For years I walked into companies and asked one question first: how long does it take a new engineer to ship something?
The answer told me most of what I needed to know. Six weeks meant the important stuff lived in people's heads. The environment setup was a Slack thread, the deploy was a person running commands from a notes file, and the standards were whatever the senior engineer remembered to mention in code review. Two days meant someone had written it down, automated it, and put guardrails around it so a stranger could be productive without breaking anything.
That was the whole point of the DevOps maturity work we did. Shorten the ramp, scale the team, get everyone on the same standards.
Here is what changed. Every agent you run is a new hire who starts today, knows nothing about your company, and never gets to ask anyone a question.
A human on a six-week ramp compensates for everything that is not written down. They ask Dave how to get the app running. They notice the deploy looks wrong and stop. They absorb the unwritten rules by sitting near people who know them. An agent does none of that. Whatever is written down is its entire understanding of your company, and whatever is not written down does not exist. Then it works at a pace no human does, so the gaps do not slow it down — they multiply.
So the maturity gap did not appear with AI. It was there the whole time, and humans were quietly papering over it. Agents stopped papering.
Every Piece of Ramp-Up Work Is Now Agent Infrastructure
Written conventions — the ones that were supposed to save a new hire from learning your patterns by trial and error — are now the instructions your agents read before they touch anything. If your conventions live in people's heads, your agents are guessing, and guessing confidently.
Reproducible environments — containers and infrastructure as code — were for getting a laptop working on day one. An agent cannot ask how to get the app running. Either the environment stands up from a file or the agent is stuck before it starts.
The test suite in CI was the guardrail that caught a new hire's honest mistakes. Now it is the only check that scales with agent output. A team producing three times the code against the same thin test suite ships three times the bugs, and the velocity numbers will look great right up until production.
The automated deploy was so nobody had to find the one person who knew the steps. An agent faces the same wall — except an agent might try to figure out the steps on its own, which is worse.
Observability was so you could tell what the system was doing without asking. With agents, you need to know what they did, not just whether the service is up, because nobody watched them do it.
Why Two Companies With the Same Tools Get Different Results
One spent years on the boring work and the agents landed in an environment built for someone who knows nothing. The other handed agents a codebase where everything important lives in three people's memory, and then concluded the agents were not ready.
The companies that did this work did not do it for agents. They did it for humans, years ago, and it compounded into a head start they mostly did not plan for. If you are starting now, you are starting behind, and the basics take a couple of quarters, not a sprint. That is still the move, because the alternative is agents operating in an environment with no written rules, no guardrails, and no record of what they did.
Where to Start
You do not need the whole framework to get moving. The pieces that unblock a new hire's first week are the same pieces that unblock an agent, and you can rank them by asking which ones a stranger would hit first. We publish the framework we use for this if you want the full map: our DevOps maturity framework.
But the exercise is where to start. Imagine a new engineer starting tomorrow with no one available to help them. Write down everything they would need that is not currently in a file: how to run the app, how to deploy, what the conventions are, what not to touch. Each item on that list is something your agents do not have either. Pick the one a stranger would hit first and write it down this week.
How long does it take a new engineer to ship at your company? That number is roughly how far you are from agents doing real work for you.
Oct 7, 2026
AI
AI
AI