Thought leadership
Research, rubrics, and field notes on governing autonomous systems.
Capital allocation and operational governance are converging. The posts below trace that convergence — from strategy and framework design to recorded benchmarks and implementation patterns.
[DRAFT] What RLCD and Jev suggest about verifiable AI systems
A reproducible rule-execution benchmark comparing Jev 1.13 and GPT-5 Nano—and what it reveals about reasoning depth, representation, model routing, and explicit control.
[DRAFT] What Clarity Looks Like in Outcomes-Oriented TokenOps
A Python debugging walkthrough from observed AI spend to durable outcomes, comparative experiments, versioned execution packages, human approval, and verified production effects.
[DRAFT] Measuring the Current Value of Merged Pull Requests
A study of 256 merged PRs in sindresorhus/ky using code survival, structural survival, and dependency evidence to estimate what remains in the repository today.
[DRAFT] You Don’t Need Uber’s Scale to Build a Better AI Factory
A practical, small-team scorecard inspired by Uber: attribute managed-agent spend, measure cost per outcome, benchmark execution, and turn successful traces into reusable operating knowledge.
[DRAFT] When the frontier learns, what remains proprietary?
Frontier AI raises the productivity floor while making scarce know-how common faster. A Romer-inspired framework for deciding what firms should rent and what they should own.
[DRAFT] Five questions every board should ask about autonomous AI
A 10-minute five-rule check for money, authority, and evidence. An interactive board governance tool co-authored with Deb Henretta.
From Campaign Blasts to Autonomous Customer Journeys
Why point-in-time CRM scoring erodes customer lifetime value, and how enterprise brands are shifting to continuous relationship planning with sequence-aware restraint.
Proverify.AI — VERIFY Runtime Rubric
A six-dimensional framework for specifying, comparing, manufacturing, validating, and operating domain-specific autonomous systems — plus a recorded retail benchmark of a standard LLM agent versus a Razor-guarded Proverify RunTime.
Control the workflow, not just the tokens
McKinsey is right that enterprise AI economics require visibility, attribution, governance and cost-per-outcome control. A skeptical read of its AI cost-control framework — and where TokenOps, RunTime and VERIFY directly address the control problem.
[DRAFT] 45 billion tokens a day changes the TokenOps conversation
AT&T reports 45 billion tokens per day, cache-aware model routing, and 90% AI cost reductions. Why that makes token economics an operating discipline, not a procurement exercise.
Proverify.ai — Autonomous RunTimes (ART)
AI is a new kind of electricity or oil, but it isn't fungible — it has to be contextual. Why the path to reliable autonomy runs through specialist AIs (ART), financed, built, and deployed via TokenOps and RunTime.
Two sides of the same control problem
Why capital allocation and operational governance for AI are the same board problem. You can't govern what you can't attribute, and you can't attribute what you can't execute deterministically.
What is a Statechart?
An interactive introduction to the diagram that keeps complex reactive systems readable: hierarchy, orthogonality, broadcast communication, and history.
Customer Journey Statechart
See the customer relationship as a living process, not a flat funnel. An interactive statechart that shows how event logs become a durable relationship runtime.
Dog Fooding at Proverify
We have been using two simple internal views—one over ChatGPT history, one over GitHub activity—to learn what TokenOps can actually measure before we ask customers for deeper integrations.
AI at build time, certainty at run time
Split knowledge acquisition from the live control loop. Compile messy operational knowledge into a deterministic runtime with explicit predicates and an audit log.
Owners, not renters
Five bets on the AI transition. The core bet: rent intelligence, but own sector-specific data and software assets.
Barton — A New Approach to the Functional Design of a Digital Computer (1961)
Barton's original 1961 paper on functional computer design. Why separating intent from mechanism is still the right way to think about autonomous systems.
NHTSA Levels of Automation
The NHTSA Levels of Automation framework. Why the oldest question in automation — who decides? — is still the right one to ask before handing work to an autonomous system.