Google Offers $1.1M, No Degree: Solve Hard Problems Instead

On Friday, October 10, a Google DeepMind research scientist posted one of the most remarkable job ads in Silicon Valley history — not on a careers page, but on X. "We're hiring exceptional engineers to join the team," Tamay Besiroglu wrote. "No degree required. All we care about is you being smart and able to solve hard problems. Total comp is $550k-$1.1M (job ad says something else, but ignore that)."

Read that parenthetical again. A Google employee, recruiting for Google DeepMind, publicly told applicants to ignore the official job ad's numbers because the real ones are better. It is the kind of sentence that only exists when a company is desperate — and it tells you exactly how desperate the talent war over AI coding agents has become.

The team is Polaris, a new DeepMind group building the evaluation frameworks — the exams — that grade how well AI coding agents like Gemini perform on real software engineering. And the pay, the perks, and the absence of every traditional hiring gate together mark a genuine break in how the industry hires. Credentials are dead. The benchmark designers are the new aristocracy.

Photo: Google DeepMind CEO Demis Hassabis delivering a lecture at EPFL, Switzerland, in May 2023. File photo: his lab's new Polaris team is hiring research engineers without a degree requirement, at total compensation of up to $1.1 million. (Alain Herzog / Wikimedia Commons, CC BY-SA 4.0)

Update, Oct. 20: the debate spread — and the equity question got an answer

Ten days after Tamay Besiroglu’s X post, the numbers haven’t changed — and the internet has been arguing about them ever since. TOI/ETHRWorld (Oct 13), NDTV Profit, news18, ET Now and Moneycontrol all relay the same facts: $550,000 to $1.1 million total comp, no degree, no LeetCode, one week per hand-built task. Two things worth adding to the record since we published: (1) the equity question got a clean answer — when an X user asked whether the ad was wrong, Besiroglu replied the ad’s numbers were ‘accurate, but just doesn’t report equity’ (news18, Oct 10) — so the ‘$1.1M’ is equity-heavy, exactly as skeptics guessed; (2) the hiring-war backdrop widened — Apple hired Google’s NotebookLM team and OpenAI accused an Apple hire of taking confidential information, in the same fortnight DeepMind was offering seven figures to people who write exams. No new statements from Google, DeepMind or Besiroglu surfaced in fresh searches this run.

The job ad that broke X's rules of the game

The listing itself is for a Research Engineer on Polaris, based in Mountain View, California. The official ad puts base salary at $164,000 to $252,000 — serious money, but not headline money. Besiroglu's correction is the headline: total compensation runs from $550,000 to $1.1 million, and when a user on X asked whether the ad was wrong, he clarified that the ad's numbers were accurate but left out equity.

That one aside — "(job ad says something else, but ignore that)" — is the most human moment in a corporate recruiting post I have seen all year. Nobody's comms team approved it. It reads like a man who has been given a hiring target he cannot miss and has decided the fastest path is radical honesty. It also, accidentally, confirmed something the industry has whispered for months: DeepMind is paying hedge-fund money for a very specific kind of engineer, and it is not finding them fast enough.

The post travelled. The Times of India, NDTV, Moneycontrol and others picked it up within days, and the coverage split into two camps: awe at the money, and a furious debate about whether a degree still matters at all. The "no degree required" line did the most damage. As one Moneycontrol piece put it, the post "has sparked debate on X over whether problem-solving skills could matter more than academic qualifications in the AI era."

What Polaris actually does all day

Strip away the compensation theatre and the job is genuinely fascinating — and genuinely hard. Polaris engineers hand-build evaluation tasks that measure Gemini's ability to perform long-horizon software engineering work. Each task is a detailed exam: a prompt telling the model what to do, plus a grader or rubric that scores its attempt. One task, from first idea to final approval, takes about a week.

That pace is deliberate, and it is a direct rebuke to the industry's default. Google's own announcement of the role — posted on LinkedIn on September 25, when the team first called for applicants — argues that "as synthetic data pipelines hit their ceiling," the hardest problem in AI engineering is "translating the full depth, nuance, and complexity of real-world software engineering into rigorous evaluation frameworks." Rather than mass-producing tests with synthetic data, Polaris builds a smaller number of tasks that capture how real software actually gets written.

This is the unglamorous engine room of the AI race. Benchmarks decide which model wins the leaderboard; leaderboards decide which model gets deployed; deployment decides revenue. The people who write the exams for the AI that writes the code are, in a very real sense, grading the future. Google clearly understands this — engineers on the team also get autonomy to launch entirely new projects inside DeepMind aimed at improving Gemini's training data and evaluations, plus the freedom to prototype architectures and publish research.

We checked Google's own LinkedIn announcement and five separate relays of Besiroglu's X post; all reproduce the same details, and the LinkedIn post adds texture the news coverage missed — a call for three archetypes, from competitive programmers and math-olympiad medalists to "AI-native builders" who already ship software inside AI coding agents.

No LeetCode, no degree, no minimum experience — and the door opens wider

The official posting lists no bachelor's degree requirement. It lists no minimum years of experience either. What it does ask for is software development experience and hands-on comfort with AI-assisted coding tools, developer agents, and LLM-based workflows. Preferred: competitive programming or maths-olympiad experience — the IMO, IOI, Putnam, USAMO — hackathons, and any background designing rubrics, evaluation frameworks, or benchmark datasets.

