Two stories broke about engineering jobs within a day of each other last week. Read only one, and you’d walk away with a completely different picture of what AI is doing to this profession.
Tata Consultancy Services announced it’s building a team of up to 8,900 forward-deployed engineers — people who sit inside client companies, learn their specific mess of systems, and figure out how to get AI working in production instead of just in a demo. TCS’s CEO was refreshingly honest about one detail: he hasn’t decided whether these will be new hires or existing employees retrained into the role.
The same week, Thomson Reuters said it’s cutting up to 500 engineering roles. Buried under that headline: they’re also planning to hire more than 250 new engineers over the next two years, and by their own account, most of those will be senior and “AI-native.”
Read separately, these look like opposite stories — one company growing engineering, one shrinking it. Pick your narrative.
Read together, they’re saying the same thing: the job hasn’t disappeared. It’s moved. Away from writing code, toward understanding context, exercising judgment, and owning outcomes when things don’t go the way the demo promised.
That part isn’t new. People have been saying “AI changes what engineers do” for two years now. Here’s the part I think most people are missing.
PwC’s 2026 AI Jobs Barometer had a number in it that stopped me mid-scroll. Entry-level roles most exposed to AI are now seven times more likely to require the kind of skills you’d normally associate with someone senior — judgment, leadership, the ability to read a room and make a call. Those “seniorised” junior roles have grown 35% since 2019. Every other kind of entry-level role has shrunk by 10%.
Companies aren’t just asking more of junior people. They’re asking junior people to already have the thing that junior people are, by definition, still building.
That’s the actual contradiction hiding under both of those stories. Everyone wants senior, AI-native engineers. Almost nobody is asking where those people are supposed to come from.
Because seniority isn’t a credential you pick up from a course or a new tool. It’s a residue. It builds up slowly, from years of making smaller decisions, watching some of them fail, getting told why, and doing it again slightly wiser. Judgment is what’s left over after you’ve been wrong enough times, in a low-enough-stakes environment, to actually learn something.
That low-stakes environment used to have a name. We called it being junior.
If AI is absorbing the small, repetitive, entry-level work — the tickets, the boilerplate, the first-draft code — and companies respond by shrinking junior headcount while raising the bar for who gets hired at all, we’re not just changing how people enter the profession. We’re removing the ground they used to learn to stand on.
I want to be careful here, because it would be easy to turn this into “companies are villains, protect every junior role.” That’s not the argument, and it isn’t even true. Roles have always evolved with technology. Nobody’s asking for the COBOL punch-card operator back. Change isn’t the problem.
The problem is narrower than that: you can’t keep raising the experience bar while bulldozing the place where experience gets built. Those two things can’t both be true for long — either the pipeline of senior talent runs dry, or “senior” starts meaning something thinner than it used to.
There’s a fair counterpoint here, and it deserves a real answer. Maybe AI is the new apprenticeship. A junior engineer pairing with a model that catches mistakes in real time, that explains the “why” behind a fix, might genuinely compress years of learning into months. I don’t think that’s crazy. But it’s unproven, and it’s a very different thing from letting the model write the code while the junior just watches. One builds judgment through fast feedback. The other removes the junior from the loop entirely. Which future you get is probably the most consequential design decision engineering leaders will make this year — and most aren’t making it on purpose.
So what does holding both of these truths look like in practice? Two things, and neither is complicated.
Retrain the engineers you already have, instead of treating “AI-native” as a label you can only buy from outside. TCS didn’t have an answer to whether their forward-deployed roles are new hires or retrained staff. That’s not a gap in their strategy. That’s the actual question every engineering leader should be sitting with right now.
Redesign what “junior” means, instead of deleting it. Not junior-as-code-writer, competing with a model that’s faster and never tired. Junior-as-supervised-owner — someone learning to read a customer’s real problem, validate an AI’s output before it ships, exercise judgment on a smaller, safer scale, with someone senior close enough to catch it when they get it wrong. That’s not a downgrade of the role. It might be a harder, more interesting version of it.
AI can shorten the distance between an idea and working code. That part is basically settled.
What it can’t do is shorten the distance between inexperience and judgment. That still takes time, and it still takes someone letting you make a smaller mistake before the stakes get bigger.
We can’t keep asking for experienced, AI-native engineers while deleting the apprenticeship that makes them. Pick one. The other is a fantasy.
Sources: Reuters on TCS’s forward-deployed engineer buildout and Thomson Reuters’ engineering restructuring; PwC’s 2026 Global AI Jobs Barometer.
