How to Land a Member of Technical Staff (MTS) Role at an AI Lab
In 2025, Workday’s chief technology officer left the C-suite to take a job titled, simply, “Member of Technical Staff” at Anthropic. No VP. No “Distinguished Engineer.” The same title a brand-new hire might carry. That single move tells you almost everything about how the frontier AI labs work, and why Member of Technical Staff (MTS) has quietly become one of the most coveted titles in technology.
I spend my days coaching senior operators and engineers, and the question I get most often now is some version of: “How do I actually get into one of these labs?” The MTS path is unusual, the comp is eye-watering, and the interview loop is unlike anything at a normal tech company. Job-board content on it is thin, mostly listings and speculation, so this guide lays out what the role really is, what it pays in 2026, what the labs test for, and the realistic roadmap to an offer.
A note up front: there is no shortcut and no secret backdoor. But there is a clear, learnable shape to what these labs want. If you are a strong engineer or researcher willing to do the work, the door is more open than the mystique suggests.
What “Member of Technical Staff” Actually Means
At Anthropic and OpenAI, MTS is the standard designation for technical staff, applied across research and engineering, and largely regardless of seniority or prior title. A first-year engineer and a fifteen-year veteran can hold the same words on their badge. The seniority lives in the compensation band and scope of work, not in the title itself.
In practice, an MTS does the core technical work of the lab: training and evaluating models, building the infrastructure that makes training possible, designing safety and alignment systems, shipping the products built on top of the models, or doing the research that feeds the next generation. The exact work varies enormously, which is the point. One flat title covers a deliberately wide range.
Why AI Labs Use One Flat Title
OpenAI president Greg Brockman has explained the structure publicly: the labs do not want to split researchers and engineers into separate organizational castes, and they do not want titles to track authority rather than contribution. A great engineer should not have to become a manager to gain status, and a researcher’s ideas should win on merit, not on the size of the title attached to them.
The cultural effect is real. When a sitting CTO can take the same title as a new grad, it signals that the work itself, not the org chart, is what the place values. For candidates, that is liberating and slightly disorienting: you are not interviewing for a rung on a ladder, you are interviewing to do the work.
What MTS Roles Pay in 2026
The compensation is the reason this title draws so much attention, and the single flat label hides an enormous range. Because one title spans new grads to seasoned leaders, the bands are wide:
- Anthropic: MTS base salaries run roughly $300,000-$405,000 for typical roles, with publicly filed ranges stretching from about $156,000 to $690,000 across the full spread.
- OpenAI: base bands of roughly $210,000-$530,000, with total compensation reaching well into seven figures at senior levels once equity is included.
- Equity is the multiplier. At senior levels, equity can be 40-70% of total compensation, so two people with the identical “MTS” title can earn dramatically different totals.
The takeaway is not “everyone makes a million dollars.” It is that the title tells you almost nothing about pay, level, scope, and equity do. When you negotiate, you are negotiating a band placement, not a title.
Before you can talk comp, though, you have to clear the bar. Strong fundamentals are non-negotiable, and the fastest way to shore them up is structured study.
The Skills AI Labs Actually Hire For
Lab interviews skew toward depth, not breadth. Whether the role leans research or engineering, a few capabilities show up again and again:
Deep Learning Fundamentals You Can Explain
You will be asked to explain core ideas from first principles, the attention mechanism is a near-universal request, not just name them. Interviewers want to see that you understand why a technique works, what it trades off, and where it breaks.
Applied Math: Linear Algebra, Calculus, Probability
Calculus, linear algebra, and probability underpin everything, and labs test whether you can apply them, not recite formulas. DeepMind’s math assessment in particular goes deeper into theory than any standard tech interview.
Clean, Correct Code Under Real Constraints
Lab engineering loops care about well-designed solutions, high-quality code, performance, and test coverage, not exotic graph puzzles. OpenAI famously cares more that you can implement an LRU cache cleanly than that you can solve a hard LeetCode problem, while Anthropic gates candidates through a CodeSignal assessment with a real passing threshold.
Research Taste and End-to-End Thinking
For research-leaning roles, be ready to discuss your own work and any publications in depth. Portfolio projects should show end-to-end thinking: data wrangling, modeling, evaluation, and (for engineering roles) deployment.
If your machine-learning fundamentals are rusty, that is the highest-leverage place to invest before you apply. Our guides to the best AI agents courses and best LLM development courses are strong places to rebuild and extend them.
The Interview Gauntlet, Lab by Lab
The most common reason strong candidates get rejected is not lack of skill, it is preparing for the wrong interview. Each lab has a recognizable shape:
- OpenAI: barely touches LeetCode-style puzzles. Expect practical coding (clean implementations, sensible tests), ML depth, and discussion of how you think about building and shipping. Judgment and clarity beat cleverness.
