships in two days.” Thirty seconds later they have 3 recommendations, a comparison, and a checkout button. They never visited a search results page. They never saw your ad. The agent already decided, and if your product wasn’t in its consideration set, you lost a sale you never knew existed.

This is agentic commerce, and in 2026 it stopped being just a demo. AI agents are completing real purchases inside ChatGPT, Perplexity, Google’s AI Mode, Microsoft Copilot and Amazon Rufus. AI platforms are projected to drive roughly $20.9 billion in retail spending this year, nearly quadrupling 2025.

The funnel is collapsing into a single conversation, and the new question for founders is brutally simple:

The part that’s good news for small teams is that agents don’t reward ad budgets. What they do reward is clean data, trust signals, and being present in the right feeds.

First, understand how an agent actually picks

AI shopping agents don’t “rank” pages the way Google’s blue links did. They assign confidence scores, every product gets scored on how well it matches the shopper’s stated criteria, and the agent recommends what it can stand behind.

To score you confidently, the agent needs 3 things:

1. Structured data it can easily read
2. Trust signals that prove you’re legitimate
3. Validation from sources it already trusts

Your Google Merchant Center feed and your organic SEO are still load-bearing. AI commerce is a new layer on top, not a replacement.

Step 1: Get into the feeds, all of them

You can’t be recommended if you’re not in the pipe. There are several entry points, and they’re cheaper to set up than most founders assume.

Underneath them, 2 open standards now move the money and the data.

1. The Agentic Commerce Protocol (ACP): co-developed by OpenAI and Stripe, now with 25+ partners — standardizes agentic checkout and powers Instant Checkout in ChatGPT.
2. The Universal Commerce Protocol (UCP): from Shopify and Google, with a Tech Council that now includes Amazon, Meta, Microsoft, Salesforce and Stripe — covers the broader journey from product discovery through post-purchase.

You don’t need to implement either by hand; your platform increasingly does it for you. But knowing they exist tells you where the rails are going.

Step 2: Make your product data machine-perfect

This is where most stores quietly lose. Agents read fields in a specific priority order, and gaps get filled with guesses, or with a competitor who left no gaps. Stores with 99.9% attribute completion are seeing 3–4x higher visibility than stores with sparse data.

Work the fields in order of impact:

Step 3: Use JSON-LD schema on every product page

The feed is one layer; your live product pages are the open-web layer that AI crawlers read directly.

Implement Product, Offer, and Review schema in JSON-LD, it’s the preferred format because it keeps structured data separate from your HTML, so AI parsers extract it cleanly without inference.

This is the difference between an agent reading your specs and an agent guessing them. Validate it, and make sure every product has it. This is a one-time engineering task with compounding payoff.

Step 4: Win on trust signals, reviews are now ranking fuel

Reviews stopped being social proof for humans and became input data for machines. The ChatGPT feed spec explicitly accepts review count and average rating fields, and agents weight them heavily when deciding who to recommend. Amazon Rufus synthesizes review sentiment and Q&A directly. Fresh, plentiful reviews give the LLM the recent, specific evidence it needs to recommend you with confidence.

Tactically: turn on post-purchase review requests, make leaving a review frictionless, and feed your review count and rating into your product feed. A product with 400 reviews at 4.7 stars beats an identical product with 12 reviews every time an agent has to choose.

Step 5: Build third-party citations (this is your Perplexity edge)

Perplexity in particular prioritizes citations from Reddit, expert blogs, and editorial sources over your own marketing copy. Agents trust what other credible sources say about you more than what you say about yourself.

This is where bootstrapped and founder-led brands have an unfair advantage. You can get mentioned in niche communities, founder newsletters, and subreddit threads in ways big brands can’t fake.

Get your product into legitimate “best X for Y” roundups, answer honestly in relevant Reddit threads, earn reviews on independent blogs. Every credible third-party mention is a citation an agent can lean on.

Step 6: Exploit the rich-media gap before everyone else does

The ChatGPT feed spec accepts video and 3D model links, and almost nobody is using them yet. That’s a textbook early-adopter window. Richer media gives the agent more to understand and more confidence to recommend, and right now adding a product video or 3D asset is a near-free way to stand out in a field of plain image feeds.

