I've built 17 apps last year at March. The year before that, I built two. The bottleneck used to be code. Now the bottleneck is everything else.

The problem no one warned me about

AI has made building software absurdly fast. Tools like Cursor, Claude Code, and GitHub Copilot have compressed what used to take weeks into days, sometimes hours. I sit down with an idea, spin up a project, wire the pieces together, and before lunch I have something working. That sounds like a dream. And it is, until you realize you now have 17 apps and no users. The bottleneck has shifted. Writing code was never the hard part of building a successful product, but it was the part that consumed the most time. Now that it doesn't, the real challenges are exposed: marketing, distribution, positioning, support, iteration based on feedback. Those things still take the same amount of time they always did. I am building faster than I can market the things I build.

The numbers back this up

The explosion in AI-assisted development is not just anecdotal. Around 41% of all code written in 2025 was AI-generated, and current projections suggest that number will cross 50% by late 2026 in high-adoption organizations. Google reported that 25% of their code is now AI-assisted. The Stack Overflow 2025 Developer Survey found that 84% of developers use or plan to use AI coding tools. Meanwhile, 4% of all public GitHub commits are currently authored by Claude Code alone, with projections reaching 20% of all daily commits by end of 2026. We are not imagining this. Code really has become too fast.

Vibe coding gets you somewhere, but not everywhere

There's a phrase floating around for this new way of building: "vibe coding," coined by AI researcher Andrej Karpathy. The idea is that developers have shifted from writing code line by line to orchestrating AI-powered workflows. You describe what you want, the AI generates it, you steer and correct. For experienced engineers, this is powerful. You already understand architecture, trade-offs, and system design. AI handles the tedious parts while you focus on the decisions that matter. For someone without that foundation, vibe coding can only get you so far. You can prototype something impressive, but when things break, when you need to debug a subtle race condition or reason about data consistency, the AI can't always save you. A recent METR study found that experienced open-source developers actually took 19% longer to complete tasks when using AI tools, even though they believed AI had sped them up by 20%. The gap between perception and reality is real. This is not a contradiction. AI makes the easy parts trivially easy and leaves the hard parts just as hard. That compression changes who can build, but it doesn't change what it takes to build well.

Software engineering isn't dying

Every few months, someone declares that software engineers are obsolete. The data tells a different story. Yes, entry-level hiring has dropped. Signalfire reported that new grad hiring fell 50% from pre-pandemic levels by mid-2025. Coding bootcamps have struggled. Dario Amodei, CEO of Anthropic, suggested AI could displace half of all entry-level white-collar jobs in the next few years. But the demand for experienced engineers who can think in systems, who can design architectures, manage complexity, and make judgment calls, that demand has not gone away. If anything, it has intensified. When anyone can generate code, the value shifts to the person who knows what to build, how to structure it, and why one approach beats another. The role is evolving from "writer of code" to "conductor of software." Just as cab drivers didn't vanish but adapted to platforms like Uber, developers are adapting to a world where AI handles output and humans handle orchestration.

The real bottleneck

Here is what I've learned from building 17 apps in three months: Building is no longer the constraint. I can ship a functional product in a weekend. That used to be the hard part. It isn't anymore. Marketing is the constraint. Getting people to notice, try, and care about what you've built takes just as long as it always did. There's no AI shortcut for building trust, writing honest copy that resonates, or showing up consistently in the places your audience already hangs out. Taste is the constraint. When everyone can build, the differentiator becomes what you choose to build and how you shape the experience. That requires judgment, empathy, and a deep understanding of the people you're building for. Maintenance is the constraint. Seventeen apps means seventeen things that can break, seventeen sets of dependencies to update, seventeen support channels to monitor. The cost of building something has dropped, but the cost of keeping something alive has not.

What I'm doing about it

I'm slowing down. Not on the coding, but on the shipping. Instead of launching everything I can build, I'm picking the two or three projects that have the best shot at reaching people and going deep on those. The rest go into a backlog. The instinct when building is cheap is to build more. The smarter move is to build less and invest the time you saved into the things that actually make a product succeed: talking to users, writing about your work, iterating on the details that matter. Code has become too fast. Everything else hasn't caught up yet. The developers who win from here won't be the ones who ship the most, they'll be the ones who ship the right things and do the slow, unglamorous work of making people care.

References