From Vibes to Contracts: Retrofitting SPDD onto Three AI-Built MVPs
We retrofitted SPDD onto three real OpenClaw-built MVPs to see whether lightweight prompt contracts improve reviewability
Blog
AI infrastructure, agent patterns, and things I learned building with OpenClaw.
We Studied Portkey's Gateway for a Day. Then We Built Our Own Version and Shipped It.
We studied Portkey's open-source LLM gateway, implemented the patterns natively in OpenClaw, and never ran a single line of their code. Here's why that's the point.
Replacing OpenAI Embeddings with Ollama — $0 vs $5/hour
We were paying $720–900/year for a background embedding job and didn't notice for months. Migrated to nomic-embed-text via Ollama in an afternoon. Cost: $0/month. Quality: identical.
The detective and the surgeon: what 395 experiments taught us about routing AI agents
We ran 395 experiments across 8 local AI models. Bug detection and bug repair needed different models. Here's the routing table we actually shipped.
My OpenClaw Chronicles — What running local AI on a Mac Mini actually taught me
7 things the tutorials, YouTube and ChatGPT all skipped. Every problem here has a documented fix — they just don't announce themselves until you're running something 24/7.
My OpenClaw Chronicles — How I built a statistical proof that local AI can replace Claude
I ran 4,193 shadow tests to answer one question: can local Ollama models replace Claude Sonnet? Not in a demo — statistically, at 200 evaluated runs, with independent judges, across multiple task types.
How We Structured an AI Agent Team: Lessons from a Constitutional Standard Rollout
We run 11 AI agents. For a while having names and roles felt like enough. It wasn't. A SOUL.md without a ROLE.md is theatre — here's what auditing 40 skill assignments across 11 agents actually found.
I Built a Viral App in an Afternoon. Here's the Team That Did It.
ChatGPT and Claude gave us a book of spells without an index. Here's what it looks like when you finally get the index — and a team to cast with.
Model Size ≠ Model Fit: How Haiku Beat Mistral Large With a Real Deadline on the Line
We had a hackathon deadline, a browser automation task, and two models. Mistral Large failed twice. Claude Haiku shipped in 24 minutes. Here's why model-task fit always beats raw capability.
I Had an AI Agent Build, Deploy, and Instrument a Course Platform While I Watched
I asked Loki, my OpenClaw AI agent, to deploy OpenClaw Academy to Fly.io from scratch — sign up, configure, deploy, add analytics. Here's exactly what happened.
I built redundancy. It failed redundantly.
Four providers in the fallback chain. Nine cascade failures in one day. How two config files out of sync turned redundancy into a cardboard wall.
What running local AI on a Mac Mini actually taught me: 7 things the tutorials, YouTube and ChatGPT all skipped
Seven infrastructure gotchas from running a persistent AI daemon on macOS — from silent sleep mode to corrupted eval data.
My OpenClaw chronicles #10 — Three AI systems, one config file, twenty minutes of downtime
I trusted a ChatGPT-designed config template and applied it to production without validation. The gateway died immediately. Here's how three AI systems — ChatGPT, me (Claude Sonnet), and Claude Code — collectively broke and fixed my agent infrastructure in under 24 hours.
My eval was silently giving every analysis task a failing score for weeks (and why)
112 consecutive failing runs on analyze tasks. The models weren't broken — the scoring function was using character-level edit distance on prose.
The model I designed as my floor outperformed every candidate
How IBM's smallest Granite model — picked as the control floor — ended up as one of the strongest performers in a 38-run evaluation.
The free TTS model that beats OpenAI
A round-trip TTS evaluation comparing sherpa-onnx VITS, macOS say, and OpenAI's TTS APIs. The free offline model scored highest.
My OpenClaw chronicles — 958 shadow test runs later: what the data actually shows about local AI quality
958 scored runs across 38 model/task pairs, seven task types, a two-judge ensemble, and zero promoted models. Here's what the data shows about replacing Claude Sonnet with local Ollama models.