evidoria

← Back to browse

Good practice Imported

Poppy — California's Statewide Generative-AI Platform for State Employees

United States of America · Sacramento · See the United States of America profile · See the Sacramento profile

Evidence: Descriptive / self-reported Top 25% 73/100 · Ask Evidence Copilot about this practice

California's CDT built Poppy, a vendor-agnostic GenAI platform offering ten models to state staff. After a nine-month pilot with 2,800 employees across 67 departments, it went statewide on 1 July 2026, and European officials are studying its model-agnostic design.

10
Underlying AI models available to state employees
2,800
Employees who tested the platform in the pilot (pilot, from 29 Sep 2025 (9 months))
67
State departments participating in the pilot (pilot, from 29 Sep 2025 (9 months))
Poppy — California's Statewide Generative-AI Platform for State Employees

Details

Maturity
Scaling
Promoter
California Department of Technology (CDT)
Period
2025–2026
Keywords
digital government, generative AI, public-sector IT

Context

California's Department of Technology (CDT) built Poppy, a vendor-agnostic generative-AI platform configured to the state's own data, security and operational requirements. Unlike a single-vendor chatbot, Poppy gives state employees a choice of ten different underlying models — including tools from Anthropic — so staff can match the model to the task at hand, view side-by-side comparisons, and use customized workspaces integrated with existing state systems.

Objectives

The platform aims to let state employees pick the right model for each task rather than being locked into one vendor's chatbot, while meeting the state's own data-security and operational requirements.

Activities

The platform closed a nine-month pilot phase (from 29 September 2025) that involved more than 2,800 public employees across 67 state departments testing the tool and providing feedback. On 1 July 2026, Poppy became available statewide across California's entire administration, and Governor Newsom's office separately announced a partnership bringing additional Anthropic tools to state agencies.

Results

French AI outlet Actuia and Government Technology magazine have both singled out Poppy's vendor-agnostic, multi-model architecture as a design pattern European public administrations are now studying.

Conclusions

Coverage to date is adoption and rollout reporting rather than an independently measured productivity or service-quality study — CDT has not published time-savings or error-rate data comparable to, say, the UK's controlled Microsoft 365 Copilot trial.

Implementation
Indicative cost
Medium (€50k–€500k) — No published budget figure; the platform licenses multiple commercial model providers, including Anthropic, rather than building foundation models in-house.
Time to results
Medium (1–3 years) — Nine-month pilot from 29 September 2025 with 2,800 employees across 67 departments, followed by statewide rollout on 1 July 2026.
Staffing & skills
California Department of Technology (CDT), which built and operates the platform, state employees across 67 departments who tested it during the pilot, Governor Newsom's office, which announced the Anthropic partnership

Conditions for success

  • a vendor-agnostic architecture integrating multiple commercial model providers under the state's own security and data requirements
  • a structured nine-month pilot with 2,800 employees across 67 departments gathering feedback before statewide rollout
  • executive-level backing, including a Governor's-office partnership announcement.

Common failure modes

  • no independently measured productivity or service-quality data has been published, so adoption scale is documented but impact is not yet.

Where it fits

Governance type
state government technology department
Scale
statewide (US state administration)
Income level
high-income

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.

Attachments

Similar practices you may find useful