Where are you on the agentic ladder?
2026-05-15
Where are you on the agentic ladder?
Hey there! Welcome to Platform Weekly. Your weekly knead of platform engineering dough. Every week, we dive into another corner of the platform engineering universe, from industry insights and community news to practical lessons and live demos.
Plus… our friends at Tailscale are organising their own awesome conference - TailscaleUp! They’ll be in San Francisco on August 26, 2026 and are giving the community discounts. You can save $100 off TailscaleUp registration now though May 31st and use PE10 for an additional 10% off your ticket.
Where are you on the agentic ladder?
I've been heads-down for months with my AI & platform co-conspirators Kaspar, Mallory Haigh, and Ajay Chankramath on a piece of research I'm honestly buzzing to share.
The four levels of agentic software development in the enterprise is our attempt to give engineering leaders a real map of where they are, where they're going, and what the platform underneath them needs to look like to actually get there. Mamma mia, surprise surprise! The conversations we're having every single day all point to the same thing. It's not the models holding teams back. It’s not the tools (shocker). It's the platform.
As your AI capability matures, and expands, your platform needs to mature and expand with it. If your AI usage is outpacing your platform. You’re fu- in trouble. We’ve known this for a while now. Last year we published data showing that 95% of platform teams were failing with AI, and the 5% were using platforms. Exciting stuff at the time. But that exciting stuff at the time is old news now. Now we need to go deep on exactly what that means for the highest performing organizations today, and what can be done for you to replicate.
Here is the ladder of maturity that we’re talking about. Each rung is defined by how the human role evolves as agents take over more of the value stream:
Level 1: Human in the loop. Agents suggest & humans still approve every line. Gains are real but linear. Review bandwidth is the cap.
Level 2: Human on the loop. Agents become parallel participants. Humans dispatch the work, and validation becomes an industrial feedback loop rather than a single gate.
Level 3: Humans as orchestrators. The platform runs as a background execution layer. Agents respond to system signals on their own. Review is exception-based.
Level 4: Fully autonomous (outlook). The production system becomes partially self-adjusting. Agents initiate, implement, validate, and promote within guardrails you've defined.
It’s the rungs of this ladder that drive the incredible ROI that teams are seeing with AI. And it’s only possible to successfully climb this ladder if you’ve got the platform maturity to support it.
And here's the further punchline. None of these rungs are obtainable without what we're calling an Agentic Development Platform (ADP). Not that the Internal Developer Platform that won the human-developer era is disappearing, but it is evolving. Probabilistic systems (models, agents) need a deterministic fabric (policies, ephemeral envs, RBAC, observability) wrapped around them, or the whole thing falls over. That's why most failed AI initiatives we analyzed fell apart.
There is far too much in these 40 pages to summarise here. So read it, find your rung, and tell me what's pinning you there. If you read one whitepaper this month, make it this one.
Quick bites
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.