§ Blog Build in public
Thinking out loud about enterprise AI that earns its place.
A working log, plus a below-the-line fejeton under each piece. Architecture, delivery discipline, and the occasional invented word.
- 16 Sidepromotion Shared bots keep chats private and skills shared. The risk sits in between: one person's correction quietly becomes everyone's answer. Promote on purpose.
- 15 Hold to release Any agent or skill ships only after named human hold-to-release and ownership of the outcome. New-law pacing is theatre. The desk still holds.
- 14 Instruction debt Instruction debt is the real skill-platform risk. Always-on AGENTS.md and skills that waste context, contradict, or grant unclear authority. Audit like code.
- 13 Approval fatigue Approval fatigue: the human gate fails when it is a click. Bounded escalate, named reviewer, revoke. Presence of an overseer is not oversight.
- 12 One stack to build the stack The personal harness is over-built so client platforms stay thin. Memory, tools and skills live in the business and can change without the builder.
- 11 Owner, verifier, gate A generated SKILL.md is not a skill until it has an owner, a verifier, and a gate. Agent-authored files can score worse than no skill at all.
- 10 Identity is the gate HiTL for callable tools is identity as the gate, not a human in the thread. Admin-once client, user-scoped token. A PAT is not a headless agent.
- 09 The description is the trigger Everyone repeats don't build an agent, build a skill. The scarce thing is a handwritten SKILL.md whose description actually fires when nobody is in the room.
- 08 Headless agents and the ones who still call them Below-liner included In a long conversation with people who ship real agent systems, the question that finally landed after forty minutes was simple. Can you explain what you mean by a headless agent? Preferably with an example.
- 07 Use cases to production, yesterday Below-liner included The enterprise AI loop — proof demanded three years ago, production demanded today. A field note on governance, platform patterns, and why bulk beats silo when the CEO wants APIs now.
- 06 Planneropause Below-liner included A glass installer with Harleys beat a six-year enterprise AI roadmap in one afternoon. Twenty euros a month, paid back in thirty minutes. The gap is not tools. It is planneropause.
- 05 Vibe consulting Below-liner included Everyone can generate AI roadmaps now. The scarce skill is vibe consulting — carrying overnight drafts into executive rooms and back, without pretending v1 is final.
- 04 Three experts, many tokens Below-liner included AI transformation is inverting the staffing model — fewer subject matter experts, heavy token use, human-in-the-loop QA. A field note on delivery economics from Big 4 to solo practice.
- 03 Two lanes for agent orchestration Below-liner included Enterprise agent platforms need a deterministic fast lane and a governed non-deterministic lane. A field note on loop engineering, model routing, skills, and the stakeholder shift.
- 02 Alignment before architecture Below-liner included In large enterprises, cross-functional alignment (not models) often determines whether AI reaches production. A field note on the ~40% nobody budgets for, and why business, risk, security, and technology must decide as one.
- 01 Building in public Below-liner included Why this blog exists — a working log of enterprise AI architecture, client-owned platforms, and the discipline of making AI earn its place in the business.