Turning software teams agentic; a leadership perspective
2026-06-09
Turning software teams agentic; a leadership perspective
Hey there! Welcome to Platform Weekly. Your weekly altitude check of platform engineering ambition. Every week, we bring you the sharpest thinking from the community, straight to your inbox.
Plus… 28.1% of teams (so far) say they are massively struggling with shadow AI. What about you? Answer our State of AI in Platform Engineering survey and let us know!
How to become an agentic organization
5 years ago when I was talking about platform engineering, it was hard to tell whether I was truly ahead of the curve or was the lunatic people kept telling me I was. For the first time in a while, I'm starting to get that feeling every day again. I love it.
My friend Ajay Chankramath, CTO at Platform Engineering Consulting and Community Ambassador ran an awesome webinar yesterday on what it actually takes to turn a software organization agentic. Not just walking through our theory of the four levels I broke down a few weeks ago, but the practical leadership decisions that determine whether your org actually climbs them. This is the same type of thing he and the team will be talking about at PlatformCon in a few weeks. If you’re in NYC or LDN, you need to come chat about it.
The four levels:
Level 1: Human in the loop
Level 2: Human on the loop
Level 3: Humans as orchestrators
Level 4: Fully autonomous
So why does this matter? Well, our data shows that 88% of firms are running AI coding experiments. 5.5% report significant business impact. I’ve said it before. It's not a model problem. It's whether your production systems are designed to capture value at organizational scale.
The majority of organizations are at L1 trying to get to L2. The tooling jump from L1 to L2 is truly tiny. What you have to do as a leader to move from L2 to L3 is what organizations are failing at.
Three bottlenecks we’ve noticed across the 100+ orgs he’s talked to this year:
Work dispatch: Your sprint model assumes sequential human work. When agents generate PRs in parallel, you need a proper dispatch path which means clear agent identity, scope definition, and workspace provisioning. Without it, agent work has nowhere to land.
Validation loops: Traditional validation is a gate. For agents, it needs to become a loop. Agents iterate through CI, security scans, and policy checks until they pass. Gate-based review at L2 volume will drown your engineering org.
Risk governance: Who approved that agent-generated change? Which non-human identity modified that API contract? Without agent-aware audit trails, you can't pass a regulatory audit. That's a current problem, not a future one.
Leadership alignment: None of these can be solved by engineering teams working independently. They require executive mandate. Consistency across the org determines whether agents run at scale, or whether you've just rebuilt tool sprawl with a language model on top.
Governance enables velocity. It doesn't restrict it. That's the operating principle you need to keep coming back to, and what distinguishes L3 organizations from the rest.
There is a reason we keep banging on about it. We're talking about operationalizing AI agents. It's the governance and cultural questions that are the most important. If you haven't read our mini-white paper on this, then that's a great place to start.
We're not in the territory now where nobody's sure how to proceed to get value.
We're in the territory of people getting value and people falling behind.
Highlight of the week
From the community
Want to watch the recordings of PlatformCon workshops and talks you weren't able to attend? We're uploading more every day on the community YouTube channel.
º