About
Brad Larson
Brad Larson started as an analyst, taught himself to code, and was a software engineer in San Francisco within two years. He was Coinbase's first UK hire, is co-inventor on five granted US patents, and ran validator infrastructure with $5B+ staked at Polychain and Unit 410. He has built with AI since 2024: 112 billion tokens so far, across a deal-sourcing company, three production systems and fifteen hobby projects. FireClaw is his UK consultancy.
Brad’s own projects
Alongside FireClaw, Brad runs his own companies and side projects: Acquirelytics, Brain, Dad Cards, LEGO Mondrian and more. See the portfolio →
Career
Since teaching himself to code at Capital One, Brad has worked across data (mostly machine learning), backend, infrastructure, SRE and security.
- 2021–2023
Unit 410 (Coinbase) · Network Launch Lead
Launched 8 networks, including Moonbeam, Near, Arbitrum and Rocket Pool, which brought in over half the revenue. $5B+ staked, often the largest validator on its networks. Made cold-storage key work up to 95% faster at 4× the throughput.
- 2020–2021
Polychain Labs · 3rd engineer
Launched Celo, Polkadot, Avalanche and Oasis, worth over $20M in revenue, with at least $5B staked. Coinbase acquired the team as Unit 410 in 2021.
- 2016–2020
Coinbase London · first UK employee
The office grew from 1 to 50. Moved 300+ passwords and keys out of the code of 150+ services. Built Codeflow, the deploy system behind 10,000+ releases a year. Migrated an acquired company from Azure to AWS.
- 2014–2016
Coinbase San Francisco · about employee #50
Built the machine-learning risk score behind the compliance patents, and the real-time pipeline into the data warehouse.
- 2011–2014
Capital One Labs
Taught himself to code here, then led a B2B software platform for merchants. Two patent applications on data visualisation.
Patents
Coinbase was using machine learning against fraud. These patents turn the same approach on compliance: train on a few known bad actors, score every account, send the top matches to investigators, and learn from their verdicts. Filed 2016, assigned to Coinbase.
- US10482470B2 Self-learning compliance determination and enforcement platform (2019)
- US10510079B2 Small sample based training applied to a large population (2019)
- US10510034B2 Investigator interface and override functionality (2019)
- US10755347B2 Corrective action realignment and feedback (2020)
- US11625769B2 Multi-factor integrated compliance determination (2023)
Two earlier applications from Capital One, published but not granted: US20160203501A1 (merchant business intelligence tools) and US20190164176A1 (processing transaction data).
112 billion tokens since November 2024
Two sources, each counted once: the coding agents Brad builds with (Claude Code, Codex, Cowork and OpenClaw), and the AI calls the apps make, as each provider billed them.
Coding agentsApps’ own API calls
Most coding-agent tokens are the model re-reading its working context, which is normal for long agent sessions. Coding agents are counted from February 2026, when Brain started recording them; app calls from November 2024. Source: Brain’s /ai/history and the providers’ own usage reports, 18 Oct 2026.
AI-written commits per month
Commits that name Claude, Codex or Cursor as a co-author, across every repository.
Source: Commits naming Claude, Codex or Cursor, all repos, from git, 17 Oct 2026.
Tell us what you need.
A few lines is plenty. We’ll reply with a clear quote, usually within two business days.