I watched our CEO and a senior technical leader debate “everyone ships code” in a Slack thread this week. The technical leader is exactly the kind of person you’d expect to embrace this—ex-founder, high agency, ships fast. He pushed back.

Our CEO cited Tobi Lutke’s tweet about shipping more code in three weeks than the previous decade. The response: he’s seen this at other companies, it leads to more time cleaning up PRs, PRs and lines of code are bad metrics, the real bottleneck is architecture and figuring out what to build. He isn’t wrong about any individual point. But I think he’s making a bet that’s already outdated. “Everyone codes” isn’t about turning PMs into programmers. It’s about removing the artificial barrier between thinkers and doers.

That barrier is now artificial. Most people at our startup are founders or ex-founders. This isn’t about effort. It’s about mental models.

Five days to one meeting

My brother called me last week. He works for a traditional software company—they build ERP systems for subdivisions and construction. He’d just walked out of a meeting where he used Claude Code in front of the entire dev team (~15 people) to build a feature they’d estimated at five days. He did the whole workflow during the meeting, all the way to PR. By his account, he overcame every objection about code quality. The code had unit tests. Their existing codebase has none.

The crazy part: it’s written in PHP and he’s not really a software developer. He’s just a smart guy who was curious enough to try. His goal wasn’t to convert all 15 developers. It was to convert one. Get one person to see it, then fight the next battle. The subtext was clear: expectations for output are about to change dramatically.

The goalposts keep moving

I’ve been watching this pattern for three years now:

2023: “LLMs can’t write real code.” 2024: “Okay, they can write code, but it’s throwaway quality.” 2025: “Okay, the code is good, but they can’t do architecture.”

Each time the barrier falls, the defense retreats one abstraction level higher. My colleague is running the 2025 playbook: “figuring out what to build and the architecture… that still happens in my head.” LLMs can help with architecture too. The skill that matters is critical thinking—asking “but why?” and “is this actually better?” That’s not correlated with years in the codebase.

I met the CEO of Nerdy recently—an education company. AI tutoring was about to eat his business. The incumbent developers kept telling him his ideas were too risky and would take years. Tons of tech debt, moving insanely slow. So he taught himself to code. Used Cursor, vibe coded a new repo, forced the devs to migrate. Now at least his business has a chance.

Done crying about bad code

In 2013, my company was acquired by Autodesk. I complained a lot about the quality of the AutoCAD web codebase—it was from a different acquisition, relatively new, and messy. One of the veterans there, Albert Szilvasy (28 years at Autodesk at that point), told me he was done crying about bad code. It was hopeless to fight and you just had to keep moving forward. I watched him sit in a board room in Tel Aviv while everyone was debating how to bring AutoCAD to web and mobile properly. He just started doing it—refactoring Windows dependencies out of the 40-year-old codebase by working through compiler errors one by one. Thousands and thousands of them.

He wasn’t waiting for architectural consensus. He was focused on the outcome he was trying to achieve, not the elegance of the code. And here’s what I observed: the elegance was a product of the good outcome. The multi-platform ambition caused a platform abstraction layer to emerge. It wasn’t clairvoyance from experts. It was a side effect of shipping. AutoCAD had more issues logged than you could ever possibly solve, code older than me, and yet it generated billions in revenue. Our “current state” is never as pristine as we pretend.

The value my colleague is defending—knowing the pipes, understanding architecture, having the schema in your head—used to be a moat. That’s exactly what LLMs are commoditizing. Even great people with high agency can have priors that don’t update fast enough in discontinuous change. The intuitions you built over a decade might be miscalibrated in months. Maybe I’m wrong—critical thinking is exhausting, and even if accessibility goes up, motivation might go down. But I’d rather bet on that than on the old moat holding.

My plan isn’t the cliché “start shipping.” It’s go experiment. No one knows how this turns out. The barrier between thinking and doing is gone—start testing what’s possible.

Process note: Drafted from a Slack thread + conversation. AI helped with structure and line edits. All claims, observations, and uncertainties are mine.