94% of engineering leaders say their AI metrics are missing. Here's how platform engineering is changing to close that gap.
Based on the InfoQ webinar "Platform Engineering in the Age of AI," featuring panelists from Harness, DKB, and Shine, August 18, 2026.
94% of engineering leaders say the AI metrics that matter most to them are missing. That single statistic framed a wide-ranging InfoQ panel on how platform engineering itself is changing as AI moves from an assistant developers use to an actor that operates inside the platform. The panel, featuring a Field CTO at Harness alongside platform engineers from DKB and Shine, split the conversation into two questions every platform team is now facing.
What should the platform provide and what should teams build themselves?
The first question isn't new to platform engineering — it's the same build-vs-provide tension platform teams have always managed — but AI raises the stakes. Should the platform standardize which AI coding tools are approved and how they're wired into the SDLC, or should individual teams pick their own? Panelists described talking directly to platform leaders about what's actually changed for them day to day, not just what's changed in theory.
How will AI consume platforms built for people?
The second, harder question: internal developer portals were designed around human workflows — a person clicking through a portal, opening a ticket, requesting an environment. As AI agents start consuming those same platforms directly, new questions about developer experience emerge that don't have established answers yet.
What does a platform's API need to expose for an agent to safely self-serve infrastructure or a deployment pipeline? What guardrails travel with that access?
The whole SDLC needs to be autonomous
One framing from the session stuck out: AI has already made the "coding" part of the software development lifecycle largely autonomous, but the rest of the SDLC hasn't accelerated at the same pace.
That mismatch is exactly where organizations are challenged: the fast part got faster, and the slow parts became the new constraint. If the “too slow” checks are circumvented or done in a slapdash way teams are feeling it with errors in production.
The missing metric: proving what AI investment produced
The 94% statistic was the panel's central challenge to attendees. Ask most engineering leaders what percentage of their AI-generated code actually shipped last quarter, or whether an AI agent in production is worth what it costs, and most don't have a ready answer.
The panel observed that this reporting gap won’t close on its own; it requires deliberately tracing token spend and agent activity to outcomes on both the developer productivity and infrastructure cost sides. This is exactly the approach Harness takes with its Cost Management Agent.
What platform teams should take away
Platform engineering isn't being replaced by AI. Today, it's being asked to do two new jobs at once:
- Decide what AI tooling belongs at the platform layer versus the team layer and
- Re-architect developer experience for an audience that now includes non-human consumers.
Teams that get ahead of both questions now will be the ones with a real answer the next time leadership asks what the AI investment actually produced.
