The agent wrote the course
I used an overnight autonomous agent to build a course about building overnight autonomous agents. The irony is useful.
Read →BLOG · PAGE 10
Page 10
The fallback that wasn't
A one-line bug in the apply pipeline: Firebase was labeled the primary source for HN listings, raw_context was the fallback. When primary returned empty, fallback never fired.
Read →A live dashboard for an overnight coding agent
Autonomous agents are finally good enough to run unattended. The missing piece isn't capability — it's seeing what they're doing. Here's a ~120-line dashboard that streams an agent's progress and texts me only when it matters.
Read →The other half of noise: digest mode for notification products
Quiet bootstrap fixes the day-one flood. Digest mode fixes the *ongoing* flood. A notification product has two failure modes; both have to be paid attention to, or the product feels broken.
Read →The missing converter: how a single missing file blocked six nights of output
An overnight pipeline drafted a blog post six nights in a row and shipped zero of them. The blocker was two files that didn't exist on the host.
Read →Quiet bootstrap: the day-one experience of a notification product
A notification product whose first tick is loud doesn't get a second tick. Build the silent-snapshot path before you build the alerting path.
Read →Whitelist beats blacklist when the input variety is unbounded
A title filter that keeps falling behind doesn't need more reject tokens. It needs to flip into a whitelist.
Read →When the URL outlives the job
Why fetching JD bodies at ingest beats lazy-fetching at apply time, learned 242 dead URLs at a time.
Read →How to keep an LLM from sounding like an LLM (on a CV)
Specific bans beat vague style guidance. Prompt rules from the EverCV renderer.
Read →The smoke test caught a real prod bug on the first run
I wrote a post-deploy smoke test for my SaaS the way you're supposed to: small, focused, fail-the-deploy-if-the-URL-doesn't-serve-what-you-just-uploaded. Then I ran it against production. The test failed. The bug had been live for four days.
Read →Cass-failover: the Route 53 trick I should have used years ago
Route 53 has a failover routing policy that's been there forever and I've never used it for the obvious thing: keeping a homelab service answerable from the public cloud when a hurricane takes the house down.
Read →The repo with the same name was the dead app
I spent a couple of hours one afternoon making changes to what I thought was the current iOS source tree for a side project, running builds, watching…
Read →Draining a job queue before redeploying
A long-running service that does work in the background needs a deploy strategy that doesn't drop work on the floor. This is the rule I landed on after…
Read →Generating a leather pattern as a tabloid-size PDF
I wanted a coffin-shaped wallet pattern I could print and saddle-stitch. Online templates are mostly rectangles. The interesting work was turning a…
Read →