Step 7: Write conversational content that matches how people actually ask

Agents answer questions, so publish content shaped like the questions buyers ask:

• “Is X good for small apartments?”,
• “What’s the difference between X and Y?”
• “Best X for beginners on a budget?”

Clear, factual, comparison-friendly pages give agents quotable, citable material. Agents reward specificity, not adjectives, so skip all the fluff.

Step 8: Measure what’s working

You finally have instruments. Google launched Search Console’s Search Generative AI performance reports in June 2026 (initially to a limited set of sites) surfacing impressions and pages appearing in AI experiences like AI Overviews and AI Mode. Watch which products show up, test feed changes, and double down on what lifts your presence. Treat it like any other channel: instrument, iterate, compound.

The 10-Minute Checklist to Get Recommended by AI Shopping Agents

Pin this somewhere visible:

• Google Merchant Center feed live and clean
• Shopify Agentic Storefronts (or your platform’s equivalent) enabled
• ChatGPT feed + Perplexity Merchant Program submitted
• Titles loaded with brand + type + feature + differentiator
• GTINs and full attributes on every product (aim for ~100% completion)
• Shipping time and returns as structured, parseable data
• Product + Offer + Review JSON-LD schema on every page
• Review count and rating flowing into the feed; review requests automated
• Third-party citations seeded (Reddit, roundups, expert blogs)
• Video/3D media added where the spec allows
• Feed price = storefront price, inventory synced
• Search Console AI report monitored monthly

The takeaway

Agentic commerce rewards the unglamorous work: complete data, real reviews, clean feeds, honest third-party presence. None of it requires a brand budget, which is exactly why disciplined small teams can out-rank incumbents in the agent’s consideration set.

Over the next few years, the brands that win will be the ones the machine can read most clearly, and that’s the opportunity in front of you. You don’t need a bigger budget, just better data than the competitor sitting one row away in the results.

So start today. Clean the feed, sharpen your titles, turn on reviews. Every field you complete gives the agent another reason to choose you.

The AI agents are already shopping. Make sure they find you first.

About the Author

Assel Beglinova is a CEO and Co-founder at Paperstack, a platform that provides working capital for e-commerce sellers. With her expertise in revenue-based financing, Assel helps merchants achieve their growth goals and thrive in the competitive world of e-commerce.

The post How to Get Your Product Recommended by AI Shopping Agents appeared first on BYVI.

My father-in-law has migraines. Bad ones, and unfortunately quite often.

He also doesn’t speak English very well. So when he sits in front of a doctor and tries to explain what’s happening inside his head, the words come out hopelessly vague: “headache… sometimes… strong.” The doctor, in return, offers equally vague recommendations (proper sleep, a good diet, less stress) and sends him home.

Since “sometimes, strong, headache” doesn’t sound like a serious case, they don’t run any diagnostics or refer him to a specialist.

A doctor can’t diagnose from a 10-minute appointment and a patient’s fuzzy memory, especially through a language barrier. But if that same doctor received 30 days of clean data (attack frequency, pain intensity, duration, and the meds he took), the case would be taken much more seriously.

So I built him a migraine diary app in a few hours. To be completely transparent, I’m a marketer who happened to co-found two software companies in the past, but I’m not an engineer. I understand how software works, UI, backend logic, data architecture, but I don’t write code. And I built the entire thing in one afternoon, with one latte and a $25 plan upgrade, using an AI app builder called Hercules.

Here’s exactly how I did it, start to finish.

Research first, prompts second

Even though I had a specific use case and a specific user in mind, I didn’t start by building. I started by checking whether apps like this already existed, and what people loved and hated about them.

The category already has a clear leader, so the idea was obviously validated. But its users complained about the same 3 things over and over.

• a bright, cluttered interface
• a locked-down free tier
• and subscription fatigue

The most valuable thing I found was a peer-reviewed study that analyzed 945 reviews of headache apps to understand what users actually want:

• fast, simple logging of symptoms, triggers, and treatments
• personalization and simple trends in their own data over time
• easy way to export data to show their doctor

And the most requested feature in the entire category was exactly my father-in-law’s problem. In my case, the doctor report was the whole point of the app all along.

