AI
Enterprise Controls for Agentic Development
GitHub’s platform announcement at Universe 2025 rests on four enterprise controls: observability (see what agents are doing), auditability (trust every action), policy enforcement (stay in command), and customization (adapt to your org’s standards). Mario Rodriguez framed this not as a technology shift but a work-pattern shift. Developers and agents now co-author software, and that collaboration only scales if the enterprise can see, audit, and control everything those agents touch.
Custom Agents Across the SDLC
Custom agents are Copilot workflows defined in a single markdown file, each baking in tool sets, coding conventions, and a CodeQL security scan for one specific job. The Amplitude custom agent end-to-end demo (07:22) ran a GitHub issue from PRD to live experiment, calling the Playwright and Amplitude MCPs along the way. ARM shipped a repo migration agent; GitHub’s docs team built one that auto-repairs knowledge-base gaps.
“Literally anyone in your organization could make a custom agent with this kind” — Kate Catlin
The live custom agent creation in VS Code (12:29) confirmed it: paste markdown, save, the agent appears instantly. Scoped at repo, org, or enterprise level, one agent propagates standards across thousands of repos.
Measuring Copilot Adoption with Metrics
Copilot Metrics launched today, designed with HP, HSBC, and Lumen. The dashboard tracks daily and weekly active users, chat requests per user broken down by feature (edit, ask, agent, custom), and completion rates per language. A downloadable JSON report slices the same data by user, IDE, model, and interaction count. Todd Manion answered a CTO’s real question live: he pulled the JSON into Excel, ran a Power Pivot table, and found that the notifications team had low acceptance counts while high-adoption teams could be cloned for coaching. A metrics API supports custom dashboards.
Governing Agents with the Control Plane
The agent control plane launched today in public preview as a single panel inside GitHub Enterprise settings. The agent control plane live demo (28:05) showed three capabilities. First, enterprise-wide agent rollout: Mario pushed the full set of custom agents to every org in Avocado Corp with one click, then used a repo ruleset to restrict edits so only enterprise owners can add new agents. Second, session inspection: admins see every running agent session across the enterprise and can teleport directly into the session’s VS Code workspace. Third, license control: access for any developer can be revoked from the same panel.
“you are already seeing how our Coding agent is the number one contributor” — Mario Rodriguez
Notable Quotes
because I believe this is the most important thing we are shipping Kate Catlin · ▶ 4:44
Literally anyone in your organization could make a custom agent with this kind Kate Catlin · ▶ 13:51
you are already seeing how our Coding agent is the number one contributor Mario Rodriguez · ▶ 35:59
Key Takeaways
- Custom agents let any team encode domain expertise, tool sets, and security scans into a markdown file deployable at enterprise scope.
- Copilot Metrics tracks per-user, per-team, per-language usage so leaders can clone high-adoption patterns and coach low-adoption teams.
- The agent control plane gives enterprise admins a single panel to install, audit, and revoke agents or developer licenses across the org.