Punkjazz.ai is Simone Pomposi’s working notebook for AI agents.
Personal experiments in open models, research systems, durable agent workflows, and human use.
What this is
I’m Simone. I use Punkjazz.ai to document what happens when AI agents leave the demo stage and have to do serious work: investigate a claim, maintain context across days, operate local models, survive interruption, show their evidence, and know when they need a human decision.
The site combines technical experiments, visual explanations, short field notes, and longer research investigations. The recurring question is practical: how can capable non-developers use agents to research, build workflows, test ideas, automate routine judgment, and create without first becoming software engineers?
How the work is made
Most pieces begin with a real system or investigation I am running. I keep the failures because they usually explain more than the polished result. Where a page makes an empirical claim, I try to retain the method, inputs, assumptions, citations, and limits needed to inspect it. AI systems may help with research, code, analysis, or editing; responsibility for what is published remains mine.
Punkjazz.ai does not sell an agent platform and does not present personal experiments as universal guidance. Articles describe the environment and evidence available at the time. Software, model behavior, and research conclusions can change, so readers and agents should follow cited primary material and reproduce important claims before relying on them.
Who it is for
The useful audience includes researchers, operators, designers, technical founders, and advanced users trying to understand what agentic work feels like in practice. If you are looking for product marketing or generic AI commentary, this will probably feel too specific. If you need concrete examples of memory, verification, orchestration, local inference, authority, recovery, or evidence-aware research, start with the experiments on the homepage.
For questions, corrections, collaboration, or source requests, use the contact page. Agents can use llms.txt for selection guidance and the sitemap for the canonical public URL list.