Thomson Reuters told its technology staff this week that it's cutting up to 500 engineering roles, about 5.2% of its 9,400-person operations and technology division, and backfilling with more than 250 net-new hires over two years, the large majority senior and, in the company's own phrase, "AI-native." Stock popped about 5%. But sit with the mechanics for a second: the company is eliminating roughly twice as many jobs as it's creating, and the jobs it's creating carry a label it has never publicly defined.
Fine. Companies resize engineering orgs constantly, and doing it around a strategy shift instead of just cost is at least honest. The official line was the usual — focus capacity where customer expectations in legal, tax, and regulatory workflows are moving fastest. I believe that part. It's the label I don't.
"AI-native" is doing enormous work in that sentence, and nobody's told us what it means.
Here's the company's own words, in full:
"We expect to hire more than 250 net-new engineering roles globally over the next two years, the large majority senior and AI-native."
— Thomson Reuters, official company statement (GV Wire)
That's the whole of it — a label, not a definition.
Read the company's stated priorities and a rough shape emerges: engineers who build with AI tools rather than around them, who can supervise and validate machine-generated output instead of just writing their own, who understand where a model's confidence and its accuracy diverge, and who can connect that judgment back to a real product without breaking a legal or tax workflow that a law firm is relying on to be right. That's a coherent skill set. It's nothing like a job description, and nothing like what most technical screens are built to find.
Here's the problem that creates. If Thomson Reuters, or anyone else running this same playbook right now, posts "AI-native senior engineer" and routes it through a standard loop (years-of-experience filters, a LeetCode-style coding round, a system design whiteboard), the screen will reward exactly the wrong signal. Fluency talking about AI is cheap; every candidate has a paragraph ready about how they "leverage AI tools daily." What's scarce is the person who can look at a chunk of model-generated code, spot the subtle hallucination: the API that doesn't exist, the edge case silently dropped, the citation that's almost right — and know it before it ships. That skill doesn't show up in a resume keyword and it doesn't show up in a whiteboard round built for a role that assumed a human wrote every line.
The people who hire engineers for a living are already saying the same thing. Janet Harrah, who runs talent acquisition at DigitalOcean, wrote it plainly after rebuilding her own loop:
"The traditional engineering interview loop (recruiter screen, hiring manager screen, technical phone screen, take-home, onsite) was designed for a different era of software development. It tests for pattern recognition and syntax recall. It stages information rather than creating genuine signal."
— Janet Harrah, Director of Talent Acquisition, DigitalOcean (DigitalOcean blog)
This is the same failure mode I wrote about with TCS's forward-deployed engineer buildout a few days ago, wearing a different label. A company commits to headcount against a title with no established interview, no comp band precedent, and no agreed definition — and does it at a scale (250-plus hires, senior-weighted) where getting the screen wrong doesn't just cost one bad hire, it costs a whole division's worth of them.
The fix isn't complicated, it's just not what's sitting in most ATS templates. Stop screening for "has used AI tools" and start screening for the actual behavior: hand a candidate a piece of AI-generated output with a planted flaw and watch what they do with it. Ask them to design the validation step they'd put between a model's output and a customer-facing legal product, not describe one in the abstract. Weight the interview toward judgment under ambiguity, not toward syntax fluency, because syntax is the part the AI already handles. If your current loop can't tell you which candidate would catch the flaw before it shipped, it can't tell you who's actually AI-native. It can only tell you who's good at saying it.
Thomson Reuters has a two-year hiring runway to figure this out in public. Most companies running this exact trade right now don't have either.
VC5 Consulting builds hiring processes and screens for roles that don't have an established playbook yet — AI-native engineers, forward-deployed engineers, and the other new titles companies are committing headcount to faster than they can define them. If your req reads right but your screen can't tell you who'd actually catch the flaw, let's talk.