Writing
Technical essays on why agents break and what it takes to fix them, published on DEV. Separately, I write Off Script, about ambition, work, and the strange years between becoming someone and figuring out who you actually want to be.
Technical
Agent memory, cost governance, reproducibility, and where structure helps.
-
Did Your AI Agent Quietly Blow All Your Money?
Why a budget enforced across a whole run catches the overspend that per-call limits miss, and what TokenOps does about it.
-
Debugging Multi-Agent Systems: Your Trace Tree Is Lying
Naive attribution builds a confidently wrong parent-child tree: nothing crashes, the trace just points at the wrong code for an hour. Opening a span before the wrapped function runs, rather than after, fixes it.
-
40% of AI Agent Projects Will Be Canceled by 2027. Token Governance Is How You Avoid It.
Gartner predicts over 40% of agentic AI projects will be cancelled by 2027, with cost among the reasons. Four steps out: instrument, attribute, surface, enforce.
-
You Recorded the Incident. Now Prove Your Fix Actually Works.
Cut-point replay: freeze the run rather than the model, drop your new code into the middle of a recorded incident, and play everything upstream back byte for byte.
-
Why Agents Forget
Models are stateless functions with frozen weights, so agents get faster at repeating work rather than better at it. Persistence is largely solved; learning into the weights is not.
-
Your Agent Failed in Prod. Good Luck Reproducing It.
Temperature zero buys sampling determinism, not system determinism. Batch composition, GPU concurrency and model drift all move the output, so record the run and replay it instead of chasing bitwise reproducibility.
-
Spec-Driven Development: When Structure Helps and When It Becomes Tax
A U-shaped cost curve. Spec it when ambiguity is expensive; past the optimum every extra rule just consumes the agent's finite reasoning budget.
Off Script
Personal writing on ambition, work, relationships, and the strange years between becoming someone and figuring out who you actually want to be.
- The Privilege Paradox
- Evolve. Evolve. Evolve.
- The Price of an Open Heart
- Did You Get This Life Just to Coast?
- Speaking at AI Engineer World's Fair, San Francisco
- The Billionaire Myth and Your 100% Survival Rate
Elsewhere
-
TokenOps: real-time, run-scoped cost control for AI agents
TokenOps featured in Microsoft's developer newsletter.
-
Peer-reviewed research
Applied machine learning and network measurement. Full list on the research page.