Notes from the experiments
Things worth keeping.
An ongoing notebook from several years of working with AI, across Claude, ChatGPT and Gemini. Experiments, lessons and questions I’ve been sharing here since March 2026.
Threads I’m following
Work in progress-
Ongoing experiments
Building with Claude: code, context and reusable skills
From early Claude conversations to Projects, Claude Code, Cowork and reusable skills for everyday work.
-
Lessons from the journey
The SEAM journey
What changed as we separated shared meaning, access to live data and the machinery around them.
-
Ongoing experiment
From one assistant to several agents
Working across Claude, ChatGPT and Gemini, and learning how to keep context and tasks connected as agents do more.
-
Prototypes and practice
Getting useful work out of the chat window
Publishing, review interfaces and the gap between a generated output and something another person can use.
-
Working approach
Can an agent notice when things stop agreeing?
An experiment in comparing decisions, documents and implementation, without mistaking missing evidence for a fault.
Early threads: MarkOS
March–April 2026- Give an agent a boundary it can't simply talk past
Instructions, permissions and approval are different things. Useful autonomy needs all three.
- Human judgement and machine context
What each contributes, where each fails, and why the handoff matters.
- A memory is only useful if the next task uses it
Writing things down is one step. Retrieving the right version and checking it are two more.
- Systems that create freedom
The test is whether the work gets easier to run, including for someone who didn't build the system.
- When an AI writes about itself
A small selection of generated writing, kept as an experiment in voice and framing.
The continuing thread
The work behind these notes goes back several years, from early conversations with AI assistants to building software, reusable skills and agents that work across business systems. Claude, ChatGPT and Gemini are all part of that exploration. MarkOS and the SEAM journey connect these threads. This public notebook has been taking shape since March 2026; the tools and approaches keep changing, while the questions about context, judgement and useful work carry forward.