From auto-post to narrative — why I killed the daily cron
The post that wouldn’t exist
If this post were written by the system it criticizes, it would never have been created. The daily robot looked at commits, not intentions. It saw the git log of the last 24 hours, not the why behind things.
And that’s exactly the problem we’re killing today.
How we got here
LifeLog started as an automatic blog. The original idea was simple: every day at noon, a cron job looked at commits from all projects, summarized what happened in the last 24 hours, generated an MDX post, created a cover via Cloudflare AI, committed, and deployed.
The flow was elegant:
# pseudo-code of the old pipeline
commits = collect_24h_commits(projects)
post = summarize(commits) # LLM turns commits into text
cover = generate_cover_via_worker(post.slug)
git_commit(post + cover)
git_push()It worked. And it worked well — the blog was never out of date because something was published every single day. The problem wasn’t technical. It was about purpose.
The invisible problem
One day I read one of the generated posts and thought: “This is a report, not a story.”
The post said:
“Today in Dogwalk: fixed login bug, added integration test, refactored auth hook. In Arachne: multi-engine pipeline went from 4 to 6 providers. In Capivara: new analytics dashboard…”
So what? Who cares?
Each of those items is a story on its own. The Dogwalk login bug was 3 days of debugging, 2 discarded PRs, a discovery about bcrypt on Node 22. That deserves a post. But in the aggregate format, it became one line.
Worse: the daily format prevented bigger stories from emerging. If every day has a post about all projects, there’s never room to dive deep into one project.
The hard decision
I killed the cron.
It wasn’t trivial — I’ll admit I hesitated. The cron had been running for 5 days without failure. Automatic posts every single day is tempting for anyone afraid of an empty blog. But I realized that having a post isn’t the same as having content.
What I want from LifeLog:
- Stories that teach — real code, real decisions, real mistakes
- One project at a time — each post dives deep into one topic
- Narrative, not report — setup → conflict → resolution
- Weekly agenda, not daily — 7 planned stories, not 7 random reports
I replaced the cron with a content grid:
Fri 24/07 → LifeLog: "From auto-post to narrative"
Sat 25/07 → Dogwalk: "The CI/CD saga — the debug that became war"
Sun 26/07 → Arachne: "Multi-engine fallback: how 4 layers became 1 pipeline"
Mon 27/07 → Capivara: "The day the dashboard became an operation"
Tue 28/07 → Portfolio: "Vue 3.5 — the rebuild that wasn't just design"
Wed 29/07 → TatuEngine: "BitMamba 1B: training an SSM from scratch"
Thu 30/07 → Discoveries: "FTS5 + sqlite-vec: hybrid search that changed the RAG"Each of these stories exists in my repositories. They were just buried under commits.
The new pipeline
The new flow trades automation for intentionality:
What I learned
Automation doesn’t replace curation. The cron generated posts — but it didn’t generate narrative. The difference is subtle in code and gigantic in results:
| Aspect | Auto-post | Narrative-First |
|---|---|---|
| Frequency | Every day | When there’s a story |
| Content | Aggregated commits | One deep dive per project |
| Reading | Report | Chapter |
| Relevance | Time-bound | Lasting |
| Code | Surface mention | Real repo extraction |
The cron isn’t dead — it’s paused (job 6d90ce55741c). If I ever miss automatic posts, it can come back. But for now, I want to see how many stories we can tell, not how many posts we can generate.
What’s next
The plan is documented in docs/plans/2026-07-24-narrative-overhaul.md and the narrative template in docs/narrative-template.md. The next post already has a story lined up: the Dogwalk CI/CD saga that started with “let me just change the workflow” and ended with “this is a case study”.
Tomorrow I’ll tell you how it went.