- Anthropic: typically starts with a CodeSignal assessment that has a specific passing threshold before you reach a human, followed by ML-depth and values-aligned conversations. Safety motivation is part of the screen, not an afterthought.
- Google DeepMind: the most theory-heavy of the three. The math assessment goes deeper than a typical FAANG loop, and research roles lean hard on publications and first-principles reasoning.
Across all three, two themes repeat: they want breadth and depth, the “why” and “how” behind each technique, and they want evidence you can do real work, not just pass tests. Tailor your prep to the specific lab and role rather than grinding generic interview problems.
Practical prompting and applied-LLM fluency increasingly show up even in research-adjacent loops. It is worth being conversant.
Your Roadmap to an MTS Offer
There is no fixed timeline, it depends on where you start, but the sequence is consistent. Here is how I would approach it.
1. Pick Your Lane: Research or Engineering
MTS spans both, but the prep diverges. Research-leaning candidates invest in math depth, paper fluency, and a publication or strong research project. Engineering-leaning candidates invest in systems, clean code, and shipping real ML infrastructure. Know which loop you are training for before you spend months on the wrong one.
2. Rebuild Fundamentals Until You Can Teach Them
The bar is being able to explain attention, backprop, optimization, and evaluation from first principles, out loud, under questioning. Study until you can teach each idea, not just recognize it. This is the single biggest differentiator in lab loops.
3. Build a Portfolio That Shows End-to-End Judgment
One or two substantial projects beat ten toy notebooks. Show the full arc: a real problem, the data work, the modeling choices and why you made them, an honest evaluation, and (for engineering) a deployed result. Write up what failed and what you learned, labs read for judgment.
4. Prepare Lab-Specific, Then Apply
Match your prep to the target: CodeSignal practice for Anthropic, clean practical coding plus ML depth for OpenAI, theory and math for DeepMind. Then apply broadly, including to the well-funded second tier (Cohere, Mistral, xAI, Sierra, Thinking Machines) where MTS-style roles are also opening up.
How to Stand Out as a Candidate
When the fundamentals are table stakes, differentiation comes from a few things:
- Demonstrated research or engineering taste. A thoughtful blog post, an open-source contribution to an ML library, or a reproduction of a recent paper signals more than a credential.
- Genuine alignment with the lab’s mission. Especially at safety-focused labs, why you want to be there is part of the evaluation. Vague enthusiasm reads as a red flag.
- Evidence you ship. Talk in specifics about something you built end to end, the tradeoffs you made, and how you knew it worked.
If you are transitioning from a strong software or data background, translate that experience into the lab’s language rather than apologizing for not having a PhD, many MTS hires do not have one.
And if any layer of the stack, Python, NumPy, the basics of PyTorch, is still shaky, fix that first. Everything in a lab loop sits on top of fluent, clean coding.
Frequently Asked Questions
Do I need a PhD to become a Member of Technical Staff at an AI lab?
No. While many researchers hold PhDs, a large share of MTS hires, especially on the engineering side, do not. Labs care about demonstrated ability: deep fundamentals you can explain, a strong portfolio, and evidence you can do real work. A PhD helps for pure-research roles but is not a hard requirement for MTS broadly.
Why do AI labs give everyone the same Member of Technical Staff title?
It is a deliberate cultural choice. Leaders like OpenAI’s Greg Brockman have said the labs do not want titles to track authority instead of contribution, or to separate researchers and engineers into castes. Seniority is reflected in compensation band and scope of work, not in the title itself.
How much does a Member of Technical Staff make in 2026?
It varies widely because one title spans new grads to veterans. Anthropic MTS base salaries run roughly $300K-$405K for typical roles (filed ranges reach $156K-$690K), and OpenAI bands run about $210K-$530K. Equity can be 40-70% of total compensation at senior levels, pushing totals into seven figures.
What is the hardest part of the AI lab interview?
For most candidates it is the depth of fundamentals. You must explain core deep-learning ideas like attention from first principles and apply linear algebra, calculus, and probability under questioning. DeepMind leans most theoretical; Anthropic gates with a CodeSignal threshold; OpenAI emphasizes clean practical coding over puzzles.
Should I prepare differently for each AI lab?
Yes. Preparing for the wrong interview is the most common reason strong candidates get rejected. Practice CodeSignal-style assessments for Anthropic, clean practical coding plus ML depth for OpenAI, and deeper math and theory for DeepMind. Tailor your preparation to the specific lab and role.
