The AI agent accountability gap nobody ships
Updated 17 October 2026. An agent has a task. A credential has an owner. A business process has someone responsible for the result. Those three facts need to…
The writing archive
AI governance
I write about AI agents, governance and product decisions. Usually because a question won’t leave me alone. Have a different take? I’d like to hear it.
58 essays & field notes58 articles
The EU AI Act’s High-Risk Deadline Just Got Pushed to 2027
Updated 17 October 2026. The high-risk AI deadline moved, but the AI Act did not disappear into a drawer. Treating the change as a blanket pause is how…
Sovereign AI needs operational control, not just location
Updated 17 October 2026. A local data centre answers a location question. It does not automatically answer who can access the system, change it or stop it. Sovereignty…
Federal AI preemption: ask what replaces the state rule
Updated 17 October 2026. Preemption debates are about who may make rules and what replaces them. A claim that a proposal will create clarity needs to explain the…
Voluntary AI governance needs evidence of participation
Updated 17 October 2026. “Voluntary” describes participation. It does not, by itself, tell you whether a programme produces useful evidence. The harder question is what participants provide and…
EU AI Act compliance 2026: are enterprises really ready?
Updated 17 October 2026. Readiness is a claim about a particular system and its evidence. It is not a percentage printed beside a green traffic light.Earlier versions of…
AI governance starts with an inventory you can actually use
Updated 17 October 2026. Before asking whether an AI system is ready for a regulatory deadline, ask whether the organisation can find it. An inventory is basic work.…
An AI deregulation headline does not settle your obligations
Updated 17 October 2026. A political signal is not a deployment assessment. “Deregulation” can describe an administration’s direction without answering what a particular organisation must do with a…
The voluntary AI governance gamble is follow-through
Updated 17 October 2026. The gamble in voluntary governance is not simply that somebody can decline to participate. It is that a serious finding may have no clear…
AI harm needs an investigation, not a blame shortcut
Updated 17 October 2026. Who should investigate, who should help the affected person and who may be legally liable are different questions. Collapsing them into a single blame…
AI accountability needs a maintained legal register
Updated 17 October 2026. Earlier versions of this article discussed Colorado’s changes as a pending delay. The legislature now records SB26-189 as enacted. A legal register needs to…
AI policy coordination should preserve a route to remedy
Updated 17 October 2026. Coordinating AI rules can be useful. The test is whether a person affected by a consequential decision still has somewhere effective to take the…
Sovereign AI procurement needs an exit test
Updated 17 October 2026. A service can be local and still be difficult to leave. Exit is therefore a useful sovereignty question: what can the organisation take with…
AI accountability must survive the prioritisation meeting
Updated 17 October 2026. It is easy to agree that accountability matters. The interesting decision arrives when the work competes with a feature somebody can demonstrate on Friday.Make…
AI Act simplification: check the control, not the label
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
AI governance needs an owner, not another deadline
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Colorado AI rules: documentation still has work to do
Updated 17 October 2026. Colorado’s 2026 changes should not be described as the disappearance of developer accountability. That earlier framing was too broad. The enacted SB26-189 includes technical…
A model evaluation is not a deployment evaluation
Updated 17 October 2026. Testing a model can reveal important capabilities and failures. It does not automatically establish what a particular product built around that model will do.Evaluate…
Turn AI governance into a control backlog
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
A later AI Act deadline is a budget decision
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Stop mixing legal uncertainty with unfinished product decisions
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
How to track AI regulatory changes without losing the plot
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
EU-aligned AI is a starting point, not a local compliance verdict
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
The autonomous enterprise still needs an exception owner
Updated 17 October 2026. An automated workflow can complete many steps and still leave one awkward decision unresolved. That is where the organisation discovers whether autonomy came with…
Measure AI governance by demonstrated controls
Updated 17 October 2026. An adoption percentage and a governance percentage may measure different things. Subtracting them can produce a dramatic headline without producing a useful management decision.…
When AI capabilities change, revisit the threat model
Updated 17 October 2026. A more capable model can change the risk of a workflow without changing its interface. The product still looks familiar. The assumptions underneath it…
AI governance belongs in the release process
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
AI standards: what to build while guidance evolves
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
AI monitoring needs a response, not just a dashboard
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Plan the critical path for AI governance
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Separate AI policy arguments from product requirements
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
How to assess an AI policy proposal
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Scenario planning for changing AI rules
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Prepare AI governance evidence for procurement
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Compare AI protections before and after a policy change
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Collect AI evidence before the deadline scramble
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Agent governance starts at the action boundary
Updated 17 October 2026. An agent’s most important boundary is often the point where a suggestion becomes an action. That transition deserves a product decision, not an accidental…
Deferring AI governance work is a decision
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
AI governance for small product teams
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Make AI governance an operating routine
Updated 17 October 2026. A governance framework becomes operational when it changes what people do on an ordinary working day. A document launch is not the same milestone.Put…
Reuse AI frameworks, not unexamined conclusions
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Answer the AI assurance question your customer asked
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
A written AI policy is not evidence of enforcement
Updated 17 October 2026. A written policy is a starting point. The question is whether a person or system behaves differently because it exists. Earlier unsourced market percentages…
Budget for the operation of responsible AI
Updated 17 October 2026. Responsible AI has operating costs after the launch deck is finished. Someone has to maintain tests, review exceptions and keep the evidence aligned with…
Make AI risk classification reviewable
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Manage AI governance exceptions properly
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Read AI readiness surveys before repeating the numbers
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Know your agent: build a useful identity record
Updated 17 October 2026. “Know your agent” is a useful product question when it means more than giving software a name. The record should explain why the agent…
Agentic AI liability: design the evidence before the dispute
Updated 17 October 2026. Imagine an agent making a consequential change that nobody intended. The first useful question is what actually happened. A confident explanation without records can…
Manage the AI compliance transition
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Prepare AI evidence for scrutiny
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Light-touch AI governance still needs a clear boundary
Updated 17 October 2026. A light-touch process can be appropriate for a limited use. An undefined boundary cannot. The useful question is whether the process matches the consequences…
AI transparency should help people decide
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Map the dependencies behind AI governance
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
There is no single AI compliance clock
Updated 17 October 2026. This article has been revised to separate the policy timeline from the product argument. See the current AI Act timeline for the category-specific dates…
Human oversight needs time, context and authority
Updated 17 October 2026. Putting a person in a workflow does not establish meaningful oversight. The person needs enough information, time and authority to make a different decision.Test…
Read AI vendor terms beside the product promise
Updated 17 October 2026. A product page describes the promise. The terms describe parts of the arrangement. Read them together before deciding what the organisation can safely depend…
AI should support judgement, not replace the problem
Updated 17 October 2026. People forget the real problem they wanted to solve. Instead of using AI to complement their judgement, they start depending on it to supply…
No essays match that search. Try another word or choose All.
In conversation
Shorter thoughts.
Longer conversations.
Some thoughts start as a post. The replies often make them more interesting. You’ll find those conversations on LinkedIn.
Read my posts ↗Customer-centricity; Products and Cultures ↗