The study pulled 945 reviews from these 15 apps.

So my vibe coding plan wrote itself:

Your first prompt is your product brief

Hercules works like this: you describe what you want in plain language, it builds the app, you look at the preview, you give feedback, and it revises. No code anywhere. If you ever briefed an agency or a freelancer, you already have the core skill.

I spent more time on my first prompt than on anything else that day, and I believe this is what made the difference. 3 things I did that I would recommend to anyone:

The prompt was solid. But my workflow, not so much.

Hercules has a Plan mode that interviews you and creates a roadmap before building, and their docs clearly recommend it for exactly this kind of big first feature. I completely forgot to turn it on and sent my prompt in the default ‘Build mode’.

I switched mid-build, expecting to have broken something.

But Hercules recovered on its own and produced a clean four-milestone roadmap: theme and home screen, the log attack flow, attack details and editing, and the doctor report.

5 credits bought the entire interface

The first version of the app cost me about 5 AI credits. I’m careful not to call it a working app, because it wasn’t one yet. My prompt explicitly asked for the front end only, with realistic mock data, no auth, no backend. I wanted to nail the user experience first and wire the logic after.

So Hercules delivered exactly what I asked: a fully interactive front end running on mock data, a dark, calm migraine diary with a color-coded calendar and a doctor report screen.

The logging flow was split in two on purpose.

Mid-attack, the app asks only two questions, how bad and where, and you’re done in seconds. Later, when you feel human again, you can open the entry and answer the rest: what you felt, what happened before it, what medicine you took and whether it helped, plus a free note for anything unusual.

You could tap through everything, but nothing saved and the PDF button was just a button. The real logic came later, and as you’ll see, that is exactly where the money went.

Still. A complete, polished interface for 5 credits, and I had 30. I was genuinely blown away.

What vibe coding actually is: a review loop

The building takes minutes. The reviewing is the actual job. The loop itself is simple. The AI ships a version, you walk through it the way your user would, you send back specific feedback, and it revises. And then you do it all over again.

The quality of the product depends entirely on the quality of what you feed the AI: your prompts, your reviews, and your feedback.

So I went through every screen the way my father-in-law would use it, one thumb, low light, zero patience, and sent back a list of revisions.

Hercules came back with an itemized changelog, ran its own error check, and took its own screenshots to verify the work before showing it to me. Watching an AI QA its own output was one of those moments where I sat back and thought: I’ve had human contractors who did less checking than this.

The real skill in vibe coding: knowing your user

Remember the red line from my first prompt? Strictly a diary, no interpretation, ever. And I want to be clear upfront: nothing here is Hercules getting it wrong. Hercules did exactly what a great AI app builder should do. It built what I asked for, it flagged what was broken, and it suggested what works for everyone else in the category. The one thing AI cannot do is know my user the way I do.

The first test came from my own product copy: “What might have caused it?”

It looks completely innocent. But my rule said the app never interprets, and “caused” quietly invites the user to build theories. And we all know that patient trigger theories are famously unreliable, people blame chocolate for their migraines when craving chocolate is often an early symptom of an attack that has already started.

My common sense says: collect facts ONLY, let the doctor draw conclusions. So I sent Hercules one small prompt, and it changed the question to “Did any of these happen before it?” and updated the matching section of the doctor report in a single pass, a few-line edit, nothing rebuilt, nothing broken.

Now the app records what happened, and the doctor decides what it means. More honest, and funny enough, more credible to the person reading the report.

Hercules audits its own work (and it’s seriously impressive)

The second test came from my favorite feature of the whole platform. Hercules has a built-in Audits feature: it takes screenshots of your app, reads through the actual page code, and reviews everything across product, design, engineering, and security, then reports its findings.

I ran the product audit and the results genuinely impressed me: 8 findings, including “nothing actually saves” (fair, I had only asked for the front end at that point) and “the PDF button doesn’t work.”

