Edition of Thursday 12 November 2026

Running parallel agents for coding is not as fast as you think

Comments on “Running parallel agents for coding is not as fast as you think”

Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked

Stackness's author timed 364 Claude Code task runs in one repo and found parallel agents averaged 1.21 working at once, peaking at 1.46 in the busiest window. A separate six-agent run of 106 tasks measured 1.82. Higher published speedups, up to 3.9, come from benchmarks that exclude tests or from search branches that never merge. Idle time waiting for review, a shared test machine and a late 92-file merge cut throughput. The data covers one developer, one repo and one harness, and the author's own figures are unsourced.

Read the original at stackness.dev. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Why it ranked

No. 7 of 106 with a score of 0.69. The panel rated it highest on new information (0.86) and technical depth (0.84), lowest on significance (0.56). Placed No. 7, No. 12 by score alone: the placement terms moved other stories past it.

AI judgeJudge 2claude-haiku-5-5
Detailed measurement of parallel coding agent throughput in one repo, with timing data and a comparison table of published figures. Single-repo sample limits generality; claims are mostly backed by the author's own data, with some overgeneralization in the thesis.
AI judgeJudge 3gpt-5.6-luna
Detailed measurements, methodology, comparisons, and operational lessons make this highly actionable for teams using coding agents, though much of the evidence comes from one repo and self-reported external runs.

Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.

Score

0.69total under rubric v2.3

Judges’ mean answer per question, on a scale from 0 to 1. Questions marked minus count against the score.
  • Significance0.56
  • Reader interest0.78
  • Audience breadth0.67
  • Practical relevance0.82
  • Substantiated0.83
  • Technical depth0.84
  • New information0.86
  • Counts against: Promotional0.06
  • Counts against: Overclaim0.24

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

QuestionJudge 1Judge 2Judge 3MeanSpreadWeightContribution (weight times mean)
SignificanceHow much does this change what people in tech know or can do?0.580.400.700.560.30+0.24+0.135
Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on?0.740.700.900.780.20+0.26+0.203
Audience breadthHow many people who build technology does it concern?0.650.600.750.670.15+0.13+0.087
Practical relevanceHow useful is it to someone who builds software or tech products?0.890.700.880.820.18+0.22+0.181
SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations.0.970.700.820.830.27+0.05+0.042
Technical depthHow technically substantive is it?0.990.700.840.840.29+0.05+0.042
New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)?0.840.800.950.860.15+0.05+0.043
PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional.0.050.100.030.060.07−0.19−0.011
OverclaimDo the headline or conclusions claim more than the body supports?0.200.400.120.240.28−0.11−0.026
Sum of contributions+0.694
Claim-check penalty0.000
Total, as published0.694

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.

Ranked #7 on Rubric, 12 Nov 2026
<a href="https://rubric.hrih.si/editions/2026-11-12/7/"><img src="https://rubric.hrih.si/badge/2026-11-12/7.svg" alt="Ranked #7 on Rubric, 12 Nov 2026" width="238" height="28"></a>

Claim check

Claim check: 7 matches source · 1 no source given. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.

  1. Matches sourceNumber

    The mythical-agent-month run of 106 tasks with six agents measured an effective parallelism of 1.82, from 64.2 agent-hours of work in 35.3 busy hours.

    The cited source says the same thing.

    The article says

    “The 1.82 comes from 64.2 agent-hours of work in 35.3 busy hours, read from 398 subagent transcripts”

    The source says

    “6 agents gave about 1.8 agents of throughput once test slots, file conflicts and merge locks queued, and it fell to 1.4 late in the run”

    Source cited by the article: github.com/MedAdemBHA/mythical-agent-month

  2. Matches sourceQuote

    The mythical-agent-month repo gives the rule of thumb "effective parallelism ≈ 0.3 × your agent cap".

    The cited source says the same thing.

    The article says

    “effective parallelism ≈ 0.3 × your agent cap”

    The source says

    “Effective parallelism ≈ 0.3 × your agent cap.** 6 agents gave about 1.8 agents of throughput”

    Source cited by the article: github.com/MedAdemBHA/mythical-agent-month

  3. Matches sourceNumber

    Cursor's team described the effective throughput of its 20-agent browser experiment as "two or three" agents.

    The cited source says the same thing.

    The article says

    “Cursor's own team later wrote "1-3" for the same 20-agent experiment”

    The source says

    “20 agents would slow to the throughput of 1-3 with most time spent waiting on locks”

    Source cited by the article: cursor.com/blog/self-driving-codebases

  4. Matches sourceNumber

    Claude Code's default BASH_MAX_TIMEOUT_MS is 600,000 milliseconds, i.e. a 10-minute ceiling.

    The cited source says the same thing.

    The article says

    “That 10-minute ceiling matches Claude Code's default BASH_MAX_TIMEOUT_MS of 600,000 milliseconds”

    The source says

    “BASH_MAX_TIMEOUT_MSMaximum timeout the model can set for a foreground Bash or PowerShell tool command, in milliseconds (default: 600000, or 10 minutes”

    Source cited by the article: code.claude.com/docs/en/env-vars

  5. Matches sourceBenchmark result

    In the Vesper paper, four agents in git worktrees reached a parallelism ratio of 3.2 to 3.9, cutting a run from about 70 hours to about 20.

    The cited source says the same thing.

    The article says

    “four agents in git worktrees reached a "parallelism ratio (total sequential agent time divided by wall-clock time)" of 3.2 to 3.9, cutting a run from about 70 hours to about…”

    The source says

    “The parallelism ratio (total sequential agent time divided by wall-clock time) ranges from 3.2x to 3.9x, reducing wall-clock time from approximately 70 hours to 20 hours in the most compute-intensive…”

    Source cited by the article: arxiv.org/abs/2605.15221

  6. Matches sourceQuote

    CooperBench found agents achieve on average 30% lower success rates when working together.

    The cited source says the same thing.

    The article says

    “agents achieve on average 30% lower success rates when working together”

    The source says

    “agents achieve on average 30% lower success rates when working together compared to performing both tasks individually, across the full spectrum of task difficulties”

    Source cited by the article: arxiv.org/abs/2601.13295

  7. Matches sourceQuote

    Claude Code's agent teams docs advise starting with 3-5 teammates for most workflows.

    The cited source says the same thing.

    The article says

    “Start with 3-5 teammates for most workflows”

    The source says

    “Start with 3-5 teammates for most workflows. This balances parallel work with manageable coordination”

    Source cited by the article: code.claude.com/docs/en/agent-teams

  8. No source givenNumber

    In the author's repo, the busiest four-day window with two lanes had 60% of its time with no agent running, and the merge 24 days later left 92 files git could not resolve.

    The article gives no link or citation for it.

    The article says

    “Git could not merge 92 files on its own, and resolving them took a Claude Code session 56 minutes and four full check runs”

    The source says

    No source to compare with.

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN 2 pts, not on front page

Discussion on Hacker News

Recorded for context only, checked 12 Nov 2026, 05:47 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition. How ranking works describes rubric v2.3.

Comments

Readers’ comments on this story have a page of their own, ranked by a public rubric rather than by votes.

Comments on “Running parallel agents for coding is not as fast as you think”

More on Rubric

From the last 7 editions, chosen by shared topic, score and date.

  1. Score 0.75

    llama.cpp v1.0