Then came the replies, and this is where Besiroglu turned a good story into a great one. Asked whether non-engineers could apply, he said quick learners were fine. Asked about vibe coders — the term of art for people who write code by prompting AI rather than typing it — he said they just need to be very good ones. He confirmed the team does not ask LeetCode questions. And when one user called Google's hiring process "laughably bad," Besiroglu replied that his own team runs the interviews, and that he thinks the process is good.

Consider what just happened here. A DeepMind scientist looked at the most ritualised hiring gauntlet in technology — the degree filter, the years-of-experience ladder, the LeetCode screen — and waived all three, on a public timeline, for a role paying up to $1.1 million. The message to the rest of the industry is not subtle: the credential ladder was a proxy for "can you solve hard problems," and DeepMind has decided to measure the thing itself.

The Mechanize backstory nobody's mentioning

There is a reason Besiroglu, specifically, is the face of this hiring push — and most of the coverage has missed it. He is a co-founder of Epoch AI, the influential AI-forecasting group, and he joined DeepMind as a research scientist after Google completed a talent and licensing deal involving his startup Mechanize, which built reinforcement-learning environments to train and evaluate AI models on real software-engineering tasks. In other words: the exact work Polaris is hiring for.

According to Startup Fortune's reporting, more than a dozen former Mechanize employees moved to Google alongside Besiroglu — most landing on DeepMind's midtraining team — while Mechanize itself keeps operating independently under CEO Guive Assadi. The licensing arrangement is non-exclusive, the same structure Google used for its earlier Windsurf and Character.AI deals. Earlier talks were reported at more than $1.5 billion, though final terms were never disclosed.

So the man telling you to ignore the job ad's numbers spent the last two years building a startup whose entire product was teaching AI to do real software engineering — and Google bought the team to do exactly that at DeepMind. Polaris is not a skunkworks experiment. It is the institutionalisation of the Mechanize playbook, with a million-dollar recruiting budget attached.

Our take: the credential ladder was always a shoddy proxy

Let me take the stance this story is begging for. The four-year degree as a hiring filter was always a crude, class-biased, shoddy proxy for the thing employers actually wanted — proof you can think. Google did not discover this on Friday; it discovered that in a market where AI writes most of the code, the only engineers worth $1.1 million are the ones who can design the tests the AI cannot pass yet. That skill has never lived reliably inside a diploma. It lives in olympiad medals, in open-source repos, in the weirdos who build benchmarks for fun.

Sameer Samat, Google's head of Android, made the same point from inside the building in a recent Business Insider interview: a computer science degree, he said, is "definitely not learning to code. It is the science, in my opinion, of solving problems." When the people running Android and the people running DeepMind's evals agree that the degree is beside the point, the debate is over — only the HR software hasn't been updated.

There is a darker reading, and it deserves one sentence: the same week DeepMind offers $1.1 million to people who can grade AI coders, the industry keeps laying off the junior engineers those coders are replacing. Polaris is hiring the examiners for a machine that is eating the entry-level ladder. The no-degree job is real, and it is colossally well paid — but it is the top of a ladder whose bottom rungs are being sawn off in real time.

For anyone watching from the outside, the actionable bit is simple, and Besiroglu said it himself: be smart, solve hard problems, and get comfortable being graded by people — or machines — who don't care where you went to school. The exam is the credential now. And at DeepMind, the people writing the exam are hiring.

Google DeepMind Polaris FAQ: the $1.1M no-degree job, answered

Does Google really hire without a degree?

The Polaris posting lists no bachelor’s requirement and no minimum experience. Besiroglu put it bluntly: “All we care about is you being smart and able to solve hard problems” (X, Oct 10, via 5+ identical relays).

How much does Polaris actually pay?

$550k–$1.1M total comp. The official ad’s $164k–$252k base figure is “accurate but doesn’t report equity” — Besiroglu’s own clarification (news18, Oct 10).

What would you actually do all day?

Hand-build eval tasks for Gemini (one task ≈ a week), improve training data and evals, prototype architectures, publish research.

Can “vibe coders” apply?

Besiroglu said quick learners are fine and vibe coders “just need to be very good ones” — and there’s no LeetCode interview; his team runs the process. The flip side of skills-based hiring: as IBM built 50,000 AI agents — then kept just 6,000, examiners get hired at the top while the bottom rungs get sawn off.

Sources: Tamay Besiroglu's X post, October 10, 2026 (via Times of India, NDTV, NDTV Profit, Moneycontrol, Startup Fortune); Google's Polaris announcement post on LinkedIn (September 25, 2026); Startup Fortune (Mechanize talent-deal reporting); Business Insider via NewspointApp (Sameer Samat interview).

By TechPulse USA — New York and Sydney desks. TechPulse USA is an independent tech publication covering the AI buildout as it happens: the money, the machines, and the grid underneath them. Related: Britain is stress-testing Google's Gemini 4; Google's Gemini 4 flagship announcement; Gemini 3.8 Live thinks while it talks.