But finding #8 suggested adding “personal insights” to the home screen. The exact feature my rule banned. The audit wasn’t wrong by industry standards. Every health app does insights. And that’s the point: the AI recommends the industry default, but your principles are your job.

The compromise ending was my favorite part. Instead of “insights,” the home screen got a snapshot with nothing but raw numbers (no analysis, no judgment). Hercules turned that constraint into a design better than the insights would have been, and I loved it.

The simple question I used: can the user verify it by counting? If yes, it’s a fact, and facts are allowed. “You’re doing better this month” is a conclusion, and conclusions belong to the doctor.

How much does it cost to build an app with AI (my honest breakdown)

I ran out of credits at 0.19 with two features left, so I upgraded to the $50 a month plan. Not just to finish this app, I’m already vibe coding my next one and knew I would need more credits.

How I saved credits on design decisions

One credit-saving trick I’m proud of: I didn’t burn Hercules credits exploring color palettes. I used Claude Design to generate a visual palette board instead: 5 warm accent options (clay, terracotta, ochre, brick rose, caramel) mocked up on real app screens. I picked my favorite and handed Hercules the final decision as a simple instruction. Old bootstrap habits die hard.

From Publish to iPhone home screen in 2 minutes, no App Store needed

I expected publishing to be the complicated part. It was the easiest step of the entire build.

Apps you build with Hercules can ship as PWAs (Progressive Web Apps). If you have not met the term before, a PWA is a web app that your phone treats like a real app: it gets its own icon, launches full screen, works offline, and can even send push notifications.

The biggest difference from a native app is that a PWA installs from a link instead of an app store. Here is the entire flow:

Two minutes after clicking Publish, the app was on my father-in-law’s home screen. And because it stores everything on the device, it works in a dark bedroom with no signal, exactly where migraines happen.

His appointment is coming up. For years he has walked into that office with “headache, sometimes, very strong.” This time he will walk in with evidence: every attack, how bad and how long, which medicines helped and which didn’t, and his own notes. All on one printed page.

What makes a good vibe coder

I speak with so many founders who have spent years waiting to find a technical co-founder before they build anything. And I want to say this as directly as I can: you are not the bottleneck you think you are. Writing the code was never the hard part, and with tools like Hercules it isn’t even a step anymore. What the AI actually needed from me was everything else.

It needed me to know exactly who the user is. Not “users” in general, but a light-sensitive senior in the middle of an attack who can barely look at his phone. It needed me to have taste and hold onto it. It needed me to understand how the app should behave, where the data lives, what happens offline, what a broken state looks like. And it needed me to defend one principle while every default quietly pushed against it.

None of that is coding. That is just running a product the way you would run any business.

So if you ask me what actually makes someone good at vibe coding, my answer is none of the things you would expect. If you know what good looks like, have common sense, and deeply care about your end user, you will be great at it. Everything else, the AI already knows.

About the author

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post How I Vibe Coded a Health App with Hercules in One Afternoon appeared first on BYVI.

I met Joe (Beiqiao) Hu at Toronto Tech Week. When I asked him what had brought him there, I braced for the usual founder answer, some polished version of “I built an app that.” Instead he said:

It stopped me. Not only the thought, but the word. Serendipity is a word I picked up only after 10 years of living and studying in Canada.

Joe Hu, Beiqiao Hu, still remembers the trick he used in primary school. When the English teacher asked the class to read vocabulary aloud, he did not know how. So he asked the classmate next to him to write the Chinese pronunciation under each English word, and he read the Chinese instead. He calls what he had back then “deaf and mute” English: he could squint at a word on a page, but he could not hear it and he could not say it.

That kid grew up to score 106 out of 150 on China’s national college entrance English exam, 469 on CET-4, and 426 on CET-6, and then quietly decided he was simply not a language person. When he thought about studying abroad as an undergraduate, the moment he pictured his English level, he gave up on the idea. He tried to improve after graduation. He says he failed every single time.

Today he lives in Ottawa, has spoken at more than 20 events around Canada, runs a company called Just Joe Technologies, and ships an AI product that tens of thousands of people use to practice speaking English.

