Blog

• Five ways to run coding agents in parallel, compared

  Git worktrees by hand, a session manager, editor agents, a hosted agent, or a desktop app. All five launch agents. The difference shows on a team.

  1 August 2026 · 10 min read · comparison · parallel agents · claude code

• Git worktrees are the easy part of running coding agents in parallel

  Eight of twelve tools we checked isolate agents with a git worktree, so isolation is the floor. What grows is cleanup, untracked files, and reading.

  1 August 2026 · 5 min read · git worktree · parallel agents · teams

• How many coding agents can one person supervise?

  The ceiling is review throughput, not hardware. About 600 lines read properly a day, against 250 in a changeset, puts most people at three to six agents.

  1 August 2026 · 4 min read · parallel agents · code review · teams

• Nobody can tell you what their coding agents cost per person

  None of the nine locally installed tools we checked documents a spend cap. None of the twelve reports spend per person. Parallelism multiplies the bill.

  1 August 2026 · 5 min read · agent spend · teams · parallel agents

• Reading somebody else's agent

  Two of twelve tools we checked let a teammate read a run before the pull request, both on a paid plan. On the other ten, it is finished before you see it.

  1 August 2026 · 4 min read · code review · teams · parallel agents

• Twelve parallel coding agent tools, side by side, with sources

  Twelve tools side by side: platforms, agents driven, worktree isolation, review surface, spend caps, and price. One source per cell, checked 1 August 2026.

  1 August 2026 · 6 min read · comparison · parallel agents · claude code

• We checked every way to run coding agents in parallel. Two problems are left.

  Twelve tools, one source per claim, checked 1 August 2026. Eight isolate agents the same way. Two problems remain: early team review and cost per person.

  1 August 2026 · 8 min read · parallel agents · teams · agent spend

• Writing got faster. Reading did not.

  Agents multiply what a team writes and leave review throughput unchanged. Three agents produce 1,200 lines a day against the 600 one person can read.

  1 August 2026 · 5 min read · code review · teams · parallel agents

• Claude Code, Codex, or Cursor: match the agent to the task

  "Which agent is best" has no stable answer. "Which agent for this task" does. How Claude Code, Codex, and Cursor differ in shape.

  31 July 2026 · 3 min read · claude code · codex · cursor

• Review your team's AI code before it becomes a pull request

  An agent can write a week of code overnight. Waiting for the pull request turns review into a signature. Read the changeset while it is still a worktree.

  31 July 2026 · 3 min read · code review · teams · parallel agents

Subscribe to the RSS feed