Services

AI & Agents

Spend that scales once a controlled pilot has measured the effect: agents and copilots for program, back-office and public-facing work, built in your agency's tenant on the cloud services your authorization covers.

How a pilot runs

Proof of Value: experimental design for AI ROI

Clinical-trial discipline for AI. Would this have happened anyway?

  1. Pre-register the hypothesis and one KPI worth paying for.
  2. Choose the design: A/B, stepped-wedge, difference-in-differences or interrupted time series.
  3. Instrument the baseline before the pilot (Databricks, Fabric or Foundry telemetry).
  4. Measure the effect size with a confidence interval against the control.
  5. Decide: scale, redesign or stop — a quick answer is still an answer.
The rules this work is mapped to

What the rules ask of your agency, and what we do about it

Each card gives the rule, its status, the words that matter and our part in the work. Your agency makes the determinations; we provide the evidence and the documentation.

U.S. federal (executive agencies)Key date: April 3, 2025

OMB M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust

In force; rescinds and replaces M-24-10. Minimum practices for high-impact AI were due within 365 days (April 3, 2026).

“the intended purpose for the AI and its expected benefit, supported by specific metrics or qualitative analysis”

What we do: We produce the evidence an AI impact assessment asks for: the expected benefit against the current process, the costs and expected savings, the pre-deployment test results and the monitoring plan, and your agency's reviewers make the determinations.

Source: OMB M-25-21 (whitehouse.gov, PDF) (opens in a new tab) · Checked November 10, 2026

U.S. federal (executive agencies)Key date: April 3, 2025

OMB M-25-22: Driving Efficient Acquisition of Artificial Intelligence in Government

In force; rescinds and replaces M-24-18. Applies to contracts awarded from solicitations issued on or after September 30, 2025 (180 days after issuance) and to options exercised after that date; it does not apply to AI acquired as a component of a National Security System.

“Ensuring contracts permanently prohibit the use of non-public inputted agency data and outputted results to further train publicly or commercially available AI algorithms, consistent with applicable law, absent explicit agency consent”

What we do: Our contract terms keep your data out of model training without your written consent, keep your data and the work we build for you portable, make knowledge transfer a deliverable, and give a performance-based contract a pre-registered measure it can use.

Source: OMB M-25-22 (whitehouse.gov, PDF) (opens in a new tab) · Checked November 10, 2026

U.S. federal (executive agencies; LLM procurement)Key date: December 11, 2025

OMB M-26-04: Increasing Public Trust in Artificial Intelligence Through Unbiased AI Principles

In force (issued December 11, 2025); agency policies were due March 11, 2026, and existing LLM contracts should be modified, at the latest, before an option extends the period of performance. The memo ceases to have effect on December 11, 2027 unless the OMB Director provides otherwise.

“No later than March 11, 2026, agencies must update their policies and procedures to ensure that contracts for the procurement of LLMs include contractual requirements addressing compliance with the Unbiased AI Principles.”

What we do: For every model we deploy, we gather the vendor documentation your acquisition team requests under M-26-04, such as the acceptable use policy and model or system cards, and set up the user-reporting route your policy describes.

Source: OMB M-26-04 (whitehouse.gov, PDF) (opens in a new tab) · Checked November 10, 2026

U.S. federal (voluntary framework; named by Texas law)Key date: January 26, 2023

NIST AI Risk Management Framework (AI RMF 1.0) and the Generative AI Profile (NIST AI 600-1)

AI RMF 1.0 released January 26, 2023 and being revised under the AI Action Plan; NIST AI 600-1 released July 26, 2024.

“The AI RMF 1.0 is being revised as part of the White House AI Action Plan.”

What we do: We map the delivery to the NIST AI RMF and its Generative AI Profile, and keep the risk screen, the test results and the review records with your team.

Source: NIST AI Risk Management Framework (nist.gov) (opens in a new tab) · Checked November 10, 2026

U.S. federalKey date: April 13, 2026

2025 Federal Agency AI Use Case Inventory (OMB)

Published repository; summary as of April 13, 2026.

“3,611 individually-reported AI use cases (all stages of development)”

What we do: We turn an agency's inventory entries into a ranked roadmap with a business case for the use cases worth funding first.

Source: 2025 Federal Agency AI Use Case Inventory (github.com) (opens in a new tab) · Checked November 10, 2026

TexasKey date: January 1, 2026

Texas TRAIGA (HB 149): disclosure when a governmental agency's AI interacts with consumers (Bus. & Com. Code §552.051)

In force since January 1, 2026.

“shall disclose to each consumer, before or at the time of interaction”

What we do: Every public-facing agent we build for a Texas agency tells people they are talking to AI before the conversation starts, in wording your counsel approves.

Source: Texas TRAIGA §552.051 disclosure (capitol.texas.gov) (opens in a new tab) · Checked November 10, 2026

Texas (state agencies and local governments)Key date: September 1, 2025

Texas SB 1964: heightened scrutiny AI systems, the AI code of ethics and vendor duties (Gov. Code §§2054.003, 2054.702, 2054.703, 2054.708)

In force since September 1, 2025; DIR's implementing rules (1 TAC chapter 219: the AI code of ethics, the risk and impact assessments and the vendor risk-framework requirement) took effect March 18, 2026, and DIR publishes the standardized notice form (AI System Facts).

“A state agency that deploys or uses a heightened scrutiny artificial intelligence system or a vendor that contracts with a state agency for the deployment or use of a heightened scrutiny artificial intelligence system shall conduct a system assessment that outlines:”

What we do: When a system will make, or be a controlling factor in, a decision about access to a government service, we document whether it meets the heightened-scrutiny definition for your agency's inventory, draft the risk assessment DIR's rules ask for, write the impact assessment Texas asks of a state agency and its vendor (or the review a local government completes), and prepare DIR's standardized notice for your agency to publish.

