Platform

A governed execution layer for Azure DevOps.

Agents4DevOps combines an agent runtime, control plane, reusable workflow packs, and industry-specific engineering logic. It installs under the Boards tab in Azure DevOps, so teams automate routine work without giving every agent unrestricted access — or treating every task as an open-ended prompt.

Architecture

Four platform layers

Each run is composed from the same layered system — from what starts it to the engineering standards it enforces.

LayerPurpose
Agent runtimeStarts and executes runs from requests, Azure DevOps events, schedules, APIs, or other workflows — with tool calls, dependencies, retries, cancellations, and outputs.
Control planeDefines scope, models, instructions, tools, permissions, approvals, verification, and run-history behavior for every workflow.
Workflow packsProvide reusable implementations for backlog, code, pipelines, tests, releases, operations, and evidence — available from a prebuilt Agent Library of templates.
Industry packsApply classifications, standards, traceability, lifecycle gates, and evidence expectations to the work agents prepare.
From event to verified result

From event to verified result.

Configure the workflow once, then run it manually, from an Azure DevOps event, on a schedule, or through an API. Each run operates within its assigned scope and control policy.

1

Trigger

A request, ADO event, schedule, or API call starts the run.

2

Context

It loads the permitted work items, code, runs, results, and policies.

3

Decide

It applies versioned rules or plans a sequence within scope.

4

Act

It analyzes, drafts, updates data, or invokes an approved operation.

5

Approve

Actions with side effects pause for the right reviewer.

6

Verify

Post-action checks confirm the expected result exists.

7

Record

The run record captures inputs, actions, approvals, and outcome.

Two execution models

Fixed logic where the rule is known. Planning where the work is open.

Use deterministic execution agents for controls you can specify, and planner agents where the work requires investigation or drafting — both run on one execution model.

Rule-based agents
  • Run deterministic calculations, validation, and link checks
  • Enforce required fields, links, tags, and routing
  • Update structured Azure DevOps data
  • Run on work-item events or schedules — Execution agents generate C# on the triggering event
Planner agents
  • Investigate across work items, code, runs, and results
  • Draft requirements, reviews, and reports
  • Select from approved skills and tools within scope
  • Pause for review — Assisted Autonomous agents are LLM-powered with manual, scheduled, or event triggers, optional approval gates, and a knowledge base
Control what each workflow can do

Configure the scope. Govern the actions.

Configure at the level the workflow requires, then decide where humans stay in the loop.

CONFIGURE

Set the boundaries of the run

  • Scope to the organization, project, team, repository, branch, pipeline, test plan, area path, iteration, or work item type
  • Select hosted or bring-your-own model options per plan and configuration
  • Attach reusable skills and permitted Azure DevOps or external tools
  • Set event triggers, schedules, manual inputs, and output destinations
GOVERN

Decide where humans stay in the loop

  • Define which actions are read-only, create drafts, update data, or require approval
  • Define post-action checks so a run isn't treated as successful until the expected result is present
  • Hold generated criteria or material changes behind approval gates
  • Every execution creates a Job with full logs and audit history — comments note when an AI agent made a change
Platform capabilities

What the runtime gives every workflow.

The same governed building blocks — runtime, triggers, skills, orchestration, approvals, and verification — apply to every agent you configure.

Agent runtime

Starts and executes runs with tool calls, dependencies, retries, cancellations, and outputs. A prebuilt Agent Library of templates gets teams to a working workflow fast.

Execution states · Retries · Outputs

Triggers & scheduling

Run agents manually, when Azure DevOps artifacts change, on a recurring cadence, or from another workflow — moving from one-off assistance to repeatable execution.

Manual · Event · Scheduled

Skills, tools & MCP

Compose agents from a reusable Skills library and permissioned Azure DevOps tools. MCP connectivity for external data and actions is on the roadmap, coming in v2.

Skills library · Tools · MCP (v2)

Multi-agent orchestration

Coordinate several agents, handoffs, and gates for lifecycle-spanning work. Agents also run from Copilot4DevOps chat in single-agent or multi-agent orchestrator mode.

Handoffs · Gates · Orchestrator

Governance & approvals

Define read-only, draft, update, and operational levels, set approval requirements per action, and enforce existing Azure DevOps permissions and project scope on every run.

Scoped access · Approval gates

Verification & run history

Post-action checks confirm the expected state, link, or result exists. Every execution creates a Job in a Jobs dashboard with full logs and audit history.

Post-action checks · Jobs dashboard

BYOD semantically indexes your work items to find similar items — giving planner agents relevant context from your own backlog. Browse workflow packs →

Get started

Evaluate the platform through one complete run.

The most useful evaluation is not a catalog tour. Select one workflow and inspect how it is configured, triggered, executed, approved, verified, and recorded.