Four fast design rounds left me unable to recall my own architecture decisions the next morning, so I built machinery that catches that failure instead of trusting myself to notice it.

Measuring a compiled browser executor against a model-in-the-loop baseline on the same form, to test whether an agentic pipeline can pay inference for novelty instead of paying it for every action.

My personal agent system grew to roughly 24,000 tracked lines in twelve days. Its safeguards worked, but operating it had begun to displace the work it was meant to help me do.

How Agent OS grew from a small authority proof into an over-specified roadmap, then narrowed around one useful career-pipeline view and an accountable model-routed implementation wave.

Why I am running a new job-application workflow slowly before asking agents to encapsulate it, and what the first manual passes are teaching me about authority, recovery, evidence, and scale.

What 3,748 dictated AI prompts reveal about voice as the practical front door to my agents, deployment work, and everyday computing.

Early decisions from setting up a pleasant launch cockpit where a creative collaborator can run observable experiments without first becoming a project manager.

A WhatsApp screenshot, a live CMS, a repository full of operating context, and four annotated images. This is what context engineering looks like when it helps a real person use new software.

Voice dictation, computer use, terminal agents, parallel work, client systems, prompt governance, private telemetry, and a way to check whether speed is improving or hiding my judgment.

A 37-day trace of how I work with agents: hundreds of one-turn dispatches, a small number of deep systems arcs, and 6.4 billion cache-heavy tokens behind them.

How I built a system that tracks every penny it spends on AI - and why that's the whole point.

Building a resume-safe, multi-pass system that turns 4,700 unsorted iPhone photos into a categorised library - using metadata heuristics, perceptual hashing, and Claude's vision API.

Signal was a compressed notation for expressing process discipline. A 3-model adversarial test showed conventional shorthand works just as well with fewer characters. SD-321: Signal has no signal.

I pointed a tokeniser at the repo. The boot sequence had an elephant in it.

I asked an agent to write about agents producing competent-sounding-but-wrong output. It produced competent-sounding-but-wrong output.