AI Made You 2x Faster, Not 10x. Plan Your MVP Like It.
The essay that topped Hacker News this week says AI coding buys you 2x, not 10x. Here is how a first-time founder should rebuild their MVP timeline, budget, and hiring plan around the real number.
What does "2x, not 10x" actually mean?
On July 31, the number one story on Hacker News was a short essay by developer Jacob O'Bryant called "2x, not 10x: coding with LLMs in 2026." His claim is simple. AI coding tools are a real productivity boost, maybe double, but the 10x dream is not arriving through better models alone.
That matters to you more than it matters to him. He has a job. You have a runway.
O'Bryant's argument rests on what he calls a staircase hypothesis. AI tools got widely adopted in 2026 because they finally became reliable enough to run in automated feedback loops: tell the model to build a button, have it click the button, check the result. Once a model clears that bar, being even smarter adds less than you'd think. To climb stairs you need to be tall enough for one step. Being tall enough for three steps at once barely helps.
So the honest planning number for a founder building with AI right now is 2x. Not 10x. And if you budgeted your launch around the bigger number, this article is your correction.
Why did this essay hit a nerve with builders?
Because it matched what people were seeing in their own repos. The comment thread filled up fast with developers saying the same thing in different words: the demo is magic, the maintenance is not.
There's data behind the feeling too. A 2025 study by METR took experienced open-source developers, gave them real tasks in their own codebases, and measured what happened when they used AI tools. Developers expected to be about 24% faster. They were actually 19% slower. On unfamiliar or simple tasks AI helps a lot. On complex work in code you know well, it can cost you time while feeling like it's saving you time.
That last part is the trap. AI coding feels fast. The feeling is real even when the speedup isn't. July's Hacker News mood leaned the same way, with builders talking less about shiny demos and more about stability and trust.
None of this means skip the tools. It means measure, don't vibe.
Where does AI coding actually save you time?
The short answer: anywhere you can define done in a way a machine can check.
AI tools shine when acceptance criteria are objective and verifiable. Make a signup form that sends a confirmation email. Write tests for this payment function. Convert this page to mobile. You describe the end state, the model iterates toward it, and both of you can tell when it's finished.
For a first-time founder, that covers a surprising amount of an MVP. Boilerplate, CRUD screens, integrations with well-documented APIs, test suites, admin panels. The stuff that used to eat weekends. This is why non-technical founders can now ship working products at all, and that really is new.
What the tools can't do well yet is answer judgment questions. Is this the right architecture? Will this structure still make sense at 10,000 users? Should this feature exist? O'Bryant's line is blunt: a working implementation used to mean a task was 80% done. Now it means 20%.
Where does the time come back out of your pocket?
Three places, and they're sneaky.
First, review and rework. The draft arrives in minutes, then you spend hours restructuring it so a human can maintain it. If nobody on your team does that pass, the cost doesn't vanish. It compounds quietly until small changes start breaking unrelated things.
Second, integration debt. Models don't know your whole app. They re-solve solved problems, duplicate logic, and wire things together in ways that work today and wobble in month three. Each generated feature sits on top of the last one's blind spots.
Third, verification. You still have to check everything that can't be checked automatically. Edge cases, security, data handling, the way the feature feels in a real user's hands. That was always the slow part. AI didn't touch it.
Add those back in and 2x is about right. Still a great deal. Just not the deal the launch-in-a-weekend posts promised you.
Your AI co-founder is ready when you are.
Foundra turns everything in this article into an actual plan. Validation, customers, pricing, launch. In one place, in your voice, in an afternoon.
Start free→3-day free trial. No credit card. Cancel anytime.
How should you plan an MVP timeline around 2x?
Take your gut estimate for the build, cut it in half, then add back a third for review and integration. In practice most founders land near 60 to 70% of their pre-AI timeline, not 10%.
