Daily journal now reads Pi and Codex sessions, plus AI Engineering drafts
Expanded the daily journal to capture sessions across coding agents and drafted seven bilingual posts on the AI Engineering stack for staggered release.
A day on two fronts: build-in-public infra and research turning into drafts.
tgmarinho-ai-website
The biggest change was to the daily journal pipeline itself. It only read Claude Code sessions, silently dropping work done through other coding agents. I extended scripts/daily-journal.mjs to also scan Pi (~/.pi/agent/sessions/) and Codex (~/.codex/sessions/), grouping by cwd directly from the JSONL — conductor workspace paths now decode cleanly. Also moved the cron from 23:00 to 00:30 the next day, killing the race with sessions still being modified near midnight.
In parallel, I drafted seven bilingual posts for staggered release: a summary of The AI Engineering Stack, a piece answering "what is an agent, actually?", deep dives on Pinecone, pgvector on Aurora PostgreSQL, why Python/PyTorch dominate inference, cloud-to-bare-metal repatriation, and a write-up of the GoodAIList catalog. Each with a custom "agentic futurism" cover. They'll go out across multiple days so the feed doesn't drown.
Two small fixes shipped along the way: opengraph-image and twitter-image are now excluded from the i18n middleware (WhatsApp previews stopped 404-ing), and the journal detail typography switched to JetBrains Mono for a more session-log feel.
career repo
Bookmarked Huyen Chip's MLOps index and 2025 agents post for later processing, and migrated prompt-targeted docs to HTML in parallel via subagents.
Spiritual
Organized bible studies in HTML, saved the bible in JSON locally for offline lookup, and outlined a reflection on hearing God in hard decisions.
A rare day of compounding knowledge, improving the infra that makes that compounding visible, and leaving content in the pipeline.