Agarwal, Shyam, Lvova, Olga, Sergeyuk, Agnia, Serova, Anastasiia, Bryksin, Timofey, Zimmermann, Thomas, Vasilescu, Bogdan · Nov 10, 2026
What Engineers' Token Spend Measures in the AI Era
Markdown · JSON · Original source
The paper’s authors ask whether engineers’ token spend—the amount of tokens used in coding agents—measures engineering output. They link five weeks of agent usage to commits by 404 JetBrains engineers and analyze a survey of 412 employees. They report that spend differed mostly between engineers, with largely the same engineers spending the most across the five weeks. In weeks with similar commit counts, the 90th-percentile spend was 39 to 240 times the 10th-percentile spend; within an engineer, doubling spend was associated with about 10% more commits. Available work metrics explained 38% of variation between engineers and 13% of week-to-week variation within an engineer. The authors also report that technical-role respondents used agents for work beyond coding in nearly nine of ten evaluations, including analysis, research, and planning. Their argument is that spend tracks agent use across different kinds of work, not the outcome or value of that work. They recommend using it at most for resource accounting and evaluating agent-assisted work with evidence about tasks and outcomes. The abstract does not establish how well such evaluation methods predict the value of work.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Nov 10, 2026 (2026-11-10).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T03:49:01.118Z. First observed: 2026-11-11T05:04:51.331Z. Retrieved: 2026-11-12T03:48:04.772Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The evidence described here is the abstract; it does not provide further detail on how token spend, commits, or the survey responses were measured. The study covers five weeks of usage among engineers at one company, JetBrains; the abstract does not establish whether the findings generalize to other organizations or periods. The abstract reports an association between within-engineer spend and commits, not that increased spend caused more commits.
Open question
What evidence about tasks and outcomes would let organizations evaluate the value of agent-assisted work, as the authors recommend?
Source-stated
The study links five weeks of coding-agent token usage to commits by 404 JetBrains engineers and analyzes a survey of 412 employees.
We link five weeks of token usage in coding agents to the commits of 404 engineers at JetBrains, and analyze a company survey of 412 employees.
Source-stated
For weeks with similar commit counts, token spend at the 90th percentile was 39 to 240 times spend at the 10th percentile.
Among weeks with similar commit counts, the 90th percentile of spend was 39 to 240 times the 10th.
Source-stated
Within an engineer, doubling token spend was associated with about 10% more commits.
Within an engineer, doubling spend was associated with about 10% more commits.
Source-stated
Available work metrics explained 38% of between-engineer variation in spend and 13% of week-to-week variation within an engineer.
Available work metrics explained 38% of the variation in spend between engineers and 13% of the week-to-week variation within an engineer.
Source-stated
The authors report that technical-role survey respondents used agents for tasks beyond coding in nearly nine of ten evaluations.
In the survey, respondents in technical roles reported in nearly nine of ten agent evaluations using agents for tasks beyond coding, such as analysis, research, and planning.
Source-stated
The authors argue that token spend measures agent use across kinds of work, rather than the outcome and value of that work.
Token spend therefore measures how much engineers use agents across many kinds of work and not the outcome and value of that work.
Source-stated
The authors recommend using token spend at most for resource accounting and assessing agent-assisted work with evidence about tasks and outcomes.
We recommend using token spend at most for resource accounting, and evaluating agent-assisted work with evidence about tasks and outcomes.
Source and contributor context
Agarwal, Shyam, Lvova, Olga, Sergeyuk, Agnia, Serova, Anastasiia, Bryksin, Timofey, Zimmermann, Thomas, Vasilescu, Bogdan · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 3c6d402b-7128-4ae2-8eab-f8d04806135d.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T03:49:01.118Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 2ac9cd65e001949a38721cbdceec0630a6887c048d5be9e841519b570143cd68. Source text hash: 7097a1f4d3747180d85ad1be148333750554d768a59e95a02acc567130358a3f.
Revision and source-check history
Version 1 · 2026-11-12T03:49:01.118Z
What Engineers' Token Spend Measures in the AI Era
The paper’s authors ask whether engineers’ token spend—the amount of tokens used in coding agents—measures engineering output. They link five weeks of agent usage to commits by 404 JetBrains engineers and analyze a survey of 412 employees. They report that spend differed mostly between engineers, with largely the same engineers spending the most across the five weeks. In weeks with similar commit counts, the 90th-percentile spend was 39 to 240 times the 10th-percentile spend; within an engineer, doubling spend was associated with about 10% more commits. Available work metrics explained 38% of variation between engineers and 13% of week-to-week variation within an engineer. The authors also report that technical-role respondents used agents for work beyond coding in nearly nine of ten evaluations, including analysis, research, and planning. Their argument is that spend tracks agent use across different kinds of work, not the outcome or value of that work. They recommend using it at most for resource accounting and evaluating agent-assisted work with evidence about tasks and outcomes. The abstract does not establish how well such evaluation methods predict the value of work.
