GSA's internal chatbot GSAi, launched March 2025, is now used daily by roughly half of GSA's employees; officials say AI adoption jumped from 15% to 70% of staff and claim ~900,000 hours saved — though the hours figure rests on unaudited self-reporting.
70 % of workforce
Staff regularly using AI (mid-2026, up from ~15%)
~50 % of employees
Employees using GSAi daily (mid-2026)
400,000 hours
Hours attributed to direct automation
500,000 hours
Hours attributed to self-identified optimisable/eliminable processes
~900,000 hours
Total claimed hours saved
25
Federal agencies onboarded to USAi (mid-2026)
3.4 million
Estimated users reached via OneGov
Details
Maturity
Scaling
Promoter
U.S. General Services Administration (GSA)
Period
March 2025 – ongoing
Keywords
generative AI chatbot, federal workforce productivity, government efficiency, multi-agency AI platform
Context
In March 2025 the U.S. General Services Administration (GSA) rolled out GSAi, an internal generative-AI chatbot giving employees access to models from OpenAI, Anthropic and Google inside a GSA-controlled environment, after an internal AI safety team evaluated vendors.
Objectives
GSAi aims to raise AI adoption and productivity across the GSA workforce, and to seed a shared-tenant platform, 'USAi', that other federal agencies can adopt without building their own AI infrastructure.
Activities
Following internal deployment at GSA, USAi was extended to other federal agencies as part of the OneGov initiative: 25 agencies had onboarded by mid-2026, with 16 more planned by year end, together reaching an estimated 3.4 million government users.
Results
GSA Deputy Administrator Michael Lynch reported in June 2026 that regular AI use across the agency rose from about 15% of the workforce at the start of the current administration to roughly 70%, with GSAi used daily by nearly half of employees. Officials attribute approximately 400,000 hours of savings to direct automation and a further 500,000 hours to processes staff identified as optimisable or eliminable with AI, a combined total described as roughly 900,000 hours.
Conclusions
These adoption and hours-saved figures come from agency officials' public statements, reported independently by Nextgov/FCW and FedManager, rather than from a published, audited methodology. Neither outlet found that GSA had disclosed how the hour estimates were calculated, and the 500,000-hour figure explicitly rests on unverified employee self-reporting rather than measured time-and-motion data.
Implementation
Indicative cost
Medium (€50k–€500k) — No public budget figures found; GSAi/USAi provides API-based access to third-party foundation models within a GSA-controlled environment rather than a custom-built model.
Time to results
Short (< 1 year) — Launched March 2025; workforce adoption rose from ~15% to ~70% and the platform expanded to 25+ agencies within roughly 15 months (by mid-2026), with 16 further agencies planned by end of 2026.
Staffing & skills
internal GSA AI safety team (vendor evaluation), GSA IT/technology staff operating GSAi and USAi, onboarding IT staff at each adopting federal agency
Conditions for success
GSA-controlled environment giving vetted access to major LLM vendors (OpenAI, Anthropic, Google)
shared-tenant USAi architecture letting other agencies onboard without building their own infrastructure
executive-level sponsorship and public reporting of adoption figures
Common failure modes
no published, audited methodology behind the headline hours-saved figures
the larger of the two hours-saved components rests on unverified employee self-reporting rather than measured data
no independent oversight body or audit trail described in reporting
Where it fits
Governance type
U.S. federal government agency, extending to a multi-agency shared platform
Scale
national, 25+ agencies and an estimated 3.4 million users
Income level
high-income (United States)
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.