How I Chained TX SOS and USPTO to Verify Entity Names Before Filing
Before filing an entity name, I built a two-step verification pipeline — TX open-data API signal + USPTO Playwright knockout — but the most importan
Models, agents, memory and the systems around them.
Before filing an entity name, I built a two-step verification pipeline — TX open-data API signal + USPTO Playwright knockout — but the most importan
Adversarial verification can kill true facts. A 2/3 'refute' vote isn't ground truth — it's a hypothesis. Spot-check anything the user specifically aske
If your background daemon alerts when an error counter crosses a threshold, it'll oscillate silently. Latch a flag the moment you cross it — and d
A test suite written by the same model that wrote the code proves almost nothing about financial logic. Spinning up Grok 4.5 as an independent revi
Pre-generate your TTS audio when you store the text — not when the caller requests it. A fan-out burst will expose your rate limits immediately, and a l
Two required gates before any auto-fix lands: match on stable content identity (not LLM-regenerated labels), then check 'is this actually a defect —
A memory routing table is a second source of truth — and second sources lie. Build a contradiction pass, or it rots silently.
When one MCP server exposes five businesses and you only want three of them, a prompt exclusion is not a wall. Put the allowlist in code, at the bou
The Setup
I did a full evaluation of omnigent as a fleet harness for my agent setup. The verdict: don't adopt. Two of its three headline features I already had. T
Two mechanical memory fixes are safe to auto-apply. One semantic fix is not — because 'looks like a duplicate' and 'is a duplicate' are different th
'Could we use X instead of Y?' is not always a migration question. Sometimes it's a learning question. Treating them the same burns everyone's time.
The Day I Found Two Source-of-Truths
When two agents write the same file concurrently, the Edit tool's snapshot guard is the wrong weapon. Use an anchored, idempotent write instead — an
I built a feedback loop into my AI dashboard before I built the inspection surface. When I finally could see what it had learned, I immediately wipe
My automated CEO briefing kept dying at exactly 8 AM. The logic was fine — it was the clock. Reproduce off-peak first. If it works, you have a conte
Apollo, my Claude-backed dev agent, was booting with full identity and rules — but zero awareness of what we'd shipped last session. A single rollin
If you're using AI agents to build, your value isn't in the keystrokes. Price the orchestration, not the labor.
A health check that says ok tells you the process is alive — not that it's running the code you think it is. Make it report identity.
Two organizing axes = everything shows up twice. Pick ONE. Then delete.
Before you touch a committed sprite (or any binary asset), run git log -- <path>. The generator script sitting in your repo is a hypothesis, not
We gated Phase 3 of an Apollo memory build on a real benchmark before writing a single line of reranker code. The gate came back no-go. That was the bes
Pillow's multipage PDF save crashes if your Python build has no JPEG codec. img2pdf + getfixeddpilayoutfun is the fix. And every build that hits
What I Did on Purpose
The Setup
A script named gensprites.py in the repo looked like the source of truth for the assets. It was the thing that got replaced. We ran it anyway,
Two layers, one store. Deterministic for instant. Batch LLM for generalization. Never re-run the pipeline on every thumb.
My RAG system silently ignores PDFs. Every PDF I'd ever dropped in was invisible — no error, just gone. The fix: extract to .md, clean the junk, reind
Two AI savers writing the same memory files in parallel will eat each other's work. The fix isn't a lock — it's an idempotent, anchor-keyed rewrite and
Don't save memory aggressively. Capture it cheaply and commit immediately — flush expensively on demand.