Articles
Essays, research notes, and architecture memos.
Writing on AI-native engineering systems, governable autonomy, context architecture, and enterprise platform strategy.
Architecture - 2026-10-09 - 6 min read
One caller, two approvers: the bypass a reader found in my approval code
A reader showed that the reference HTTP server in aine-control-plane let one caller grant itself the approver role and fill a two-person quorum by sending two names. What the fix does, what it costs, what is still open, and why a pilot's stop condition for bypassed approval depends on it.
Leadership - 2026-10-09 - 9 min read
Choosing a First AI Pilot Across Three Workflows
Choose a first AI pilot with three synthetic workflows, a reusable assessment table, clear handoffs, and a baseline that includes review and rework.
Architecture - 2026-09-30 - 8 min read
A frozen field with one witness is still a claim
A field captured at the call site and frozen, exactly by the rule, can still prove nothing, because nothing else in the record holds a value that could disagree with it. How to find the second witness for each field, the field in my own system that had none, and where a second witness goes blind.
Architecture - 2026-09-28 - 7 min read
The credential your record names is not the one that made the call
A record that reads the credential from configuration is wrong for exactly the calls a rotation left behind, and it reads as complete while being wrong. What the record has to hold instead, where the bound stops working, and the field my own decision record is missing.
Architecture - 2026-09-25 - 9 min read
What your records cannot tell you six weeks later
An audit log with the right columns can still fail the question it was built for. Two tests: whether you can answer without reconstructing, and whether anything notices when the record stops being written.
Architecture - 2026-09-21 - 9 min read
The eight questions a manager asks before any of this gets budget
Part 6 and the last in the series: the eight questions a manager asks before enterprise AI adoption gets budget, four of which the reference diagram answers with a specific component and four of which stay judgments.
Architecture - 2026-09-21 - 9 min read
Stage 5 of enterprise AI adoption: governance hardening
The fifth stage of enterprise AI adoption compares declared policy against what the records actually show, records which policy version allowed each action in an append-only ledger, and stops a run that is already out of bounds rather than refusing its next call.
Architecture - 2026-09-20 - 7 min read
Stage 4 of enterprise AI adoption: write and send behind approval
The fourth stage of enterprise AI adoption splits write connectors from send connectors, puts the refusal where a tool enforces it rather than in the prompt, and leaves an approval record the system writes instead of one the agent reports.
Architecture - 2026-09-16 - 5 min read
Stage 3 of enterprise AI adoption: agents that can read
The third stage of enterprise AI adoption runs an agent inside a sandbox with read-only connectors scoped by the same roles as stage 2, and treats completion as a fact the system checks outside the agent, never a sentence accepted from it.
Architecture - 2026-09-11 - 6 min read
Stage 2 of enterprise AI adoption: data and retrieval
The second stage of enterprise AI adoption indexes internal documents and scopes retrieval by role, so the same question from two roles returns different data and the model has no part in deciding what it may share.
Architecture - 2026-09-10 - 7 min read
Stage 1 of enterprise AI adoption: identity and the AI gateway
The first stage of enterprise AI adoption puts identity, roles, secrets, and an AI gateway in place so every later stage can answer who is asking and whether a call was logged and within budget.
Architecture - 2026-09-09 - 6 min read
One diagram for enterprise AI adoption: five stages from identity to governance
A vendor-neutral reference architecture for enterprise AI adoption, read in adoption order: identity and gateway, data, agents, write and send, governance.
Architecture - 2026-09-01 - 7 min read
Change-Intent Governance: The Missing Layer
Your IDP knows what exists and your agent gateway knows what a tool call did. Neither can say which change touched which projects on whose authority.
Leadership - 2026-08-23 - 4 min read
I Said Closed Source Isn't a Moat. I Still Keep the Runtime Closed.
One question sorts everything I build into open or closed: build-time artifact, or runtime service? And the condition under which I'd move the line.
Leadership - 2026-08-17 - 3 min read
Closed Source Isn't a Moat. Ask What Happens When the Model Gets Better.
Secrecy buys a head start that erodes on someone else's model timeline. The better question: does this asset gain or lose value as models improve?
Architecture - 2026-08-15 - 3 min read
The Agent Says It Only Touched One File. Prove It.
An agent's report of what it touched comes from the process you're trying to verify. So enforce scope at the OS level and attack the sandbox yourself.
Research - 2026-08-14 - 3 min read
Turning \"What Even Is a Legitimate Organization\" Into Something You Can Validate
Most org design is a canvas nobody can validate. A schema with constraint rules forces you to name who is liable when an agent gets it wrong.
Leadership - 2026-08-13 - 3 min read
Why Code Is About to Get Cheap — and What Actually Becomes the Moat
If your edge is making code generation cheaper, the next model release eats it. Here is the filter I use to decide what to build instead.
Research - 2026-07-11 - 7 min read
Accountable Generation: Intent as the Artifact of Record
If agents write most of your code, line-by-line review can't keep up. Version, diff, and sign off on the intent instead, and treat code as output.
Architecture - 2026-06-12 - 2 min read
Designing an Agent Runtime for Real Engineering Work
A practical framing for moving from tool-level AI assistance to governable AI-native engineering systems.
Research - 2026-05-28 - 1 min read
Context Engineering Is Architecture
Why context design should be treated as a first-class architectural concern in AI-assisted software delivery.
Leadership - 2026-04-18 - 1 min read
Governance-Driven Engineering for AI-Native Teams
A model for combining autonomy, review, policy, and evaluation in AI-assisted engineering organizations.