Is your 2023 startup already legacy software?

On Saturday, investor Josh Elman talked to several founders who have spent the last three years building their products. They started with code written by hand. On Sunday morning he posted what they told him: they "feel behind and slower than companies that started this year with all code written by agents," and he has "never seen such an order of magnitude difference."

By Sunday evening the post had more than 170,000 views and 860 likes, and the timeline had turned it into a sport.

• The hot take: Jake Gold predicted that "even OpenAI and Anthropic are going to start looking like pre-AI 'legacy' companies," and said startups that are pre-product-market-fit and started before 2026 "should probably just re-found the entire company/code/team."
• The counterpoint: Richard Artoul quoted Elman and pointed at the real difference: people "who have customers already with real enterprise workloads they can't break" move slower than those who don't. "Nothing slows you down more than having paying customers."
• The fine print: In the replies, Elman said the faster companies he's hearing about have "real traction" and are "not built in a weekend slop," but he didn't name them because he was told in confidence. He also said rebooting the code "is a good way to reboot the team too."

What this is and isn't. It's a handful of founder conversations relayed by one investor, plus reactions. Nobody published numbers, and "feel behind" is a feeling. Treat "order of magnitude" as a vibe, not a benchmark.

Our take. Artoul has the sharper point. A company that started this year with agents has no customers to break, no migrations, and no on-call pager, so of course it ships faster. That doesn't mean the older company should throw out the code it has. The practical move is the one in between: let agents take over the boring layers (tests, glue code, migrations, internal tools) where your existing customers won't feel a mistake, and keep a human reviewing anything that touches their data. If you're pre-product-market-fit and your codebase is mostly guesses you'd make differently today, a rewrite is cheaper than it used to be. If people pay you, they're the reason you're slow, and also the reason you exist.

What we did and didn't do: we read the three posts linked below and Elman's public replies on October 25, 2026. We didn't talk to the founders he mentions and can't verify the speed difference. Not sponsored; no affiliate links.