Google DeepMind has listed a new engineering role with a pay band most people will never see in a job advertisement: total compensation between $550,000 and $1.1 million. More striking than the number is what the listing does not ask for. There is no degree requirement, and no minimum years of experience either.
The listing, live on Google's official careers site, is for Polaris, a new DeepMind team that builds evaluation frameworks for AI coding agents. The post went viral this weekend after being shared by Tamay Besiroglu, a research scientist at DeepMind, and has reignited debate about how the world's best-funded AI labs decide who is worth hiring. For more context on this story, see our ongoing artificial intelligence updates.
What the Polaris Job Actually Is
Polaris occupies an unusual corner of DeepMind. Rather than training models, the team designs the tests that decide whether the models are any good. According to the listing, the role involves building evaluation frameworks for AI coding agents, prototyping new evaluation architectures, and publishing research along the way.
It is a meta-job of sorts — evaluating the evaluators — and an increasingly consequential one. If a benchmark quietly misleads a lab, every downstream decision built on that benchmark inherits the error. Polaris sits at the choke point that determines how quickly DeepMind can ship better coding agents without a broken test lying to the whole team.
The compensation structure is unusually transparent for a role of this seniority. The listing shows a base salary range of roughly $147,000 to $210,000 plus bonus, with total compensation between $550,000 and $1.1 million once benefits and equity are included, according to reporting on the listing by Startup Fortune. That top end works out to roughly 10.9 crore rupees, which explains the eye-catching "Rs 11 crore job" headlines that spread across Indian media within hours of the post going viral.
The Mechanize Connection
The people recruiting for Polaris did not arrive at DeepMind through the conventional academic pipeline. Besiroglu joined after Google completed a talent-and-licensing deal involving his startup, Mechanize, which built reinforcement learning environments for training and evaluating AI models on real software engineering tasks — precisely the kind of work Polaris is hiring for.
Talks over that deal were reported at more than $1.5 billion, though the final terms were never disclosed. More than a dozen former Mechanize employees moved to Google alongside Besiroglu, according to Business Insider. The structure echoes Google's $2.4 billion Windsurf arrangement, which was likewise organized as a technology license plus talent hire rather than an outright acquisition.
That context matters because it explains what DeepMind is really shopping for. Polaris is not a generalist engineering job with a famous logo attached. It is a targeted hire for a specific, provable skill — building evaluation harnesses for coding agents — where a portfolio of real work answers the question a transcript cannot.
No Degree, No Experience Floor
Frontier labs dropping formal degree requirements is not new. What is new is doing it this loudly, on a role attached to a number big enough to make outsiders look twice. The listing states no education requirement and no minimum experience threshold; the bar is built entirely around demonstrated ability.
Besiroglu framed the team around a simple pitch: problem-solving ability over paper credentials. For candidates without computer science degrees, the posting is a rare explicit invitation from one of the world's most selective research organizations.
The Numbers Behind the Talent War
Polaris's ceiling looks almost modest against the wider market. Business Insider reported from H-1B filings that Anthropic listed base pay as high as $1.38 million for technical roles in fiscal 2026 — cash alone, before equity or bonus. Meta offered researcher Ruoming Pang a package reported at more than $200 million to leave Apple. Business Insider has also reported that some DeepMind engineers concede Anthropic's Claude Code outperforms their own coding models, which is precisely why outside evaluation tooling has become strategically important.
Against packages like those, the signal from Google is less about the dollar figure and more about which gate the lab is willing to drop to win: not the compensation ceiling, but who is even allowed to apply.
What It Signals for AI Hiring
For the broader technology labor market, the listing is another data point in a slow unbundling of the degree-as-proxy system. When a hiring process can test the actual ability directly — as an evaluation-engineering role can — the credential becomes the cheapest thing to cut.
Whether that philosophy spreads beyond elite AI labs is an open question. Most employers cannot run million-dollar hiring processes to find non-credentialed talent. But for the specialized, high-leverage roles at the center of the AI race, the transcript is losing ground to the portfolio, and DeepMind's newest job ad is the loudest evidence yet.
A Live Test of Skills-First Hiring
The timing is not accidental. The listing appeared while Google is under pressure on the coding front, with internal assessments reportedly showing rivals ahead. Evaluation quality is now a competitive weapon: labs increasingly differentiate not just on raw model capability but on how honestly they can measure it. A team that builds trustworthy evaluations can iterate faster and market results that hold up to outside scrutiny.
For job seekers, the listing is also a template worth reading closely. It asks for evidence — open-source work, published evaluations, demonstrated problem-solving — rather than a specific line on a diploma. That is a meaningful shift in how a top-three AI lab defines qualification, and other employers watching the talent war will notice which gates Google decided it could drop without losing quality.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
