Three pieces to start with.

Beyond the demo: what it actually takes to put AI into production.

A working list of the things you discover only after the first customer signs — evaluation, fallback paths, support tooling, monitoring, the long tail of model failure modes — and how to plan for them up front instead of in postmortems.

Coming soon

Human-in-the-loop is not a fallback — it's the architecture.

How to design HITL workflows that scale instead of becoming the bottleneck. Notes from running multi-LLM HITL pipelines for an AI workforce platform — including when to keep humans in, when to phase them out, and how to instrument the handoff.

Coming soon

From RFP to ROI: a CIO playbook for AI pilots that don't die.

Why most public- and enterprise-sector AI pilots stall in the gap between procurement and production — and what changes when you treat the pilot as the first 90 days of operation rather than a feasibility study.

Coming soon

Other recent writing.

What technical diligence actually looks for in an AI acquisition.

Beyond "does the model work?" — what to look for in data rights, partner dependencies, eval harnesses, model-IP ownership, and the team's ability to maintain quality at the buyer's scale.

Coming soon

Why the integration team is your AI strategy.

The best AI product in your category loses to the average AI product that integrates cleanly with the systems your customers actually run on. A short argument for putting integration on the roadmap before the model.

Coming soon

Hiring for AI product teams in 2026: skills that age well.

Which skills compound, which depreciate fast, and how to write scorecards that don't assume today's stack will still be the stack 18 months from now.

Coming soon

Lessons from a $25M public-sector program — translated for AI.

What a UK welfare IT transformation taught us about change management, partner negotiation and benefit realization — and how those lessons map onto putting AI into a regulated, large-organization environment.

Coming soon

The roadmap question every AI exec gets wrong.

"What should we build?" is almost never the right opening. A better starting question — and how to use the answer to sequence a roadmap that survives its first six weeks.

Coming soon

What 95% retention on an AI platform actually requires.

Spoiler: it's not the model. Notes on customer-success patterns from an AI workforce platform serving 1M+ users — the rituals, the metrics, and the conversations that hold the number together.

Coming soon

Get new pieces in your inbox.

Once or twice a month. No marketing list — just the writing. Unsubscribe in one click.