This new job title needs an apprenticeship program, yesterday.
Call the role whatever you want. “Orchestrator” may not be the title we end up with, and the language will change. The job is already here: someone who directs agents, tools, and generated code into a system that has to work in production, and who can still dig into the details when it doesn’t.
Yesterday, I introduced the emerging role of the “Orchestrator.” One response struck a nerve: we need to build its career path now, because junior developers — the future experts who would fill it — are already disappearing.
The short-term hiring logic is hard to argue with. Why bring on a junior developer when one senior developer with AI can clear the backlog before lunch? That question sounds like efficiency. It is also how you dig a crater under your own industry when your senior developers start to realize that retirement exists.
Press enter or click to view image in full size
Generative AI has automated a lot of the work that used to train people. Recent graduates are discovering that those first jobs literally don’t exist anymore.
What feels like a productivity gain today will turn into a talent crater in about a decade.
Stop hiring juniors for a decade, and you don’t get leaner engineering. You get an automated industry full of systems nobody under forty understands, staffed by a shrinking group of people who learned it the old way. Let’s call this the Talent Crater of 2036. It will take five to ten years to become obvious, but if you squint and look far into the future, you can see it slowly forming.
Automation has already done this to manufacturing
Factories automated routine production, apprenticeship programs thinned out, and companies counted the labor savings. Then they couldn’t find anyone who could diagnose the machine when it started making a noise nobody recognized.
Automation did not eliminate the need for skilled workers. It eliminated the work that made them. That shortage is biting countries like the U.S. now as they try to rebuild domestic manufacturing and discover that the pipeline of people who can run the plants has dried up.
Germany kept running apprenticeships while the factories were filled with robots. When automation removed the entry-level work, they replaced the training path instead of pretending they wouldn’t need skilled people later.
You didn’t learn the stack from a course. You learned it by breaking things.
I learned how to build and operate websites from decades of failure. Mistakenly running rm -rf * at the root of one of Jim Cramer's web servers in 1999. Being blissfully unaware that Netscalers could actually melt two decades later.
Those mistakes, and a thousand smaller ones, are how you learn a system end-to-end. Small bugs, flaky tests, tracing a 500 through four services, sitting through a deploy that goes sideways. That work shipped software, and it also produced the next round of senior engineers.
Not only is the industry not giving people opportunities to fail in 2026, but many students are switching out of CS degrees because the jobs aren’t there.
AI can do most of that junior work now. If you hand all of it over and don’t replace the training that came with it, the training disappears. For a few years, that will look like efficiency and productivity. Ten years out, it looks like a lot of software running with nobody able to understand it if the AI encounters a problem.
News from 2036: The automated power grid is having problems rebalancing workloads during a Winter storm, and the power company is struggling to find experts to debug the software. Comment from CEO: “AI isn’t always perfect, sorry.”
Keep hiring junior developers anyway
Today’s orchestrators can do the job because they spent years doing the work AI is now taking over. That experience is the qualification, and there’s no way to hire it if nobody is producing it.
So while your balance sheet might be allergic to it, keep hiring junior engineers, even if the latest frontier model can replace them. You’re going to need them in a decade. Give them real systems. Let them inspect code, chase failures, sit in on incidents, and take on responsibility under people who have already broken and fixed a few things. Sometimes that means handing over work a model could finish faster. That isn’t waste. The task is how they learn.
If your staffing plan depends entirely on people who already know how the whole machine fits together, you’re spending down a supply that nobody is replacing.
Build the apprenticeship into the redesign
As you begin redesigning your organization and teams to accommodate orchestrators, build an apprenticeship program early. Pair entry-level programmers with senior orchestrators. Develop a process that exposes them to the operations and implementation details they need to support the systems being automated: DNS, deployment, database, load balancer, and the 2 a.m. incident.
Even in a world where everything is automated and magically agentic, NIST’s AI Risk Management Framework still calls for the ability to supersede, disengage, or deactivate AI systems and keep operating when they fail.
Imagine the worst case: the models you use to automate operations disappear due to a security threat or a power grid failure. One objective of an Orchestrator Apprentice program should be to make entry-level developers responsible for understanding how to run the manual overrides.
By the way, no matter how much you automate, you should be working on a manual override backup plan.
The other reason to do this right now? Our industry is facing a bit of a crisis, and it’s time to invest in people, and recent grads are struggling.