This section is deliberately the lightest-weight part of the site: applied notes on AI developments that actually change something about building, hosting or ranking a website — not a summary of every model release. Most of it exists to feed short commentary elsewhere, not to carry its own heavy publishing schedule.
Key takeaways
- Filtered, not comprehensive. Coverage here is deliberately narrow — one lens, applied consistently.
- Practitioner-written, not aggregated. Notes come from actually using these tools to build things, not from summarising press releases.
- Feeds distribution, not the reverse. This section's real job is short-form material for LinkedIn, not standalone traffic.
- Publishes irregularly, on purpose — unlike Hosting and Visibility, which run on a steady cadence.
The filter this section applies
Three questions decide whether something gets written up here:
- Does it change how something should be hosted or deployed? New agent tooling, new deployment patterns, changed pricing from a major AI-adjacent platform.
- Does it change what gets crawled or cited? Shifts in how AI systems retrieve and attribute content — the same territory as the technical SEO pillar, viewed from the "what just changed" angle rather than the "how it fundamentally works" angle.
- Does it change how software actually gets built day to day? Coding agents, context management, tooling that changes a working developer's actual workflow — covered as a practitioner, not a spectator.
What doesn't make the cut. General model-capability announcements, benchmark
leaderboard shuffling, and speculation about AGI timelines. Plenty of other places cover that
territory; duplicating it here wouldn't be worth a reader's time.
Notes published so far
- 01 How AI coding agents actually decide which files to read The mechanics behind context selection, from map-then-fetch to on-demand retrieval →
- 02 What MCP (Model Context Protocol) Actually Does Protocol, not product: how a standard tool-calling convention replaced one integration per agent per tool →
- 03 Prompt Caching: The Real Mechanics Behind Lower AI API Costs What's actually cached, why writes cost more than reads, and why prefix order decides whether it works at all →
- 04 Why AI-Generated Code Fails Silently, and How to Catch It Confident-wrong output happens because fluency and correctness aren't the same property, and what actually catches it →
- 05 What SWE-bench Actually Measures (and What It Doesn't) Real GitHub issues, a fixed set of Python repos, pass/fail on tests — what that setup captures and what it structurally can't →