Moving to an English-speaking country did not do that for him, and he is blunt about it. Living abroad, on its own, changed nothing. What changed things was building the tool he wished existed, then aiming it squarely at himself, every day. Serendipity, maybe. But the kind you have to build with your own hands.

Starting over, one word at a time

The restart came in 2020, at home during COVID. Joe went back to the foundations, vocabulary one word at a time, which felt faintly ridiculous for a grown man who read English on Twitter every morning and assumed he was basically fine. Then he sat a reading mock in May 2021 and scored 5.5. It took him 3 hours. Reading and listening came back with effort. Speaking and writing were a different category of difficulty. He had a full-time job, so he hired one-on-one tutors and bought himself some technique. In March 2022 he scored a 6.5 on IELTS, the number he had set as a private border: clear it or stay home. He still calls it one of the best days of his life.

Speaking would not let go of him, though. He got a school offer in 2023 and spent a good while quietly unconvinced he belonged. What finally loosened that doubt was not an exam. It was a trip to New Zealand, where eleven native speakers, one after another, told him his English was good. One American teacher said it was better than that of most Chinese speakers she had met. Something settled in him: his English was already enough for a real conversation. He filed his study permit.

“Living abroad didn’t magically improve my English. Dedicated practice did.”

The idea that arrived in the shower

Joe was already the type who builds his way out of a problem. He had wired up custom GPTs to grade his own IELTS reading and writing. Speaking, the part he actually needed, had nothing good behind it. Then in November 2023 OpenAI gave ChatGPT a voice, and Joe had the idea in the shower that would eat the next two years of his life: simulate the entire speaking exam, examiner and all.

He built the IELTS Speaking Simulator GPT that December and open-sourced it. His read on why it worked is sharper than it first sounds. A speaking exam ships with a rubric, which means the AI’s output has something concrete to be measured against, which makes the whole task an unusually clean fit for the technology.

“Even hallucination is a feature, not a bug,” he says, and in this one corner of the world he is correct: an examiner that improvises fresh questions is simply doing its job. He used it daily, shipped a V3, posted it to RedNote, and watched it go viral. Over time roughly 50,000 learners ran on it.

He also caught the flaw that would quietly become his company’s entire thesis. After each session he copied the feedback into Notion, then never opened it again. Same topics, same mistakes, on a loop.

The problem, he worked out later, was never how much he practiced. It was how well.

Knowing when to quit, and when to wait

In August 2024 Joe landed in Canada for a master’s, and got a small pleasant surprise at the door: a classmate already knew his simulator GPT. Joe wanted to turn it into a real product and did not believe he had it in him, so he set himself a low bar, refresh the question bank each season and call that maintenance.

Then the model moved. OpenAI shipped a real-time voice API late in 2024, and Joe built a demo at an AGI Venture Canada hackathon.

It worked, and it was expensive: a two-minute conversation cost more than $2. Worse, every time he fixed one bug, three new ones appeared, which is the specific suffering of building software while leaning on AI to write the parts you do not fully understand. So he stopped. On purpose. He decided to wait for better models rather than ship something he was afraid to touch.

Most founders would file that under failure. Joe filed it under timing, and timing turned out to be the skill that separated this attempt from the last one.

A trip to the Bay Area in mid-2025 handed him the line he kept repeating to himself: build something that belongs to yourself.

By then he had spent the year crossing from non-technical to genuinely shipping front-end work, first with Cursor, then with Claude Code. He flew home to Ottawa with a plan and a stack of free AI credits.

Mark, and the two-minute report that became a company

The center of this story is not a model release. It is a man named Mark. Non-native English speaker who sounds native, former army officer, economist by trade, onetime United Nations employee, and, in his own words, a full-time grandpa and part-time citizen in action. They met before a networking event in Joe’s first month in Canada. When Mark heard that Joe wanted to get better at English, he started sending him voice messages and handed him a deceptively small assignment: report your day, out loud, in no more than two minutes.

Joe did it inconsistently at first. Last summer he committed, sending Mark a report every day and getting feedback every day in return. Mark even gave the habit a name to make it stick: Reporter Joe Speaking.

