Where a hundred billion tokens went
In May I built a dashboard to watch a week of token burn. This month I asked it about the whole year. It told me a third of the evidence was gone, and then i...
I’m a Director of Engineering at Samsara, working across Safety and Telematics. I write about edge AI, firmware-to-cloud systems, engineering leadership, and the practical systems I build around coding agents.
Read the AI engineering series · About my leadership scope
This site is a record of the work behind the outcomes: the decisions, experiments, tools, failures, and operating practices that make the next result more trustworthy.
Recent writing
What the tokens bought
At work the pull-request chart is a hockey stick. At home I have the same chart in lines of code, and I believe it about as much. Here is what I measured ins...
A ledger that outlives the logs
A usage dashboard that only reads what is on disk forgets everything the tools forget. Version 0.5 keeps a ledger instead, and the first run of it found that...
How I use AI (November 2026)
I use AI from two seats. The weekend seat is where I find out what agents can do. The weekday seat is where I find out what it costs to trust them.
2,161 commits later, here is what I measure
The Git history is busy. The harder question is how much of that activity becomes a result worth keeping.
The guardrails that let me run multiple AI agents
One agent per level worked well. Eight agents assigned broad themes across the same codebase did not. The difference was ownership.
AI gives me speed. My job is to give it direction.
I can keep several agents busy. The harder job is making sure their work adds up to a product I actually want.
Keeping teams steady while the organization changes
An organization chart can change quickly. Rebuilding clarity about priorities, ownership, and growth takes more deliberate work.