Original source
cs.AI · Article
Source details
Original source date: Nov 10, 2026 (2026-11-10).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-12T03:49:01.118Z. First observed: 2026-11-11T05:04:51.331Z. Retrieved: 2026-11-12T03:48:04.772Z.
Limits
The evidence described here is the abstract; it does not provide further detail on how token spend, commits, or the survey responses were measured. The study covers five weeks of usage among engineers at one company, JetBrains; the abstract does not establish whether the findings generalize to other organizations or periods. The abstract reports an association between within-engineer spend and commits, not that increased spend caused more commits.
Open question
What evidence about tasks and outcomes would let organizations evaluate the value of agent-assisted work, as the authors recommend?
Source-stated
The study links five weeks of coding-agent token usage to commits by 404 JetBrains engineers and analyzes a survey of 412 employees.
We link five weeks of token usage in coding agents to the commits of 404 engineers at JetBrains, and analyze a company survey of 412 employees.
Source-stated
For weeks with similar commit counts, token spend at the 90th percentile was 39 to 240 times spend at the 10th percentile.
Among weeks with similar commit counts, the 90th percentile of spend was 39 to 240 times the 10th.
Source-stated
Within an engineer, doubling token spend was associated with about 10% more commits.
Within an engineer, doubling spend was associated with about 10% more commits.
Source-stated
Available work metrics explained 38% of between-engineer variation in spend and 13% of week-to-week variation within an engineer.
Available work metrics explained 38% of the variation in spend between engineers and 13% of the week-to-week variation within an engineer.
Source-stated
The authors report that technical-role survey respondents used agents for tasks beyond coding in nearly nine of ten evaluations.
In the survey, respondents in technical roles reported in nearly nine of ten agent evaluations using agents for tasks beyond coding, such as analysis, research, and planning.
Source-stated
The authors argue that token spend measures agent use across kinds of work, rather than the outcome and value of that work.
Token spend therefore measures how much engineers use agents across many kinds of work and not the outcome and value of that work.
Source-stated
The authors recommend using token spend at most for resource accounting and assessing agent-assisted work with evidence about tasks and outcomes.
We recommend using token spend at most for resource accounting, and evaluating agent-assisted work with evidence about tasks and outcomes.
Source and contributor context
Agarwal, Shyam, Lvova, Olga, Sergeyuk, Agnia, Serova, Anastasiia, Bryksin, Timofey, Zimmermann, Thomas, Vasilescu, Bogdan · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 3c6d402b-7128-4ae2-8eab-f8d04806135d.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-12T03:49:01.118Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 2ac9cd65e001949a38721cbdceec0630a6887c048d5be9e841519b570143cd68. Source text hash: 7097a1f4d3747180d85ad1be148333750554d768a59e95a02acc567130358a3f.
uncertain · 2026-11-12T03:49:01.118Z · anthropic/claude-opus-5.5
Publication: 3c6d402b-7128-4ae2-8eab-f8d04806135d
attribution — pass: The findings and recommendation are consistently attributed to the paper's authors. The draft says the study covers 404 JetBrains engineers and 412 surveyed employees, which matches the source.
We link five weeks of token usage in coding agents to the commits of 404 engineers at JetBrains, and analyze a company survey of 412 employees.
summary — uncertain: The draft omits the source's stated stakes: organizations already use token spend to monitor AI adoption, set budgets and measure productivity. Without this, a general reader cannot see why the finding matters. Smallest repair: add that sentence. The central terms 'token' and 'commit' are also unexplained. Saying 'the amount of tokens used' does not tell a non-specialist what a token is. Add term-library tooltips for 'token' and 'commit'. The paper's framing as the AI era's 'lines of code' is lost from the title. The figures and conclusions are otherwise faithful.
Organizations use engineers' token spend to monitor AI adoption, set budgets, and measure productivity, but the metric's relationship with engineering output is unclear.
date — pass: The originalPublished date of 2026-11-10 matches the v1 submission date on arXiv. The 2026-11-11 feed date is correctly treated as a later announcement.
[Submitted on 10 Nov 2026]
support — pass: Each claim is quoted closely from the abstract. The limitations correctly note abstract-only evidence, a single company, and association rather than causation. The question follows from the authors' recommendation without adding unsupported assumptions.
Within an engineer, doubling spend was associated with about 10% more commits.
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct access to the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv