Tuesday November 17, 2026; 10:58 AM EST
How Claude code allowed me to think big

As before I asked Claude.ai to do the show notes. As always if you really want to know what I really said you have to listen to the podcast. More of my notes at the end. DW

Recorded while driving to breakfast in Kingston, the podcast is Dave Winer’s account of how Claude Code changed what one developer can attempt, and how that led to Bonbon, which is about to get its first public release. The Mac version is well tested, and the Linux version hasn’t even been booted, which he offers as a measure of his confidence in the process.

He opens with a grievance: the tech industry talks about him in the past tense, though he never stopped building. What he lacked was users, because writers moved to Twitter, Facebook and Substack. He compares the last 15 years to the PC market just before the web: a few dominant vendors, each with its own incompatible formats, and no room for anything new. The web broke that logjam because it needed no one’s permission: put a file in a place, type its address, and there it is.

Claude Code is the second such break, in his telling. He started using it at the end of 2025, after using ChatGPT as a consultant. His test was Frontier: he pointed Claude at the old C source and asked whether it could port the UserTalk language. By the next morning it had, and the object database, rebuilt on SQL, followed the same way. The hardest part was that Claude had quietly bent the design toward Python, and finding and undoing that may have been as much work as everything else. His general observation is that Claude understands existing things extremely well but tries to turn new ideas into familiar ones, which is where an experienced developer still earns his keep. It also forgets and needs patient reminding.

That success gave him the chutzpah to think bigger. The big companies’ defense, that a browser must do everything Chrome does, looks to him like the Maginot Line (the name he couldn’t recall): you can go around it. The trigger was Chrome’s warnings on scripting.com for not using HTTPS, which he regards as a lie about a site that is just writing, and as part of a pattern, after Microsoft, of big companies breaking the web’s role as a historical record. He also faults the EFF for not objecting.

Bonbon is his way around: a browser for a web cut back to roughly 1994. Its main features:

Markdown only: documents are Markdown with links, and any embedded HTML is stripped.

Reader-controlled style: the writer has no say over presentation; the reader picks one by dropping CSS files into a stylesheets folder.

No interruptions: no scripts, pop-ups or interstitials.

Open: it will be open source, with plugins for rendering other file types.

His goal is “locked open,” the way podcasting has stayed open for 20-plus years.

This is Dave again. As Claude surmised, the name of the French wall was the Maginot Line. And the third founder of the EFF was John Gilmore. My apologies for failing to call it up in real time. Shit happens all the time nowadays. ;-)