Writing

Notes from the practice of parallel agentic coding. Things that only really become visible once you're running 8-9+ Claude Code sessions in parallel — the rest of the time the pain stays hidden.

Posts

• Code review when four engineers merge 844 PRs a month — 2026-07-27 — In June 2026 the team I lead opened 892 PRs and merged 844. The per-PR triage we ran to check whether that number means anything, the rework it found, and what we changed about reviewing.
• Why MCPs Don't Scale with Heavy Agent Herding — 2026-05-16 — How I'm getting fewer interrupted Claude Code sessions. Custom MCP servers disconnect mid-flow, spawn per-tab at parallel scale, and force restarts to clear tool/settings caches. The replacement pattern: skill + CLI.

Elsewhere

Longer pieces on Positively Fred, where the practice gets written up at team and organisation scale:

• This is getting ridiculous: I shipped 490 pull requests in June — 2026-08-06 — a follow-up to the Playbook with real throughput numbers for every stage, the per-PR triage that checked them, and what stops a flying team once the technical bottlenecks are gone.
• The AI-Native Engineering Playbook: Crawl, Walk, Run, Fly — 2026-03-04 — the four-stage model this course and bootcamp are built on. 1,100+ reads.

Guides

Longer-form, kept up to date rather than dated:

• What is AgentHerder? — the practice, and cctabs, the terminal-tab tool behind it.
• Running Claude Code sessions in parallel — what breaks between three sessions and twelve.
• Claude Code and tmux — panes, terminal tabs, and where Agent View fits.