Source: Texas SB 1964 (capitol.texas.gov) (opens in a new tab) · Checked November 10, 2026

U.S. (state and local governments)Key date: April 20, 2026

ADA Title II web and mobile accessibility rule: compliance dates extended (DOJ interim final rule)

IFR (91 FR 20902) effective April 20, 2026: 50,000 or more → April 26, 2027; under 50,000 and special districts → April 26, 2028; WCAG 2.1 AA stands as adopted in 2024. DOJ plans future rulemaking on the substance, and NFB v. DOJ, No. 1:26-cv-02007 (D. Md., filed May 21, 2026), which asks the court to vacate the extension, is pending.

“The compliance date for public entities with a total population of less than 50,000, or any special district government, is extended from April 26, 2027, to April 26, 2028.”

What we do: Public-facing apps, agents and dashboards we build for state and local governments are built and tested to WCAG 2.1 Level AA, and remediation against your compliance date becomes a costed line in the business case.

Source: ADA Title II web and mobile accessibility rule (govinfo.gov) (opens in a new tab) · Checked November 12, 2026

Every rule on this site, by service: the regulation map.

Platforms

Availability in government clouds

Availability differs by cloud, so each row carries its own status and source.

Platform availability in government clouds for AI & Agents
PlatformStatus in government cloudsSource
Azure Government: authorizations and regionsUS Gov Arizona, US Gov Texas and US Gov Virginia: FedRAMP High P-ATO and DoD IL2, IL4 and IL5 PAs (IL5 with workload isolation); Azure Government Secret: DoD IL6. FedRAMP Marketplace (November 10, 2026): FedRAMP Certified, Class D (High), package F1603087869.Azure Government (learn.microsoft.com) (opens in a new tab)
Checked November 12, 2026
Azure OpenAI in Azure GovernmentIn audit scope and authorized: FedRAMP High, DoD IL2, IL4, IL5 (workload isolation) in Azure Government; DoD IL6 in Azure Government Secret.Azure OpenAI in Azure Government (learn.microsoft.com) (opens in a new tab)
Checked November 12, 2026
Azure Databricks in Azure Government; Databricks on AWS GovCloudAzure Databricks in Azure Government (US Gov Arizona and US Gov Virginia): FedRAMP High, DoD IL2, IL4 and IL5 (workload isolation), within Microsoft's Azure Government package (FedRAMP Marketplace F1603087869). Databricks on AWS GovCloud: FedRAMP Certified, Class D (High), since February 26, 2025 (FedRAMP Marketplace FR2324740262); a separate GovCloud DoD environment for DoD agencies with IL5 requirements.Azure Databricks in Azure Government; Databricks on AWS GovCloud (learn.microsoft.com) (opens in a new tab)
Checked November 12, 2026
Microsoft 365 Copilot in U.S. government cloudsAvailable in GCC, GCC High and DoD; features differ per environment.Microsoft 365 Copilot in U.S. government clouds (learn.microsoft.com) (opens in a new tab)
Checked November 12, 2026
Where to start

A first milestone your budget office can plan for

Each offer is the first milestone of an engagement and ends in a decision meeting, before any further budget is committed. All offers: offers for agencies.

AI Value Experiment (Proof of Value)

Did the AI move the mission measure, and is the evidence strong enough to scale it?

What we look at

  • Pre-registered hypothesis and one mission KPI
  • A/B, stepped-wedge or difference-in-differences design
  • Instrumented pilot on services your agency has authorized, such as Azure OpenAI in Azure Government, with Azure Databricks or Microsoft Fabric where the data authorization fits
  • Effect size with confidence intervals, written up as pre-deployment testing evidence
  • A measure your contract can carry in its quality assurance surveillance plan
  • Scale-or-stop decision

What you get

  • Experiment protocol
  • Measurement dashboard
  • Readout with effect size for your AI impact assessment
  • Scale-or-stop decision

Each milestone is quoted in writing before your team chooses how to buy it. Cloud consumption, software licenses and equipment are separate line items or are furnished by your agency.

AI Operating Model & Governance Diagnostic

Which tasks deserve agents, which stay with people, and how does each AI use case move through your agency's governance?

What we look at

  • Task-exposure analysis across programs and back-office functions
  • AI governance workflow: use-case inventory, evidence for high-impact (federal) or heightened scrutiny (Texas) determinations, and the generative AI policy put into practice
  • Human-AI workflow design with your staff, building on the AI training they complete
  • Model documentation pack and the user-reporting route your LLM procurement policy describes
  • AI center of excellence design

What you get

  • Operating-model blueprint
  • Governance workflow and use-case register
  • Role cards
  • Change plan

Each milestone is quoted in writing before your team chooses how to buy it. Cloud consumption, software licenses and equipment are separate line items or are furnished by your agency.

Evidence

Delivered work, labelled by type

Figures from commercial client delivery and our own R&D, each labelled; clients are described rather than named.

10+
Public sources monitored by our own opportunity-intelligence platform on Azure
Internal R&D
See the case
93
Backend tests in our own opportunity-intelligence platform
Internal R&D
See the case
15–20 days → 3–5 days
From first lead to outreach at a manufacturing sales organization
The client's own before-and-after figures; directional.
Client delivery · commercial
See the case

Bring one business case.

In a 45-minute Discovery Session we look at its requirements and your procurement situation, and identify the most efficient way for your agency to buy the first milestone.