A concrete way to do this: list every feature in your MVP and sort each one into two buckets. Bucket A is verifiable work (screens, forms, integrations, tests) where AI gets you real speed. Bucket B is judgment work (data model, pricing logic, architecture, anything touching money or user trust) where you should assume human speed. Apply 2x to bucket A only.
Then write the plan down where you can revisit it monthly, because your estimates will be wrong in instructive ways. A spreadsheet works. So does Notion, or a structured planning tool like Foundra that walks first-time founders through scoping a product plan section by section.
The founders who get burned aren't the ones using AI. They're the ones who promised investors and customers a 10x schedule and delivered at 2x.
What does 2x mean for your budget and first hires?
The good news: 2x is enough to change your cost structure. A technical founder can now cover ground that used to need a second engineer. A two-person team can ship what took five people in 2021. Your seed round buys more product than it used to, and investors know it.
The trap is hiring as if the tools replace judgment. They replace typing. If you're non-technical and your whole product was generated without anyone understanding it, your first technical hire will spend their first quarter paying down debt you didn't know you had. Budget for that openly, or better, bring someone in early enough to review as you go, even part-time.
And when an agency or freelancer quotes you a timeline that assumes AI makes everything instant, ask them which parts of the estimate are bucket A and which are bucket B. Their answer tells you whether they've shipped and maintained AI-assisted code or just demoed it.
Can you get more than 2x out of the same tools?
Probably, but not by waiting for smarter models. O'Bryant's bet, and it's a sensible one, is that the next gains come from retooling: better workflows around the capability that already exists.
A few practices are emerging. Write declarative specs of the end state you want instead of chatting back and forth. Run agents in sandboxed environments so you're not approving every action. Invest in tests early, because every test you write makes more of your codebase verifiable, which is exactly where AI is strong. Tell the model not to write comments and docs, then write them yourself; you'll catch what you don't understand.
Notice the pattern. Every one of those practices is a founder skill, not a model feature. The teams pulling ahead in 2026 aren't the ones with secret access to better AI. They're the ones who built verification into their process so the 2x compounds instead of leaking away.
Key takeaways
AI coding tools are worth roughly a 2x speedup on real products, not 10x. The gap between the demo and the shipped product is review, integration, and verification, and that work still runs at human speed.
Plan with two buckets: apply AI speed only to work with objectively checkable outcomes. Assume human speed for architecture, data, and judgment calls. Expect a working first draft to mean 20% done, not 80%.
Budget for a maintenance pass even if you built solo. Ask contractors how they estimate AI-assisted work. And treat process improvements (specs, sandboxes, tests) as the real multiplier, because they are.
Speed still matters. But in a year when every founder has the same tools, the edge isn't generating code faster. It's knowing which half of your product AI can actually build.
FAQ
Is the 2x number scientific? No, it's one experienced developer's calibrated estimate, but it lines up with measured results. The METR study found experienced developers were 19% slower with AI on familiar codebases while believing they were faster. For unfamiliar work, gains are real. 2x is a reasonable planning average.
Should a non-technical founder still build with AI tools? Yes. Shipping a testable product without hiring engineers is a real advantage. Just treat the output as a prototype until someone who codes has reviewed it, and don't put payments or sensitive data on unreviewed code.
Does this mean vibe coding is dead? No. It means vibe coding is a prototyping method, not a production strategy. Use it to validate demand, then rebuild the parts that must not break.
Will next year's models get us to 10x? Nobody knows. The staircase hypothesis says model gains matter less now that reliability thresholds are passed, so bet your roadmap on today's capability, and treat anything better as upside.
How do I explain a slower timeline to investors? Show the two-bucket breakdown. Investors have read the same essays you have. A founder who plans at 2x with clear reasoning looks stronger than one promising 10x.
You just read the theory. Ready to build the thing?
Foundra is your AI co-founder. It turns an idea into a validated business plan, a go-to-market, and your first 10 customers. In an afternoon, not a semester.
3 day free trial. No credit card. Works in 20 languages.