01Signal
Architecture pattern
Reusable deterministic-first decisions from shipping real systems.
Back to home
Insights
Writing on deterministic AI.
Long-form on AI-native product architecture, the economics of inference, and lessons from building MDS in public. Posts arrive when they are worth your time.
Latest
Published writing.
012026-06-04
Workflow
What a Workflow Diagnostic actually finds
Most businesses do not have an AI problem - they have a workflow problem. A diagnostic maps the work step by step and labels each one: rule, AI, or human.
2026-06-04 · 5 min read
022026-06-04
VCOS
AI employees are theatre. Build the operating layer instead.
The "AI employee" pitch sells autonomy you cannot trust with anything that matters. The durable version is quieter: a deterministic operating layer with human gates.
2026-06-04 · 6 min read
032026-06-04
Deterministic AI
Deterministic over generative: the economics of software that runs at ~$0 per request
Most knowledge work is rules plus a little judgment. Wrapping an LLM around every request is the expensive way to ship it. Here is the cheaper, more honest one.
2026-06-04 · 6 min read
What will land here
Useful patterns, not a content treadmill.
02Signal
Build log
What changed, why it mattered, and what survived production constraints.
03Signal
Margin breakdown
Cost structure, runtime economics, and when AI API calls are not needed.
Publishing protocol.
We would rather ship a product than ship a blog post about shipping a product.
01Protocol
Ship the product
The product shelf is the first publishing surface because working software carries the proof.
02Protocol
Extract the lesson
Only patterns that survived implementation, verification, or deployment become candidates.
03Protocol
Publish when useful
No content treadmill. A post lands here only when it teaches something reusable.
Current channels
The work is the writing. Here is where it shows up.
Want a heads-up when the next post drops?
Drop us a line