JerkyTreatsLLMs, Agentic AI, unauthenticated MCP servers? We’re gonna need more Platform Engineers.
It’s about two weeks into my serious dive into the current state of AI developer tooling.
So far, I am not afraid of being replaced.
Quite the opposite.
For reference, here’s what I’ve built so far:
This is a very platform eng kind of project.
How can I start building things if I don’t have a platform to build things?
But through this project I’ve formed some very strong opinions on the future of Platform Engineering:
The future of Platform Engineering is AI Infrastructure
Performance, Security, Developer Experience
These are my three pillars for platform engineering.
It’s how I communicate our role is upper management, other teams we work with, and shapes the metrics I measure a platform teams success.
AI presents a massive challenge to Performance & Reliability and Security.
AI presents a massive opportunity to Developer Experience.
The challenges are pretty clear to me.
Vibe coding is like letting the guy from Memento fully developer your apps or features:
Memento is one of Christopher Nolan’s first movies:
It’s 25 years old, which makes me old.
It’s clear that AI tooling can massively increase code velocity.
This presents the clear risk of decreased code quality.
Platform engineers are the most incentivized to manage an orgs AI tooling.
Our pillars inform metrics that will require us to build tools that will protect developers from themselves.
As an example, this weeks hot new topic in AI is the Model-Context-Protocol developed by Anthropic.
It allows AI Agents to make tool calls to do incredibly complex things: Create a database, fetch web content, manage Git repos.
As a platform engineer incentivized to think about security, all I hear is that you’re letting a chatbot with retrograde amnesia make unauthenticated web requests.
It’s not good, folks.
We Must Build Their Tools
Somewhere between full brain-rot vibe coding and DIY joy-of-programming is a sweet spot.
A spot where the velocity increase can be quality controlled.
AI tooling is a net productivity gain, full stop.
A platform teams incentives are to maximize productivity.
Productive devs are happy devs, and I measure a platform teams success on developer happiness.
Which means a platform team is now a AI tooling team.
We need to build platforms that prevent Claude from making wild, unauthenticated API requests.
And just as importantly, we need to build platforms that allow devs to achieve the increased productivity gains that making those API requests would provide.
Platform Agents
To close, imagine the usefulness of semantic infrastructure:
The dev writes “I need a test environment for this new feature”
And the Platform Agent decomposes that into the actual workflows your org uses for environment setup.
Imagine allowing devs to write that right in their IDE.
This isn’t the future, this is now.
This is what I’m building in the image above. Each service in my little platform exposes an API, and the next step is an AI agent that can generate complex workflows using the available routes.
The goal being the ability to casually bootstrap any number of prototype ideas straight from chat.
If I can do that from side projects, imagine how powerful an entire team of devs would be building out these capabilities.
AI is a huge opportunity for platform engineer.
And that means we’re gonna need bigger platform teams.