Colleague AI Implementation Study:
2026
Read Colleague AI Implementation Study: (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-16277.txt.
Introduction
Esbenshade and eight colleagues warn that links between teacher use and student traits here are early signals only. Just some of the 12 Washington State districts supplied midyear administrative data. The report follows teacher use of Colleague AI, blending platform records with district records. The window runs September 1 to December 31, 2025. Districts run from a few thousand pupils to 30,000 and span rural, suburban, and urban settings.
Catalog record
| YEAR | 2026 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35052 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.16277
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Colleague AI Implementation Study: Abstract Cross-District Midyear Analysis \KOMAoption captionstableheading,figureheading See pages 1 of midyear_report.pdf Executive Summary This mid-year report summarizes teacher use of Colleague AI across 12 Washington State school districts from September 1 to December 31, 2025. Produced jointly by Colleague AI and AmplifyLearn.AI at the University of Washington, this report aggregates platform data and district-provided administrative records to provide an early look at how teachers engaged with AI during the first half of the 2025-26 school year. The districts vary in size from small districts with a few thousand students to large districts with up to thirty thousand students. The districts are rural, suburban, and urban. Only a subset of districts were able to provide mid-year administrative data, and findings that link teachers’ use of Colleague AI to student characteristics should be interpreted as preliminary signals. Key Findings • Adoption grew steadily over the fall. Across 12 participating districts, 1,438 teachers used Colleague AI in September and a total of 2,891 teachers had used Colleague AI by the end of December, exchanging over 412,304 messages with the AI assistant. We have observed continued growth in platform adoption; as of March 31, 2026 adoption across these districts has increased by 61% to 4,644 teachers. • Among active users, teachers averaged 12 sessions and 3.7 hours on the platform. Per session, teachers spent about 18 minutes. Teachers with more sessions also tended to have longer sessions and send more messages. • Claire AI and lesson planning dominated usage. Conversations with the Claire AI assistant accounted for the large majority of platform activity, and lesson planning features made up 60-99% of usage in every participating district. Frequent users concentrated even more heavily on Claire, while lighter users and later adopters explored a broader mix of features. • Teachers serving different student populations engaged with AI differently. While overall usage intensity was similar across classroom contexts, teachers serving higher shares of English Language Learners or Special Education students referenced Student Profiles and Differentiation & Accessibility more often in their requests. Teachers serving more SPED students and those with wider variability in their students’ test scores also made relatively more use of Lesson Delivery features like Generate Image and Generate Interactive. • Small, positive relationship between Teacher AI Use and Student ELA scores. Examining 12,005 students with matched beginning- and middle-of-year test scores, we found that average tested students’ teachers used Colleague AI for approximately 3 hours, and this use was associated with a statistically significant 0.025 standard deviation increase in reading interim test scores; the corresponding association in math was not statistically distinguishable from zero. This is a preliminary correlation, and the small effect size is an encouraging signal for AI adoption. Analysis of pedagogical topics in teacher request messages to the AI platform found that across all districts, Planning and Explicit Teaching were the most common pedagogical topics in teacher requests, followed by Assessment, Critical Thinking and Inquiry, and Student Profiles. Non-educational conversations were rare. Comparing teacher requests to AI responses, we found that AI responses consistently surfaced additional pedagogical topics—particularly around Feedback, Student Profiles, and Differentiation—suggesting that Colleague AI is not only responding to what teachers ask but also broadening the pedagogical scope of the conversation. The one topic more prevalent in teacher messages than in AI responses was Discourse Continuity, reflecting teachers’ active role in following up, refining, and redirecting AI output across conversational turns. Teachers serving different student populations used Colleague AI in distinct ways. While overall usage intensity was similar across teachers regardless of the proportion of English Language Learners (ELL) or Special Education (SPED) students in their classrooms, the features and topics teachers engaged with differed meaningfully. Teachers serving higher shares of ELL students referenced Student Profiles, Differentiation and Accessibility, Engagement and Motivation, and Project-Based Learning more frequently in their requests. Teachers serving more SPED students similarly emphasized Student Profiles and Differentiation, and were notably more likely to use Lesson Delivery features such as Generate Image and Generate Interactive. Teachers whose classrooms had lower average test scores, or wider score spreads, were also more likely to reference Differentiation, Student Profiles, and Engagement, pointing to AI as a potential resource for supporting instructional differentiation in heterogeneous classrooms. Preliminary analysis links teacher AI use to modest gains in reading scores. Analyzing 12,005 students, we examined whether students whose teachers used Colleague AI more intensively had higher mid-year scores, controlling for beginning of year scores, student characteristics, and assessment and district fixed effects. In reading, three hours of teacher AI use was associated with a statistically significant increase of approximately 0.025 standard deviations in mid-year scores, a result that held across model specifications and alternative measures of AI engagement. The corresponding association in math was not statistically distinguishable from zero. These findings are correlational and reflect only the first four months of a voluntary adoption period, but this offers an encouraging early signal. We note that teacher adoption of Colleague AI was voluntary, and we may be observing a selection effect where the teachers using Colleague AI differ from other teachers in important ways that also influence student learning. Districts included in this report agreed to participate in the study prior to the beginning of the 2024-25 school year and completed Data Use Agreements with Colleague AI confirming their participation. Among these, some districts paused their AI implementation plans during the study period or elected to pilot alternative AI tools, resulting in variable levels of platform engagement. The districts included in this study represent a small subset of Colleague AI’s overall user base. Individual districts are not identified in this report. This research has been approved by the University of Washington Institutional Review Board. Colleague AI has provided training resources and professional development opportunities for teachers, all teacher use of Colleague AI is voluntary. All data analyzed here has been de-identified. Several of the twelve participating districts provided mid-year administrative records. These records enable analyses of usage patterns by student demographics and student achievement. These supplementary analyses are presented in later sections with sample sizes noted (see Appendix: Data Sources for the breakdown of data availability by file type.) For questions about this report please contact [email protected]. Teacher Usage Levels Usage of Colleague AI Increased Steadily Over the Fall 2891 teachers used Colleague AI as of December 31, 2025, representing nearly 30% of the teachers in the participating districts. For context, a nationally representative survey of K--12 science and mathematics teachers found that in May 2025 approximately half of teachers reported using generative AI tools, and most of those (44% of surveyed teachers) reported using ChatGPT rather than an education-specific platform (Esbenshade et al., 2025). In four of the districts participating in this study, teacher use of Colleague AI surpassed the national average rate of ChatGPT usage. 1 1 1 A note on the participation denominator: Platform rostering data reflects all accounts provisioned by the district on Colleague AI, which in many cases substantially exceeds the number of active classroom teachers. In districts that did not configure automated rostering, no denominator is possible as all observed accounts were manually created by individual teachers. We use the National Center for Education Statistics (NCES) full-time equivalent (FTE) teacher count for