Coding

$28,335 Over 7 Months, 9,350 Commits: What One Developer's Coding Agent Logs Showed About Running Agents in Parallel

  • Level Advanced
  • Setup1–3 hrs (estimated)
  • Evidence Some evidence
  • CostNot stated in source
  • No-codeNo
Claude CodeCodexSource-linked

What they did

A developer analyzed seven months of his own Claude Code and Codex logs: 1,432 sessions and about 200,000 tool calls. Spend jumped 7.4x in April when he started running agents in parallel git worktrees instead of one conversation at a time. He stepped in about every 15 tool calls and interrupted runs 2,047 times, and only 0.7% of activity happened overnight. He had about 130 skills installed but used them 52 times, mostly one release chore.

How it went

dev.to post by the developer (Oct 26, 2026, cross-posted from his blog): $28,335 over 7 months across five harnesses, 1,432 sessions, 200,478 tool calls, 15 tool calls per human turn, 2,047 interrupts, 9,350 commits across 145 repos (he warns the commit count overstates output), 0.7% of tool calls between midnight and 6 AM, 130 skills installed vs 52 invocations

Worth knowing

  • dev.to post by the developer (Oct 26, 2026, cross-posted from his blog): $28,335 over 7 months across five harnesses, 1,432 sessions, 200,478 tool calls, 15 tool calls per human turn, 2,047 interrupts, 9,350 commits across 145 repos (he warns the commit count overstates output), 0.7% of tool calls between midnight and 6 AM, 130 skills installed vs 52 invocations.

Prompt

Cleaned version rewritten from the source — not a direct quote.

Look at my coding agent session logs in [folder] for the last [N] months. Count sessions, my turns, tool calls and interrupts per tool, total spend by month, and activity by hour of day and weekday. Tell me how many tool calls the agent makes per message from me, which months my spend jumped and what changed in how I worked then, and which installed skills I actually use versus never touch. Put the results in a short report with a few charts.

Recommended

Try it in Grok Bot

  1. Copy the prompt (button below).
  2. Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
  3. Connect the listed tools (as needed) as connectors when Grok Bot asks.
  4. If the task repeats, ask: "Make this a daily routine."

Alternatives: the same prompt works in Claude Code, Codex, Claude, ChatGPT Agent, OpenClaw.

Set it up

Stack: Claude Code / Codex

  1. Read the original source for full context.
  2. Connect the tools your agent needs.
  3. Paste the prompt and replace [bracketed] placeholders.
  4. Run once manually and check the output before scheduling it.

Generic suggested steps, not from the source.

Evidence

dev.to post by the developer (Oct 26, 2026, cross-posted from his blog): $28,335 over 7 months across five harnesses, 1,432 sessions, 200,478 tool calls, 15 tool calls per human turn, 2,047 interrupts, 9,350 commits across 145 repos (he warns the commit count overstates output), 0.7% of tool calls between midnight and 6 AM, 130 skills installed vs 52 invocations

Source: Web · 2026-10-26 · Verified by us: not yet · Report broken or outdated

Related use cases

ContentX · 2026-01

Turn YouTube channels into a weekly EPUB magazine

Every week, pulls new videos from a list of YouTube channels, rewrites their transcripts as magazine articles, and emails them to you as a newsletter plus an EPUB file.

  • Level Intermediate
  • Setup1–3 hrs*
  • Evidence Strong evidence
Claude CodeSource-linked
CodingX · 2026-10

Hermes Agent Nearly Doubled Speech-to-Text Accuracy on Hard Words: 99 to 183 of 240

A user turned his 15K saved Wispr Flow dictations into a test set for the local Whisper model behind his Hermes Agent. The agent turned his Wispr dictionary into a hint list, diffed raw vs corrected text to find words Whisper keeps mishearing, made find-and-fix rules, and tested every rule against the full history. With no fine-tuning, dictionary words went from 99/240 to 183/240 on Wispr clips and 19/45 to 43/45 on sentences he read aloud. It now archives the data nightly and looks for new fixes weekly.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Hermes AgentSource-linked
CodingReddit · 2026-10

1 Minute to 1 Second: Claude Code Built a Faster Plex App for a Samsung TV

On a 2022 Samsung TV the official Plex app took about a minute to open and search lagged about 10 seconds. A Redditor used Claude Code to build a free, open-source web-app client that opens in about a second, searches as you type and plays 98.7% of his library directly in hardware.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Claude CodeSource-linked