Foundation series
How organisations should place AI into real work, argued from where capability has to live. Boundaries, lineage, and governance, set as architecture rather than as prompts.
The AI Mainframe Trap
Enterprises are bolting AI onto their systems the way they once bolted everything onto the mainframe: no boundaries, no lineage, no governance. We learned this lesson once and spent decades refactoring our way out. The AI era is rhyming.
Read essay →FoundationVertical AI Foundation
Enterprise architecture patterns for governed AI systems: Steps and States, Causal Lineage, Control Plane isolation, AI Airlocks, and Contract-First Models. The way out of the mainframe trap isn't better prompts. It's architecture.
Read essay →FoundationRead the Foundation series
Both Foundation essays and the patterns they introduce, collected under Enterprise Architecture.
Open Enterprise Architecture →Frontier Series
The forward research: what you can do with the lineage a governed system records. Evaluating AI against the mechanism is the first published piece. CARL, causal-refinement learning, is where the rest is still being worked out.
AI Eval against Causal Generators
A right answer can describe impossible dynamics. Causal generators test AI claims against validated models of mechanisms, with explicit limits and room to say inconclusive.
Read essay →Causal Refinement Learning (CARL)
Frontier research on learning from causal lineage. No published piece yet; the Frontier series collects this work as it lands.
forthcomingRead the Frontier series
Causal generators, Stewardship and the X-Ray lineage essay, with CARL collected as it publishes.
Open Research →Patterns
A collection of named, evergreen, interlinked patterns, revised in place rather than dated.
Thin Boundary
Share governed capabilities while letting teams compose the behaviour their work requires.
Open pattern →PatternControl Plane vs. Execution Plane
Keep policy and authority in trusted services; give each execution only the context and capabilities its task requires.
Open pattern →PatternCausal Lineage
Preserve the evidence connecting a decision to its inputs, permissions, actions, and observed outcomes.
Open pattern →PatternAI Airlocks
Validate a proposed action before granting it authority to change anything.
Open pattern →