On September 4, 493 Redmond-based Microsoft employees reported to a campus that had just formally eliminated their positions. On the same campus, a dozen open roles in the MAI Superintelligence Team's evaluations engineering function advertised salary bands running to $274,800. Both facts describe Microsoft in the Seattle metro in Q3 2026. They are not in contradiction — but understanding why requires tracking two separate AI talent markets that share a zip code.
The macro number is well-documented. Microsoft's July 2026 restructuring produced 4,800 job cuts globally, with 605 Washington state positions eliminated — 493 of them in Redmond — effective September 4. Chief People Officer Amy Coleman framed the reductions around role realignment, not AI displacement, though the cuts concentrated in commercial and Xbox units that are furthest from Microsoft's core model infrastructure push. Total US headcount has declined from approximately 125,000 to 121,000 over the past year.
The micro number is less widely tracked. The MAI Superintelligence Team — MAIST internally, staffed initially from the March 2024 Inflection acquisition that brought Mustafa Suleyman and senior researchers including Karen Simonyan from DeepMind — is hiring evaluations engineers at a published Redmond band of $142,800 to $274,800. The post-training function, which encompasses data collection, evaluation protocol design, RLHF pipeline development, and finetuning, represents MAIST's operational core for making frontier model capabilities usable inside Microsoft's Copilot surface.
At a glance
Microsoft's MAIST evaluations engineering band: $142,800–$274,800 in Redmond. The $208,800 midpoint exceeds Seattle's Q3 2026 RLHF specialist median of $155,000. Roles are office-first (four days per week on campus, per the January 2026 RTO mandate). The 493 simultaneous Redmond job cuts came from commercial and Xbox units — not the evaluation function. Two parallel labor markets now share the same zip code.
What the MAIST Evaluations Build-Out Looks Like
The MAI Superintelligence Team's evaluations function is not a vendor management operation. The published job descriptions position evaluations engineers as technical contributors working directly on alignment and RLHF — "model architecture, data curation, training and inference infrastructures, evaluation protocols, alignment and reinforcement learning from human feedback" — at the same technical tier as pretraining researchers.
That framing matters for reading the Seattle AI evaluation market. Microsoft is not routing its Copilot evaluation work to Outlier or Mercor at $18–$45/hr contractor rates. It is building a durable, high-seniority internal evaluation org in Redmond, positioned structurally below MAIST's research arm but above the contractor annotation tier that feeds most enterprise AI teams. The comp band reflects that positioning: the $208,800 midpoint of the evaluations engineering band exceeds the $155,000 median full-time RLHF specialist salary tracked by ENTRA's Job Signal Index for the Seattle metro in Q3 2026, and the ceiling at $274,800 approaches principal-engineering territory at Google and OpenAI.
Simonyan, who co-invented VGGNet at Oxford and led applied research at DeepMind before the Inflection move, has described MAIST's current focus as "Humanist Superintelligence" — capable AI systems that remain safety-aligned and anchored to human values. The evaluation function is not incidental to that goal. Measuring whether a model is safe and aligned requires exactly the structured human feedback infrastructure that the broader AI trainer economy is being built to supply. MAIST is building that infrastructure as a core internal competency rather than outsourcing it.
The RTO dimension sharpens the picture. Since January 26, 2026, Microsoft AI employees who live within 50 miles of a designated Microsoft office are required to work from that office at least four days per week. For MAIST evaluations engineers based in or near Redmond, this means office-first work on the Redmond campus — a structural distinction from the remote-first contractor model that characterizes the rest of Seattle's gig evaluation economy. The January mandate is not explicitly a talent filter, but it functions as one: MAIST is selecting for evaluators willing to be in Redmond four days a week, which de-emphasizes the remote-native workforce that platform competitors like Outlier and Scale AI primarily recruit.
Why the Bifurcation Is Accelerating
$142,800 to $274,800 in Redmond. $18 to $45/hr on Outlier's platform. $155,000 median for Seattle metro full-time RLHF specialists. These numbers describe one metropolitan area and three entirely different labor markets operating in parallel — and they are diverging, not converging.
