SWE-CoWork: Benchmarking Coding Agents as Digital Employees in Organizations
Comments on “SWE-CoWork: Benchmarking Coding Agents as Digital Employees in Organizations”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
TokenWave AI's SWE-CoWork benchmark places coding agents in simulated engineering organizations with 26 modeled colleagues, across 101 assignments and repositories. Claude-Opus-5.5 and GPT-6 Astra pass about 63% of fail-to-pass tests but complete under 8% of tasks. Supplying full requirements upfront raised test pass rates but left task success unchanged. Claude-Opus-5.5 reaches the right owners yet obtains 64% of colleague-held requirements, and 31% of requirements it reported done failed tests. These figures are the authors' own, without independent verification.
Why it ranked
No. 14 of 144 with a score of 0.69. The panel rated it highest on new information (0.92) and reader interest (0.81), lowest on significance (0.66). Placed No. 14, No. 15 by score alone: the fourth AI & machine learning story of the day (−0.020), and close to “SlopCodeBench: Benchmarking How Coding Agents Degrade over Long-Horizon Tasks”, No. 6 on Sunday 15 November 2026 (−0.012). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Benchmark paper on coding agents in simulated organizations; substantive new evaluation with clear findings on requirement tracking, though results are self-reported and the scope is research-focused rather than landmark.
- AI judgeJudge 3gpt-5.6-luna
- A detailed primary benchmark study of coding agents in simulated organizational workflows, with executable tests, trajectory analysis, and concrete failure modes. It offers substantial new evidence relevant to AI-assisted software engineering, though some broad novelty claims are difficult to independently verify from the excerpt.
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
- Significance0.66
- Reader interest0.81
- Audience breadth0.72
- Practical relevance0.67
- Substantiated0.80
- Technical depth0.80
- New information0.92
- Counts against: Promotional0.09
- Counts against: Overclaim0.25
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.69 | 0.45 | 0.82 | 0.66 | 0.37 | +0.24 | +0.157 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.72 | 0.80 | 0.92 | 0.81 | 0.20 | +0.26 | +0.211 |
| Audience breadthHow many people who build technology does it concern? | 0.79 | 0.60 | 0.78 | 0.72 | 0.19 | +0.13 | +0.094 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.63 | 0.55 | 0.82 | 0.67 | 0.27 | +0.22 | +0.147 |
| 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. | 0.94 | 0.60 | 0.86 | 0.80 | 0.34 | +0.05 | +0.040 |
| Technical depthHow technically substantive is it? | 0.85 | 0.60 | 0.94 | 0.80 | 0.34 | +0.05 | +0.040 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.88 | 0.90 | 0.98 | 0.92 | 0.10 | +0.05 | +0.046 |
| 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. | 0.09 | 0.15 | 0.04 | 0.09 | 0.11 | −0.19 | −0.018 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.21 | 0.35 | 0.18 | 0.25 | 0.17 | −0.11 | −0.027 |
| Sum of contributions | +0.691 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.691 | ||||||
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.
<a href="https://rubric.hrih.si/editions/2026-11-19/14/"><img src="https://rubric.hrih.si/badge/2026-11-19/14.svg" alt="Ranked #14 on Rubric, 19 Nov 2026" width="245" height="28"></a>Claim check
Claim check: 2 matches source · 6 no source given. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
Matches sourceNumber
Anthropic reports that as of May 2026, over 80% of the code merged into its internal codebase was authored by Claude.
The cited source says the same thing.
The article says “Anthropic reports that as of May 2026, over 80% of the code merged into its internal codebase was authored by Claude”
The source says “As of May 2026, more than 80% of the code we merge into Anthropic’s codebase was authored by Claude”
Source cited by the article: anthropic.com/institute/recursive-self-improvement
Matches sourceNumber
In the second quarter of 2026, the typical Anthropic engineer merged 8 times as many lines of code per day as in 2024.
The cited source says the same thing.
The article says “in the second quarter of 2026, the typical engineer merged $8\times$ as many lines of code per day as in 2024”
The source says “In the second quarter of 2026, the typical engineer was merging 8× as much code per day as they were in 2024.4”
Source cited by the article: anthropic.com/institute/recursive-self-improvement
No source givenBenchmark result
Claude-Opus-5.5 and GPT-6 Astra pass 63.84% and 62.91% of F2P test nodes but complete only 6.93% and 7.92% of tasks on SWE-CoWork.
The article gives no link or citation for it.
The article says “Claude-Opus-5.5 and GPT-6 Astra pass 63.84% and 62.91% of F2P test nodes but complete only 6.93% and 7.92% of tasks”
The source says No source to compare with.
No source givenBenchmark result
Supplying complete requirements upfront raises pooled F2P from 39.09% to 55.76% for Seed Evolving and from 62.91% to 78.84% for GPT-6 Astra, while complete task pass rates stay unchanged.
The article gives no link or citation for it.
The article says “raises pooled F2P from 39.09% to 55.76% for Seed Evolving and from 62.91% to 78.84% for GPT-6 Astra (Table 3), yet complete task pass rates remain unchanged”
The source says No source to compare with.
No source givenNumber
Claude-Opus-5.5 directs a mean of 93% of first questions to the accountable owner but obtains only 64% of the 2,176 colleague-held requirements.
The article gives no link or citation for it.
The article says “directs its first question to the accountable owner for a mean of 93% of queried cards across runs (Table 7), yet obtains only 64% of the 2,176 colleague-held requirements”
The source says No source to compare with.
No source givenNumber
Among 826 requirements Claude-Opus-5.5 reported as completed at handoff, 255 (31%) fail their corresponding tests.
The article gives no link or citation for it.
The article says “Among 826 requirements reported as completed at handoff, 255 (31%) fail their corresponding tests”
The source says No source to compare with.
No source givenNumber
Of 450 deferred owner-card threads, 153 received no follow-up from Claude-Opus-5.5, with a median of 7.1 hours of execution budget remaining at deferral.
The article gives no link or citation for it.
The article says “The remaining 153 threads receive no follow-up (Figure 8b). At the initial deferral, these abandoned threads have a median of 7.1 hours of execution budget remaining”
The source says No source to compare with.
No source givenGeneralization
Requirement acquisition is strongly associated with F2P across eleven models (Spearman rho=0.94), whereas tool-call counts show little association (rho=0.08).
The article gives no link or citation for it.
The article says “Requirement acquisition is strongly associated with F2P (Spearman $\rho=0.94$ ), whereas tool-call counts show little association ( $\rho=0.08$”
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
Recorded for context only, checked 19 Nov 2026, 05:47 UTC. It played no part in the ranking.