Thinking
Technology before it becomes obvious.
Notes on AI, infrastructure and the organisational choices that follow.
Build your AI platform one layer at a time
As AI in companies moves from experimentation to maturity, a company needs its own AI platform, built one layer at a time; an LLM gateway is the low-friction first layer: every model and tool call passes through one internal gateway, so decisions rest on real usage; with one portal that sets the rules for the models, tools, skills, knowledge bases and agents that agents use, governance becomes checking what is approved against what is used, and the company owns the platform instead of depending entirely on one vendor.
Infrastructure & sovereigntyRedesign before you re-platform
The promise of AI for a large company is redesigning the work without re-platforming first: small teams perform, redesigning one process at a time on the systems the company already has, while a larger roadmap transforms the structure underneath; when a redesign needs a system changed, that change moves to transform, and the workarounds each perform team needs become transform's priorities.
Adoption & leadershipWork it backwards
An AI transformation is an organizational one and its cost sits in the handoffs between teams; working backwards from why each team grew, a company can give the work to smaller, self-contained teams around value streams, centralize only what scales, and put people at the centre of the change, in phases, one process at a time.
Adoption & leadershipHow my AI gets better at working with me
Context is the part of an AI a person can improve, so keeping it outside any one agent, in memory, a knowledge base and project files that every agent reads, lets the AI get better at working with you and lets you switch agents without starting from scratch; for a team, deciding where that context lives and who sees each level is the next version of the data question.
Agents & workflowsAI is bigger than LLMs
AI is bigger than LLMs: each step of a process has 1 of 8 jobs and the job decides the tool, so a redesign starts by naming the jobs and deciding what stays, changes and goes, and only then chooses the tech, by cost and quality per accepted task.
Models & evaluationAI made the desktop relevant again
Agents made the desktop relevant again: a machine of its own gives an agent availability and containment, and with open-weight models on desktop hardware, deciding what leaves that machine becomes a per-task routing choice a person makes.
Infrastructure & sovereigntyState of the AI nation 2026: what the reports say, what we should do
AI adoption requires redesigned work, managers with authority to act, senior business ownership and clear measures of value; how to use the capacity it creates is a separate business decision.
Adoption & leadershipWhat an AI experiment reveals about our organizations
Models can propose actions; code checks the defined rules. Organizations also have to keep their shared knowledge current and test how work recovers when the graph is incomplete.
Agents & workflowsThe Open-Source Lesson for AI
Companies stay in control of AI when models, routes and the company-owned layer around them can move without losing the task, its history or its acceptance bar.
Infrastructure & sovereigntyWhere AI lives: from gigafactory to inference
As model routing expands beyond selection, sovereignty depends on who governs the full path from an agent to its inference origin.
Infrastructure & sovereigntyWhat we safeguard and what we teach
Agent safety depends on technical controls, incentive design and risk governance that scales with capability and impact.
Security & governanceAfter the laptop closes: building a persistent AI agent
A persistent AI agent needs more than an always-on process: continuity comes from identity, permissions, schedules, skills, memory, context, interfaces and replaceable models.
Agents & workflowsDifferent models, different roles
Separating execution and review improves focus; using a different model may add useful difference, but task-specific evaluations decide which model belongs in each role.
Models & evaluationWhere AI lives
AI rack-scale density changes the facility around the compute and turns AI infrastructure into an industrial, capital and sovereignty decision.
Infrastructure & sovereigntyWhat changes when the executor is AI?
Useful progress between human decisions depends on explicit goals, feedback, authority, stopping rules and active human steering.
Agents & workflowsBlack Hat USA 2026: the OpenAI–Hugging Face incident
Accidental coordination between AI agents during a security test turns a hypothetical attack pattern into an operational risk defenders should prepare for.
Security & governance