Joe often could not feel his own progress, but Mark and a mentor could see it, and more than any band score, he noticed he could now describe his own life without flinching. This, far more than the exam, is why the product exists. He wanted somewhere to keep his daily reports and get them graded. “If I hadn’t met Mark, I might not have built this,” he says. “Thank you, Mark.”

The pressure that turned a habit into a company was bureaucratic. Joe needed a CLB 9 to chase permanent residency after graduation, so he went back to practicing, even treating networking events mainly as speaking reps rather than for the networking. His score would not move off roughly 6.5. He got critical of the AI, went back and read the feedback properly, and conceded it was right.

The weak link was him. He did not have the patience to run the same topic twice, or the stomach to listen back to his own voice. That admission became the product’s spine: feedback and repetition over an endless feed of fresh questions.

80 days to a working product

He folded the whole thing into his master’s research under professor Tony Bailetti at Carleton, and paired it with a scoping review of the actual learning science, mechanisms with names like interleaved practice and elicited feedback.

One project, several problems solved at once. He also made a clear-eyed case for why he was the right person to build it: an AI tinkerer who owned the exact pain and understood it from the inside.

He started building in mid-November 2025 on free Claude Code credits. When Opus 4.5 launched, it moved what he believed was possible, and he could suddenly see the thing shipping. When the free credits ran out he paid for a $200 subscription and has zero regret about it: without it, he says flatly, the product would not exist. The voice integration nearly broke him. He budgeted one week to wire up Gemini and spent five. His first 3 paying customers hit a brutal first impression, the connection dropping after ten minutes, and he made stability the only thing that mattered until it held.

What came out the other side is Joe Speaking: a real-time AI examiner for IELTS and CELPIP that hands you a band estimate, a transcript, a replay, and section-by-section feedback for about a dollar a session, against the $25 to $60 an hour a human tutor charges.

Underneath the exam mode sits the habit Mark taught him, recording your day, reviewing it, running the same topic again, comparing the takes. It was enough of a business to put his name on real paperwork. He registered Just Joe Technologies, tagline included: Building What Was Impossible Yesterday.

The number he had chased for years

In May, after months of rehearsing on software he had written himself, Joe sat the test and scored CLB 9 on his first attempt, the number he had carried around for years. He refuses to treat it as a finish line. The real target, he will tell you, lives outside any rubric: saying what he means, clearly and without fear, so people understand both his point and who he is.

That is the part worth sitting with. He still says his English is far from great and his app is firmly in alpha. He is not selling a tidy before-and-after. He is selling a method, and he happens to be its first and most honest test case.

What builders can learn from Joe

Be the market, not its observer 

Joe never surveyed learners. He was the learner, which is exactly why he could feel it when the product was lying to him. That kind of judgment cannot be outsourced; it comes from needing the thing yourself.

Run on your own software

He practiced on Joe Speaking daily and earned his score with it. Using your product as its harshest user surfaces the truth faster than any round of customer interviews.

Ship the rough version in public

The open-sourced GPT was raw, and it still built him an audience of tens of thousands who were already waiting when the real app arrived. Those RedNote likes were also the fuel that kept him going on the hard nights.

Treat quitting as a tool

He walked away from the 2024 build on purpose, because the models were not ready, then moved the moment they were. Knowing when to wait is not the opposite of ambition.

Aim past the score

The line he borrowed from Mark runs under everything: it is not your band score that matters, it is how clearly you can say what you think and how confident you are while you say it.

Sell the loop, not the feature

His own bad habit, practicing without ever reviewing, told him where the value was hiding. He bet the company on repetition and feedback rather than on streaks and badges.

About the Author

Lidia Vijga

Lidia Vijga is a content engineer focused on building in-house growth engines and training AI agents that help companies scale content and dominate brand visibility across traditional and AI search. She is also co-founder of DeckLinks, bringing firsthand founder experience in building and scaling technology products.

The post Joe Hu Couldn’t Beat the IELTS Test, So He Built His Own Examiner appeared first on BYVI.

