I build products with teams of AI agents

2026

1. Oct 8, 2026BuildingWebsite

   Glassworks

   This page. A living tech CV: everything I build with AI agents on one timeline, with my real token usage read from the agents' own logs.

   • Built by agents
   • Usage read from agent logs
   • Self-updating CV (next)

   You're here
2. Oct 2026BuildingSoftwareOpen source soon

   Fleetglass

   A cockpit for servers that run AI agents: machine health, services, Claude Code and Codex sessions, quotas and costs on one self-hosted screen.

   • Agent observability
   • Usage & cost tracking
   • Self-hosted

   Coming soon
3. Oct 2026LiveWeb appOpen source

   The Data Driver · Data Lab

   The analysis workspace behind The Data Driver. Ask questions of real F1 data with your own model key, or plug the data into Claude and ChatGPT.

   • Public MCP server
   • Bring-your-own-key LLM
   • Tool calling
   • AI eval bench

   thedatadriver.app↗
4. Oct 3, 2026LiveWebsite · SaaS

   EpistIQ

   Financial planning and reporting for finance teams.

   • Built by agents
   • MCP server

   epistiq.com↗
5. Oct 2026BuildingWeb app

   The Data Driver · Rig

   A new Rig tab inside The Data Driver: Rig Planner compares sim racing hardware, Rig Doctor serves sourced force-feedback settings.

   • Agent-maintained catalogue
   • Sourced recommendations

   Coming soon
6. 

   Sep 2026InternalInfrastructure

   Agent Platform

   The setup that builds everything on this page: one coordinator agent splits the work, short-lived workers ship it in parallel, and Claude Code and Codex review each other's code.

   • Multi-agent orchestration
   • Cross-model code review
   • LLM-free guardrails

   Orca

   Runs daily
7. Qd

   Sep 2026InternalResearch

   Quant Desk

   Systematic research desk on a paper trading account, limited to a UCITS ETF universe with strict drawdown limits.

   • Agent-assisted research
   • Backtesting

   IB Gateway

   Not public
8. LA

   Sep 2026InternalTool

   LLM Arena

   An evaluation arena for AI models, built to decide between local and cloud inference before buying hardware.

   • LLM evaluation
   • Local vs cloud inference

   gpt-ossQwenGLMKimiGemma

   Internal
9. 

   Sep 2026RetiredInfrastructure

   Hermes

   My first multi-agent system, built on Nous Research's Hermes Agent: eight specialised agents running around the clock, each with its own role and services.

   Why retired

   Always-on agents kept spawning work and burned tokens even when idle, and coordinating eight of them cost more than it saved. I replaced it with one coordinator and short-lived workers that start for a task and stop when it ships.

   • Always-on agents
   • Token budgeting
   • Lessons learned

   Hermes AgentPaperclip

   Retired
10. Mar 2026LiveWebsite

    The Data Driver · F1

    Machine learning predictions for Formula 1, published as probabilities with a public track record. The main site of The Data Driver, home to Data Lab and the upcoming Rig tab.

    • ML race predictions
    • Automated data pipeline
    • Agent-run monitoring

    LightGBMXGBoost

    thedatadriver.app↗