AI
The Adoption Gap in Enterprise Enablement
Principal Financial Group migrated 30,000 repositories to GitHub, then discovered pipeline sprawl: dozens of CI/CD configurations, no consistent release process, and auditors unhappy with the inconsistency. An enterprise standard pipeline built on reusable GitHub Actions workflows was the answer on paper. Adoption stalled anyway. Teams said they’d prioritize it in a few months. What looked like a 20-minute task took 20 or more hours once invisible prerequisites surfaced.
Prompt libraries didn’t move the needle either. Connor Uhlman heard “this is amazing, we’ll get to it” repeatedly before accepting the real constraint: teams need you to own the outcome, not just hand over the tool.
“The easiest change I have is one that someone’s already done for me.” — Connor Uhlman (quoting Alan)
Making Work Visible: The Insights and Actions Portal
Principal built an internal developer portal called Insights and Actions: one place where every compliance gap, pipeline issue, and security finding lives, tied to the software catalog and its defined ownership. Engineers, team leads, and executives all see the same data. Remediation steps are specific to the repo, not generic. A Java app gets different instructions than a .NET app or a JavaScript web app.
The portal replaced scattered emails and Teams messages with a structured format. Platform teams bring data; the enablement team manages the experience. Teams no longer build their own dashboards.
Agentic Pipeline Migration: Button to Pull Request
When an engineer clicks “Automate with AI” in the portal, a custom workflow on AWS Bedrock fires. See the agentic pipeline migration demo (08:17). The workflow runs in steps: evaluate each prerequisite against markdown-defined criteria, format GitHub Issue specifications, create the issues via API, then hand off to GitHub Copilot, which starts the code changes automatically. Each step gets fresh context and a fine-tuned prompt.
Letting one agent run the whole task end-to-end failed in testing. The team settled on step-by-step execution with human checkpoints, mirroring spec-driven development. Engineers review each PR and merge when ready.
Scaling AI: Small Bets and ROI Measurement
Principal ran small bets over six months and changed strategy repeatedly. What shipped at GitHub Universe 2025 bore little resemblance to the original plan. AI amplifies weaknesses just as readily as strengths, so iterating fast kept failures small.
ROI measurement was built into the product from the start. An issue closed via the “Automate with AI” button typically closes in a day. An issue left unactioned often takes two weeks. At scale, that gap is the proof. Have a measurement answer ready before anyone asks whether the investment was worth it.
Notable Quotes
The easiest change I have is one that someone’s already done for me. Connor Uhlman (quoting Alan) · ▶ 3:48
Key Takeaways
- Offering prompts and tools is not enough. Teams need the work done for them to drive real adoption.
- Breaking agentic workflows into discrete steps with human checkpoints produces more reliable automation than end-to-end agent runs.
- Build ROI measurement into the product from day one so issue-close time proves the value automatically.