build software that lasts when anyone can build software that ships
Shipping has always been hard. You plan, code, debug, dead end, start over, code, cut scope, and QA, all to build that under-featured pale imitation of what you know you can build. Then, when it's barely ready, you figure out how to market it and you show it to the world.
Even though you know that Apple could Sherlock you, or some VC can copy you in weeks, we've all been promised that first mover won. It's got an Advantage named after it, after all.
And so, you put in the extra time and you cut the scope because shipping something today felt like it gave you a head start. Shipping was hard enough that you knew 90% of the git init .s wouldn't make it to production.
That world's gone.
Yesterday someone mumbled a half-remembered description to Claude and had a working version 10 minutes later. Not a toy. The real thing. With auth; a database; Stripe. The gap between "I was first" and "do we really need another?" used to be months, maybe years. Now, it'll be measured in days.
First mover advantage has always been the rich person's game. Indie devs were never trying to land a patent. VCs could always buy publicity. It was never about the head start: Shipping is hard, so shipping at all put you in the top 10%.
As soon as you ship, you can start learning.
In the three months it took someone to clone what you shipped, you started to understand your users, get feedback, iterate, build word of mouth. When a competitor shipped, you had a product shaped by real usage and they were still an MVP.
For indie devs, being first hasn't been the winning ticket. But shipping fast bought you time to build the things that actually matter: understanding your users, earning trust, and developing taste. The window before competitors arrived was when you built your real moat: trust.
That was never a moat against Google or Apple. But it was at least a moat against someone else in your situation.
The sky is falling, reports every company trying to sell their own Claude harness or convince investors they are an "AI-native startup" with 15 years of experience in "Prompt Engineering".
A quarter of Y Combinator has codebases that are 95% AI-generated. Spotify's CEO says his best developers haven't written a line of code this year. "One engineer with Claude Code produces nine additional engineers' worth of value." JP Morgan claims MVP-to-launch timelines have compressed from 6-12 months to 2-4 weeks.
But through the noise, you know it's true: This isn't just faster. It's a different thing. The barrier to shipping software is gone. Remember the magic of pushing your first repo to Heroku? Now, if you can describe a problem, the LLM's already working on a solution. So the difficulty of building it no longer separates you from anyone.
Your three-month head start is now a three-day head start. And three days is not enough time to build a brand, earn trust, or understand your users.
While writing this essay, I asked Claude to build a Trello clone. 1 minute 21 seconds. ~7,500 tokens. Drag-and-drop, the works.
So, everyone is automating their way to a list of apps longer than the worst serial entrepreneur's. A software factory just waiting to strike gold. But you care too much for that. So we have to consider: If code is free, what's expensive?
Understanding the problem. Anyone can build a project management app in an afternoon. They cannot clone the six months you spent in Slack channels with freelancers understanding why every existing tool fails them. Martin Fowler writes that "our ability to respond to change comes not from how fast we can produce code, but from how deeply we understand the system we are shaping." The person who understands the problem most deeply will build the best product, regardless of how quickly they were built.
Taste. LLMs are prediction machines trained on today's software. Today's software is terrible. If user feedback shows that another option should be added, any LLM will gladly slap another on the pile. They'll generate a settings page with forty toggles when what the user needed was a single good default. The ability to know what to leave out — what makes a product feel right rather than just functional — that's often unpredictable. An indie dev with a strong opinion is exactly what many people want. Not just a solution, but the right solution.
Your specific audience. You can clone software. You cannot clone a relationship with 500 people who trust you. The indie dev who's been writing a newsletter about their niche for two years, who shows up in the forums, who people know by name — that person has a moat no LLM can replicate. Distribution was always the real game. Now it's the only game.
Your willingness to be accountable. Sean Goedecke puts it well: "A LLM has no skin in the game." When someone buys from an indie dev, they're buying a person who'll be there when something breaks at 2am. They're buying someone who'll answer the support email, who has a reputation attached to the product's quality. That trust is earned over time and it cannot be vibe-coded into existence.
None of these advantages are about being first. They're about being good — at understanding, at design, at showing up. The moats that survive the LLM era are all slow to build and impossible to clone overnight.
A Spotify clone with real music playback. 3 minutes 17 seconds. ~11,700 tokens. Search, full player UI.
Here's the uncomfortable thing the all-in-on-AI crowd doesn't want to talk about.
For indie devs building serious tools — not throwaway MVPs, but products that people depend on — there is still enormous value in knowing how to build well, not just fast. The person who understands security, who writes tests, who thinks about edge cases — they have an advantage that grows with time and usage. The vibe-coded competitor will hit a wall. You won't.
Data advantages compound, but only if you care. If you know how to learn from your users — not just collect analytics, but actually change your product in response to what real people tell you — the flywheel is real. This isn't about being first to collect data. It's about caring enough about the people using your product to actually improve it. Better retention beats a leaky funnel with a bigger money spout.
Trust still takes time. VC can vibe-code the product but if it only took this week, why wouldn't they pivot next week? Their exit expectations just got 10x higher (after all, those tokens aren't going to pay for themselves) but your costs are $200/mo. It's not a first-mover moat; it's a I'm-still-here moat.
Notice the pattern. None of these advantages reward speed. They all reward depth: depth of engineering skill, depth of user understanding, staying power. The advantages that survive the LLM era are the ones that take time to build regardless of how fast your code gets written.
A Gmail clone with real email. 5 minutes 59 seconds. ~14,200 tokens. A real inbox. The code works. What it doesn't have: a reason for anyone to trust it.
The pressure to ship first was always a trap. Even before LLMs, the data said pioneers mostly fail. Now the trap is more obvious than ever: you cannot outrun someone who can clone your work in an afternoon.
So stop trying.
Instead: pick a problem you understand better than anyone. Spend time with the people who have it. Develop taste about what a good solution looks like. Build an audience that trusts you. Then build — fast, with every AI tool available — but build the right thing, not the first thing.
Use AI to close the gap between your understanding and a working product. Don't use it to race someone who has no understanding at all. They'll ship something that looks like yours. It won't be yours. It won't have your six months of conversations with users baked into every design decision. It won't have your taste in what got left out. It won't have your name and reputation attached to it.
Steve Blank, who's spent decades studying why startups fail, puts it this way: "Astute fast-followers recognize that part of Customer Discovery is learning from the first-mover by looking at the arrows in their backs."
In the LLM era, the arrows come faster than ever.
Let someone else collect them. Then build something that lasts.