May 2026 – Ugly code runs successful companies

This month I want to highlight a series of Mastodon posts by Jonny Saunders. When the Claude harness source code leaked in March, 2026, the author took a deep look at the code to see what made Claude tick. What he uncovered was a complete mess of code, and his series of posts documents his shock and bemusement. There’s something to learn from this moment, and the lesson is not that Claude should be ashamed of their codebase. Instead, what we should learn is:

High-quality code is not required to make a great product. In fact, high-quality code may be a detriment to launching a business, as it can slow down finding product-market fit. Sure, Claude Code may be a mess of a codebase, but it’s also the backbone of a company with a hefty ~$1T dollar valuation. Why do we insist on creating “correct” code when we could create “wrong” code and achieve faster results? I have long since abandoned my ideals about delivering clean code as a primary output and now focus on how teams can deliver the right value, with clean code as a secondary consideration. When we’re threading the needle well, we’re a little unhappy with our code, but very happy with the tradeoffs we made to deliver value.

These reflections may sound like I have given up, but I haven’t. Instead, I am sharing my hard-earned realization that what matters most in business is 1\ how fast we move, and 2\ how much the customer enjoys the product. If clean code is a prerequisite to those outcomes, then we should create clean code. However, we mostly find ourselves writing “good enough” code to meet our speed-to-market constraints, and can rarely justify going back to clean it up later. With Claude’s source code leak, we have yet another data point to suggest that meeting customers’ needs matters more than anything else.

“I have read a lot of unrefined, straight from the LLM code, and Claude code is a masterclass in exactly what you get when you do that – an incomprehensible mess.”


Leave a comment