The human side
Hi, I'm Geoff. I build products for people, and lately, for machines.
Senior Product Manager at Google Cloud (AI2) for the GKE control plane and gigascale AI clusters. Google for Startups coach. Builds AI-native tools on the side. Toronto, Canada · Bogotá, Colombia.
Human as a Service
Your AI agent can ask me what it cannot compute, and I answer by hand.
01
What I care about
- Systematic forex trading on USD/JPY (the Ninja): backtesting, Monte Carlo position sizing and fully automated execution, with a hard rule that the edge is proven before money is risked
- How websites get consumed by agents and foundation-model providers instead of browsers
- Machine-payable services (HTTP 402 / x402)
- Executive assessment and decision-making benchmarks for AI
- Semantic memory for regulated B2B agents
- Games and interactive fiction (Godot): Yield 2041, built on the whisper.garden lore
02
Writing
Updated daily · charts and a dataset
Token Economics report
What a million tokens costs, and why the ceiling and the floor are moving in opposite directions
Harness report
Who is running the loop, and what it costs them
October 24, 2026
The shirt knew who was wearing it
Two founders had two visions for the world's first biometric shirt. One metric, read off the box and the app, told us who was actually buying it. It was nobody either of them had pitched.
October 24, 2026
Product management is splitting in two
Pragmatic Marketing drew the PM job as a spectrum from market problems to sales enablement in 1993. For thirty years the industry hired people to stand in the middle of it. AI is pulling that middle apart, and I picked the engineering end.
October 18, 2026
A machine paid a human ten cents. Here is every request it took.
whisper.garden now has a door where an AI agent can pay a person 0.10 USDC to interpret something for it. The payment runs over x402, a status code from 1997 that waited 29 years for a customer. It went live on Base mainnet on 17 October and the first real petition settled the next morning.
03
Recently pushed
Straight from GitHub. Public repos, most recent activity first.
Buyer-first real-estate and mortgage site for an Oakville broker: Next.js on Firebase with a full lead pipeline, email/SMS automation and an AI home valuation.
MCP server for Memanto, a different take on RAG-style recall: typed semantic memory with opinions, built for agents in regulated industries where what gets remembered has to be auditable.
A GKE cluster configured and tuned for AI workloads: accelerators, networking, storage and scheduling set up the way inference and training jobs want them.
A load generator for GKE that ships with its own observability. One script deploys a Python workload simulator that stresses CPU, memory, network or storage in burst, sustained or variable patterns, alongside Grafana dashboards fed by Google Managed Prometheus, so you can watch the control plane (API server, scheduler, controller manager), the nodes and the workload itself react while the pressure is on. Built for capacity and scaling tests where the question is not just "does it hold" but "what does the control plane do while it holds".
Named after Dead Hand, the Soviet-era system built to answer a nuclear strike on its own if the people who would normally give the order were gone. This one is a dead man's switch for a person: it checks in with you on a schedule to verify you are alive, escalates through tiers of contacts and channels when you do not answer, and, once it is sure, carries out your last wishes. Work in progress: the tiered communication and verification are built; executing the wishes is not yet.
A Rust-based, AI-native API gateway that generates typed contracts between backend services and their consumers on the fly.