mod_wsgi Creator Graham Dumpleton: AI May Do the Surface Work, but Deciding What Isn't Good Enough Is Still Down to Me
Graham Dumpleton, creator of mod_wsgi, argues that in his AI-assisted development of Muru the human's remaining role is judging sufficiency—deciding what output isn't good enough yet—and that this judgment draws on twenty years of accumulated domain-specific failures.
Graham Dumpleton, who created mod_wsgi twenty years ago and has spent that entire period thinking about how to run Python web applications in production, has a specific answer to the question of what humans still do when AI writes the code: they decide what isn't good enough yet.
His new project, Muru, is an AI-assisted Rust server designed to serve ASGI and WSGI Python applications. Over three weeks in November 2026, he published an unusually detailed series documenting its construction—worker architecture, import strategies, HTTPS handling, canary deployments, Kubernetes integration, scaling to zero, toolbar tracing. The writing is part of his development process, not a separate documentation effort afterward.
Dumpleton's claim is narrower than the general assertion that AI cannot replace humans. He does not say AI produces bad code or that human review catches bugs in the ordinary sense. His stated position is that the AI may be doing a lot of the work on the surface, while his own contribution is the judgment of sufficiency—knowing what isn't good enough yet, based on twenty years of getting things wrong with mod_wsgi. The experience he names is failure experience: accumulated knowledge of what goes wrong in production Python servers, which informs the decision of when output crosses the threshold of acceptability.
This is one practitioner's account of his own project, and it should be read as such. Dumpleton is describing what works for him, in a domain where he has twenty years of failure data. It does not follow that every AI-assisted engineering task requires equivalent accumulated expertise to judge output quality, or that Muru will succeed in production. His commitment is explicit: he intends to keep polishing until it is something he would trust. The argument concerns the quality-control layer, not a prediction of the result.