Writing

Solving LLM lock-in, governance, and cost control for modern enterprises. Why a governance-first control plane — not another multi-provider router — is what organisations need when commercial models, private LLMs, and internal RAG all have to coexist under one policy.

The hardest part of building production software with AI agents isn’t getting them to write code — it’s making the process reliable, auditable, and scalable. How I run a Strategist + Coordinator + Worker protocol (v3.8), the guardrails that catch mistakes early, and what it took to ship end-to-end voice booking in SlotSense.

The real prompt-engineering skill isn't writing one good prompt — it's capturing what you and the agent learn while working together, so the next session starts smart instead of relearning. How I turned that into a versioned collaboration protocol, what goes in it, and why the same discipline also cuts token costs.

What retrieval-augmented generation actually is, why enterprises need it, and what it takes to make one good — drawn from building a production HR-RAG system on GCP. How hybrid retrieval works, how to measure quality, the tuning that actually moves the needle, and the guardrails and costs that decide whether it survives contact with real users.

What actually changes when you build software with AI agents — drawn from several production projects. Where agents excel, where the integration points break, the pod-programming team model, a three-level agent setup with human gates, and an honest look at skill, trust, technical debt, and security.