The driver is task stratification. AI evaluation work has split into at least three tiers since 2024. Generalist rating — does this output follow instructions, is it safe — can be completed by credentialed non-experts and pays $15–$30/hr on major contractor platforms. Skilled domain evaluation — is this Python code correct, does this chemistry explanation contain errors — requires field expertise and commands $50–$100/hr. Expert alignment feedback — the work MAIST is hiring for — requires ML research fluency, the ability to identify model failure modes at a mechanistic level, and the judgment to write corrective guidance that a training engineer can act on. That tier has no established gig market. It is an employment market, and it pays accordingly.
Microsoft's July layoffs did not touch the evaluation function. ENTRA's review of the publicly disclosed Washington state WARN Act filing — the 605 positions notified in July — shows concentrations in commercial software sales, Xbox game development, and enterprise operations roles. The functional categories being cut are the ones most susceptible to Copilot-driven productivity consolidation. The functional categories being hired are the ones building Copilot itself — specifically the evaluation infrastructure that determines whether Copilot outputs are trustworthy enough to deploy into enterprise workflows.
The geography compounds the dynamic. The cuts are happening in Redmond but touching roles distributed across the broader Seattle metro's Microsoft satellite offices in Bellevue and Kirkland. The evaluations engineering hires are concentrated on the Redmond main campus, where MAIST sits. Two different physical corridors of the same metro are running opposite trajectories simultaneously — and both trends are attributable to the same underlying force: AI infrastructure investment.
Scale AI has a Seattle-area contracting operation, but it is primarily remote-first and draws on a national pool. Outlier recruits in the metro but similarly does not anchor to Redmond office space. The MAIST evaluations function is the only significant in-office, senior-level evaluation employer in the Seattle metro at this comp level, which makes Redmond the benchmark for what frontier-grade evaluation talent costs in the Pacific Northwest.
Three Things to Watch in Q4 2026
MAIST headcount pace. The team has not published a target size, and Microsoft does not break out MAIST headcount separately in its workforce disclosures. The number of active evaluations engineering postings — currently in the low double digits, based on ENTRA's tracking of microsoft.ai open roles — understates the actual build-out pace, because MAIST has historically filled roles through targeted outreach before posting externally. Watch for Levels.fyi Seattle-area Microsoft AI compensation disclosures in Q4: a clustering of new data points in the $180K–$250K base range would indicate accelerated MAIST hiring.
The RTO effect on contractor-to-FTE conversion. The January 2026 four-day mandate created a novel pressure on Seattle-area AI contractors who work for Microsoft through staffing arrangements: they are increasingly required to be office-present on the same terms as FTEs without the compensation structure that makes that workable. Two staffing attorneys familiar with Microsoft's contractor reclassification process told ENTRA in September that the RTO requirement has accelerated direct-hire conversion discussions for high-performing evaluation contractors — because contractors subject to behavioral office requirements face reclassification risk that neither party wants to litigate. If MAIST converts a significant cohort of contractors to FTE in Q4, it will show up as a net headcount increase against a backdrop of continued broad cuts.
Whether MAIST evaluations rates set a floor for Seattle's contractor market. When a dominant employer in a metro raises the compensation expectation for a specific function, contractor platforms eventually respond. Greater Boston's Mercor dynamic — $94/hr median for senior annotators — established a floor that regional tech employers now compete against. If MAIST's $275K ceiling becomes the widely-known benchmark for Seattle-area evaluation talent, Outlier and Scale AI face pressure to raise their expert-tier rates in the market to retain the credentialed workforce that MAIST is not taking at contractor rates. The rate adjustment, if it comes, will lag MAIST's hiring by two to three quarters.
The headline read on Microsoft in Seattle this fall is a company contracting. The operating read, below the WARN Act filings, is a company doing surgical expansion in the one function it has decided to own — AI evaluation — while outsourcing the attrition to the roles its own models are replacing. The workforce effect in Redmond is a net reduction in headcount and a simultaneous increase in per-head investment. That combination defines what the AI trainer economy looks like inside the largest tech employer in the Pacific Northwest.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.