If you are planning a startup journey or already in the middle of one it’s really important to take a step back and think through your personal finances. Startup life can be unpredictable, and not having your finances in order can add a lot of unnecessary stress and affect not just your personal life but also to how you run your company. Having a solid financial plan gives you the peace of mind to focus on building and growing your startup. This is the reason I built Planwell, to help people plan their finances for major life decisions.

Here is the definitive guide for the major financial decisions every founder should think about before and during the early startup phase.

Decide how long you will commit to your startup

It is wise to give yourself a certain period of time to work on your startup and evaluate whether it will work out. Set clear milestones that you need to achieve within a certain period of time.

This time-bound approach creates focus and prevents the startup from becoming a never-ending project. It also helps you plan your finances better.

How does your stage of life affect this decision?

Your financial strategy will almost certainly differ depending on where you are in life. Early career founders often have lower fixed costs and may still be on a parent’s health insurance plan, which makes a shorter runway more manageable. Founders with families face higher fixed costs such as childcare, schooling and family health insurance. However, they may also have a higher level of savings accumulated and potentially a partner whose income or insurance provides a buffer. Take stock of your full picture and try to maintain a higher personal reserve before committing fully. 

Analyze your cash flow and budget requirements

Before leaving your salaried job, analyze your personal cash flow and understand how much you spend every month. In particular, try to figure out the discretionary expenses that you could cut while you are working on the startup. 

What can you do to prepare financially while still in a W2 role?

If you have the leeway, try to keep your fixed costs low. If you have the dream of doing a startup down the line, do not buy the most expensive home and do not max out your lifestyle, even if you can afford it. The flexibility you preserve now becomes the runway you have later.

What are your health insurance options as a founder?

Unless you have a spouse or parent whose health insurance you can rely on, you need to figure out health insurance options and budget for them before you make the leap. You can purchase coverage through the Individual Marketplace via the Affordable Care Act (ACA). Typical monthly premiums vary by age and state:

Early-stage teams sometimes use services like Take Command Health to navigate options.

Estimate whether you have funds to cover your budget

How many months or years of runway do you have to cover your budget? Does this match the period of time you gave yourself to work on your startup and take it to the next level?

This calculation determines whether you can bootstrap or whether you need other income sources.

How can you model your personal financial runway before quitting your job?

Start with the calculation above, but keep in mind that most founders underestimate costs like healthcare, quarterly taxes and irregular expenses.

A tool like Planwell’s Career Break Planner  lets you input your exact income, expenses, assets and goals and stress-test your runway under different spending scenarios. You can also model how a major decision like buying a home changes how long your savings will last. Unlike a generic AI tool like ChatGPT, Planwell uses its own financial models and runs your exact numbers, so the output is specific to your situation rather than a general answer that could apply to anyone.

Build a founder-sized emergency fund

The standard advice, three to six months of expenses in reserve, was written for people with stable salaries. As a founder, aim for at least twelve months of personal expenses in liquid savings. Keep this separate from whatever you have allocated to the business. This will help ensure that an unexpected expense or a delayed funding round does not force a bad business decision.

If You Don’t Have Enough Savings, Consider Your Options

Review legal and tax obligations

When you set up your startup in the US, you have to make sure to file Form 83(b) which lets you pay taxes upfront on the restricted stocks at the value at grant rather than when the stock vests. This is because stocks appreciate upon vesting and you will have to pay ordinary income tax upon vesting. The form 83(b) has to be filed within 30 days of the grant and this is absolute. Missing filing the 83(b) within this 30-day period is usually irreversible and can lead to massive unexpected tax bills. Set a calendar reminder the day you sign your founding documents.

What are your retirement savings options as a self-employed founder?

As much as possible, do not deplete your retirement funds while building your startup. As a self-employed founder you have access to several tax-advantaged vehicles:

Understand your tax position as a self-employed founder

Your tax obligations depend on how your business is structured and how you pay yourself. If you’re operating as a sole proprietor, LLC, or partnership, taxes typically aren’t withheld automatically, and you may need to make quarterly estimated tax payments. Depending on your structure, you may also owe self-employment tax. By contrast, founders of C-Corporations who pay themselves a salary generally receive a W-2, with payroll taxes withheld just like any other employee. As a starting point, many founders set aside 25–35% of income for taxes, although the right amount depends on your income level and state of residence. Before choosing an entity structure, speak with a CPA who works with startups. The decision can have long-term implications for taxes, fundraising, and equity compensation.

