AI Coding Agent Integrations
Assign work to coding agents and track AI delivery hand-offs.
Additional cost — usage-based, BYOK supported
Overview
Connect Productivity Workbench work items to coding agents such as GitHub Copilot Coding Agent, Claude Code, and custom runners. Track run URLs and status, feed vulnerabilities into Tech Debt, and measure hand-off effectiveness without bundling in the separate inline AI Assistance bolt-on.
Features
- Assign work items to AI agents (GitHub Copilot Coding Agent, Claude Code, custom agents) with run history on the item.
- Pre-fill GitHub issue creation and Claude prompt hand-offs from the story or work item context.
- Bug & vulnerability scanner inbox — feeds Tech Debt with proposed remediations.
- AI Effectiveness report: hand-off volume, acceptance and completion trends, and recent AI-touched work items.
- Audit-ready provenance for AI-touched work items and agent runs.
- Safe-by-default agent workflows with explicit assignment, status updates and reviewable links back to the run.
Benefits
- Reduce the friction of handing backlog items to coding agents and external automation tools.
- Keep delivery teams working from the same source-of-truth story while still using external agent platforms.
- Make agent work auditable and reviewable — fit for regulated enterprise.
- Surface bugs and vulnerabilities continuously, with a fix proposal attached.
- Measure the impact: know exactly which AI-assisted delivery paths paid back and which did not.
- Adopt coding-agent workflows independently from the separate AI Assistance bolt-on.
How it works
- 1. Turn on the bolt-on
Enable AI Coding Agent Integrations on the Capabilities page. The toggle is available on every subscription tier; turning it on adds a usage-based line to the workspace bill.
- 2. Configure the integrations
On the Preferences and Integrations pages, configure the AI provider for prompts where needed and the external destination such as GitHub or a custom agent endpoint.
- 3. Assign work to an agent
On supported work items such as stories, bugs, spikes and tech-debt entries, assign GitHub Copilot Coding Agent, Claude Code or a custom agent and capture the run link.
- 4. Track delivery and review
Update run status, record the run URL, and keep independent review and acceptance inside the normal delivery workflow.
- 5. Measure with the AI Effectiveness report
Track agent-assigned work volume, recent AI-touched changes, and whether the hand-off pattern is paying back.
See it in action
These mockups show the coding-agent workflow surfaces that appear when this bolt-on is turned on.
Agent Assignment & Vulnerability Inbox
Assign stories to GitHub Copilot or Claude Code in one click. Scanner-fed vulnerabilities land in Tech Debt with CVE, CVSS and SLA deadline pre-filled.
AI Effectiveness Report
Track acceptance rate, cost per feature, agent-assigned items and escaped defects. Cost dashboard breaks down spend by model and provider.
Frequently asked questions
- Does turning on AI Coding Agent Integrations change anything for users who do not use agents?
- No. The bolt-on defaults to off everywhere. With it off, no AI controls render and no AI calls are made — non-AI users see no change at all.
- How is this different from the AI Assistance bolt-on?
- AI Assistance controls the inline drafting buttons inside Productivity Workbench. AI Coding Agent Integrations covers external coding-agent hand-offs, issue creation, vulnerability inbox workflows and agent run tracking. They are separate bolt-ons and can be enabled independently.
- Can I use GitHub Copilot Coding Agent and Claude Code?
- Yes. The bolt-on is designed to support GitHub Copilot Coding Agent, Claude Code and custom agent targets, with the run URL stored back on the work item.
- Is it safe for regulated industries?
- Yes, by design. Agent assignments and AI-touched workflow events are captured with provenance in the audit trail so reviewers can see what was handed off, where it ran, and what happened next.
- How is it priced?
- AI Coding Agent Integrations is a paid bolt-on on top of your subscription. Pricing is usage-based; talk to sales for committed-use pricing and governance options.
AI Coding Agent Integrations vs. manual hand-offs
Productivity Workbench supports both approaches. Here is the difference with the bolt-on turned on:
- Product owner copies stories manually into external coding tools
- Agent runs are tracked in chat threads, not on the work item
- Implementation hand-offs depend on copy-paste and human memory
- Vulnerabilities discovered during testing, not upfront
- No record of which work items were handed to agents and what happened next
- Engineering team triages bugs as they appear in production
- No data on whether coding-agent workflows are paying back
- AI agents assigned directly to stories; GitHub Issues pre-filled automatically
- Claude Code or custom-agent prompts launched from the same source-of-truth work item
- Vulnerability scanner feeds the Tech Debt backlog with CVE, CVSS and SLA deadline
- Every AI-touched workflow event logged to the Audit Log with provenance
- Run URLs and statuses stay attached to the story instead of living in chat or email
- AI Effectiveness report shows hand-off volume and completion trends
Turn on AI Coding Agent Integrations
Sign in to your workspace, open the Capabilities page and toggle the bolt-on on.