Learning Log

Learned

• 01Most AI integration work is data normalisation, not modelling.
• 02Cloudflare Workers genuinely make tiny backends fun again.
• 03A good llms.txt is becoming the new robots.txt for AI search.

Read

• Anthropic guide on building effective agents, second pass
• Several articles on how teams are evaluating AI features in production

Struggled with

Designing a clean evaluation harness for an agent loop where each step is non deterministic. The metrics that matter are not the metrics that are easy to measure.

Learned

• 01Generating 145 monthly articles teaches you about your own voice patterns.
• 02A tech enthusiast who writes every month for a decade is a different person than one who writes for a year.
• 03Hosting personal content on your own domain still matters in the AI search era.

Struggled with

Resisting the urge to over polish older articles. Past me made mistakes. Present me has to leave them.

Learned

• 01Daily Bible reading and daily writing share a structure. Both are small commitments that compound.
• 02Recording short devotionals on an iPhone with a Hollyland mic gives surprisingly clean audio.
• 03The hardest part of a daily practice is week three, not week one.

Learned

• 01Power Automate is more powerful than its reputation suggests.
• 02The line between low code and real code blurs when the integration problem is hard enough.
• 03Most teams underuse the desktop flows part of Power Automate.

Struggled with

Debugging a flow that worked in dev and failed in production with no useful error message. The lesson, again, is to add logging early.