Separate personal and company finances

It is wise to get a separate bank account for the business and not mix your personal and professional accounts. Even if you are the one funding the business, having a separate account helps you track how much you are actually spending on the business.

Think Carefully About Bootstrapping

Bootstrapping can be a great way to maintain ownership, but it also concentrates risk. Avoid putting all your personal assets into the company. Instead, allocate a specific budget for bootstrapping. 

Tie that budget to milestones, such as:

Consider part-time or side gig income

Even if you go full-time, you may want to consider part time or contract work on the side if finances are a crunch. You can consider consulting, contract work or freelance projects. This income can extend your personal runway while you are working on your startup.

Plan your founder salary if you have raised funding

Once you have raised funding, the question of how to pay yourself becomes more structured. Here is what founders typically do.

When should you start paying yourself?

Most founders begin paying themselves once the company raises funding or the business generates consistent revenue. The goal is not to match a big tech salary but to cover living expenses so you can focus fully on the company.

How much should you pay yourself?

Founder salaries typically depend on the funding stage and location. Many early-stage founders pay themselves enough to cover basic living expenses, often in the range of $70k to $120k in expensive cities, though this varies widely.

Investors generally expect founders to live comfortably but not extravagantly. Investors do not want you to be anxious and distracted due to money troubles, and encourage founders to take a decent salary that helps support their living expenses but otherwise want you to deploy all your resources into growing the business.

Think about equity as part of your personal finances

Your equity stake may represent the majority of your theoretical net worth on paper. It is also completely illiquid and might end up being worth nothing. Planwell’s lifetime financial projection shows your actual net worth at every stage of life, which is a useful reality check when equity makes your paper wealth look larger than your actual financial position. Plan your personal finances as if equity does not exist: maintain your savings, keep contributing to retirement, manage your fixed costs, and treat equity as optionality rather than income.

Protect your personal credit

Your credit score matters more as a founder than it ever did as an employee. Business credit lines and financing can touch it. Keep personal and business spending cleanly separated, pay personal obligations on time even when business finances are tight, and avoid personal guarantees until you fully understand what you are signing.

Plan around major life events

Buying a home, having a child, supporting an aging parent. These do not pause because you are building a company, and each one can materially change your personal runway overnight. Founders make the mistake of treating these as separate from their startup finances whereas they are deeply connected. Planwell is built for exactly this: you can run what-if scenarios to see how a single major decision like taking on a mortgage ripples through your cash flow, retirement timeline and other goals simultaneously, before you commit.

What to do if your startup fails financially

Startup failure is common and it is not something to be anxious about, but something to plan for financially. Often the startup experience can be a good value add to your resume and show that you have the drive and resourcefulness to handle high levels of ambiguity. So use your startup experiences and skills to obtain a leadership role at a company that values these skills.

In the process of building your startup, you may have done a lot of networking and met smart people. You can reach out to them and see if they can hire you or recommend you for a role. Many VC firms hire ex-founders and operators so do not forget to keep in touch with your VC friends.

The bottom line

Founders who navigate personal finances well tend to share one habit: they treat their own money with the same rigour they bring to their business. They know their runway, they separate their accounts, they plan for the costs that do not show up until they do.

The startup bet is already a bold one. The goal is to make sure your personal finances are stable enough that when hard decisions come, and they will, you are making them with a clear head. If you want to model your own situation, Planwell lets you input your exact income, expenses, assets and goals and get a personalized plan in minutes, without the generic advice that does not account for how founders actually live.

About the Author

Arundhati Sampath is the founder of Planwell, a financial planning tool designed to help people model life transitions such as startup journeys, career breaks, and early retirement.

The post How to Plan Personal Finances as a Startup Founder appeared first on BYVI.