A practical learning journal
Notes on software, products, languages, money, and the awkward details that only become obvious after doing the work.
Recent lessons, rewritten for clarity instead of left to disappear in a chat window.
npm packages can now have multiple trusted-publishing OIDC configurations. Here is how to split stable, prerelease and staged releases, when to keep staging, and how to migrate away from long-lived publish tokens safely.
GitHub CLI’s old Linux package-signing key expired on September 5, 2026. Here is who is affected, how to verify the replacement key, and the safest APT, DNF, Yum, Zypper, and Docker fixes.
AI infrastructure is tightening memory and storage supply, but the effect on consumer PCs is more complicated than 'AI is buying all the RAM.' Here is what is actually happening, when buying now makes sense, and when waiting is still reasonable.
Direct explanations, technical walkthroughs, and lessons from building things.
What do straight, arched, thick, thin and uneven eyebrows mean in traditional face reading? Explore the symbolism for fun and see how brow shape changes the story.
Production AI agents should be treated as workloads with delegated authority, not scripts holding copied human credentials. A practical pattern for unique identities, short-lived tokens, least privilege, user delegation, approval gates, and audit logs.
A staged search architecture for small AI apps: start with PostgreSQL full-text search, add pgvector when semantic recall is actually missing, combine both for hybrid retrieval, and move to dedicated vector infrastructure only when the workload proves it needs it.
GPT-6 Astra launched on September 3, but access is still staged and OpenAI's product pages differ on Plus availability. Here is what is confirmed for ChatGPT, Codex, Work and the API, what it costs, and when Astra is worth using over GPT-5.6 Sol.
No clean U.S. catch-up entry before Friday's open. Datadog still has a large cloud-software deficit but failed to confirm the Snowflake read-through, UPS's apparent repair reversed, while SMCI, MongoDB and Chevron fail event-contamination or relationship-quality tests.
U.S. tech enters Friday with futures roughly flat ahead of the August jobs report after a strong Thursday rally. Nvidia's $12.9 billion Hugging Face deal, Meta and Microsoft's relative strength, Tesla's Cybercab rollout, and Broadcom/AMD weakness create a sharply split setup inside tech.
What do round, oval, square, heart and long face shapes mean in traditional face reading? Explore the symbolism for fun, plus why your face rarely fits just one category.
A newly disclosed Git configuration flaw affected several AI coding agents because background Git commands could execute repository-controlled helpers outside the agent sandbox. The safest response is less about prompts and more about how you receive, inspect and isolate repositories.
Nvidia has agreed to acquire Hugging Face for about $12.93 billion. The practical developer question is not the headline price but whether Hugging Face stays genuinely open across models, clouds and competing accelerators.
A broad notebook still needs a map. Categories are the rooms; tags are the drawers.
