Small teams. Every paper has an author you can write to.
The research programme is deliberately small. Every published result names the team that did the work, and that team will answer a question about the method, rather than a press address that routes to nobody.
Who works on what.
Mohit Bhardwaj
Head of Research, 2023 to March 2025 · Skill measurement · Agent evaluation · Study designLed the research programme from January 2023 to March 2025: the comprehension item bank, the statistics of code-model leaderboards, review-state telemetry, variance in agent evaluations, and the design of the twelve-month cohort study. Author of the five reports listed below.
- Designing a twelve-month cohort study of agentic coding: power, confounding and attrition by simulation · AWL-2024-03
- One run is an anecdote: variance in repository-level agent evaluations · AWL-2024-02
- Reading the review state: a hidden Markov model of how engineers accept AI suggestions · AWL-2024-01
- Pass@k is not a ranking: the statistics of code-model leaderboards · AWL-2023-02
- An item bank for code comprehension: calibration, adaptive testing and fairness checks with item response theory · AWL-2023-01
AI Work Lab Measurement team
Research group · Skill measurementBuilds and validates AI Work Lab’s comprehension instruments and ran the twelve-month cohort study. Papers are written jointly and reviewed by the external advisors before publication.
AI Work Lab Learning Science team
Research group · Learning scienceDecides where friction and retrieval go in the product, and runs the placement and scheduling experiments behind those decisions.
AI Work Lab Evaluation team
Research group · Agent evaluationTests the teaching layer across the models engineers bring, and owns the concept tagger that populates the skill graph.
AI Work Lab Org Insights team
Research group · Org signalsWorks on team-level signals, such as how concentrated understanding of a module is, and on whether a warning arrives early enough to act on.
AI Work Lab Security & Privacy team
Research group · Measurement integrityOwns what a product holding per-engineer comprehension data may show, and the re-identification analysis that sets those limits.
People who tell us when we are wrong.
Three external advisors review methodology before publication. They are paid for their time, they hold no equity, and they are free to disagree with us publicly. Two have.
Prof. Elena Rusk
Learning sciences, Utrecht University (visiting)Reviews study design before publication and is the reason the comprehension rubric is being externally validated rather than defended.
Dr. Adaeze Nwosu
Empirical software engineering, independentAdvises on the cohort methodology and has twice told us a result would not survive peer review.
Hiroshi Tanabe
Data protection and worker rights, independent counselAdvises on what our aggregation constraints should mean in practice, including where they cost us revenue.
Research roles here ship.
Findings go into the product or they go into a paper explaining why they did not.
You publish what you find
No internal review can suppress a result for being commercially inconvenient. Three of our thirteen papers contain findings that are bad for us.
Prior work stays yours
We do not assign anything you built before you arrived.
Real cohorts, real constraints
You will work with engineers at companies who agreed to be measured, under commitments about what we can publish. More interesting and more restrictive than a public dataset.