The evidence gap
Zero G Talent's board shows 33 salaried roles at Together AI. Zero G Talent's board reported a salary band of $160k–$298k, and its data shows the Strategic Account Executive role topping out at $370k. The open roles: Strategic Account Executive, Director of Tax, Staff Engineer for Distributed Storage and HPC & AI Infrastructure, Director of Data Center Operations, Research Engineer for Large-Scale Training, and Research Engineer for Post-Training Inference. All are San Francisco-based; none are marked remote. No product-manager or engineering-manager titles appear.
That board is the most concrete public artifact of how the company operates. The third-party sources indexed under "Together AI" cover a 2025 body-horror film, not the AI infrastructure company. No public reporting describes sprint cadences, review processes, on-call rotations, or how product decisions flow from research to release. The board confirms they hire senior systems engineers and pay them accordingly. Everything else — autonomy levels, prioritization mechanisms, burnout signals — remains unverified.
Values and operating principles
The clearest public articulation comes from a Bloomberg Tech interview with Together AI's CEO (approximately two years ago). In that interview, the CEO framed open source as a structural bet, not a marketing angle: "Open source A.I. is really history repeating itself. Businesses have been leveraging open source in the form of databases and other core underpinnings of how they develop their software for decades."
Price performance was cited with specific multiples: inference services six times cheaper than OpenAI's API pricing at the time, training services four times cheaper. The CEO described the company as research-driven: "We are research driven company. We invest a lot in, you know, career research to build better models, better model architectures by better data sets and build a better platform."
On the customer problem: enterprises want model ownership and data control without the operational burden of running distributed GPU clusters. The CEO said Together aims to "make that easy and take it off the hands of enterprises so they can realize all those benefits without the operational burden of before."
On talent, the CEO described the founding team as "a world class team... that brings together this unique combination of systems, engineering and research talent that we think is the most potent combination to go and build a market leading company in the space."
No published "operating principles" page or internal motto appears in the public record. The gap between "research driven" and "price performance" (fundamental architecture on quarter-to-year horizons versus continuous kernel-level optimization) is not documented in terms of how prioritization conflicts resolve.
What the hiring bar selects for
The AI talent market is exceptionally tight. Per a CNBC Television interview with Blue Signal Search CEO Matt Walsh: "Oh, it is Uber competitive. I've never seen something like this... there is not enough people to fill all these jobs." Senior machine learning engineers command total packages with Bay Area offers nearly double the national average. OpenAI CEO Sam Altman publicly noted Mark Zuckerberg was offering $100 million signing bonuses to poach talent. Candidates ask for equity, dedicated compute (100 GPUs on signing), publishing rights, remote flexibility, and product influence; they get it.
Together AI's board reveals what the company funds: distributed storage, HPC, large-scale training, post-training inference, data-center operations. Broader market data reinforces the profile: recruiters report the typical candidate needs four to five years of direct experience atop a strong technical foundation. "When you think about it, it's only been mainstream for a couple [of years]. So that pool is small," Walsh noted. A PhD graduate with relevant training work can command out of school.
No public source documents Together AI's internal interview rubric or a founder's on-the-record breakdown of predictive traits. The hiring signal is most legible in the roles it funds. Candidates who thrive are those who have already solved versions of these problems elsewhere.
| Source | Role / Category | Metric | Value |
|---|---|---|---|
| Zero G Talent board | 33 salaried roles (aggregate) | Median | $270k |
| Zero G Talent board | 33 salaried roles (aggregate) | Band | $160k–$298k |
| Zero G Talent board | Strategic Account Executive | Top of band | $370k |
| Blue Signal Search (via CNBC) | Senior ML Engineer | Total package | >$500k |
| Blue Signal Search (via CNBC) | PhD graduate (relevant training) | Starting range | $200k–$300k |
Who thrives and who burns out
The board data — 33 roles, $270k median, concentrated in distributed storage, HPC, large-scale training, and post-training inference — signals a hiring bar centered on deep systems engineering. That profile, combined with the zero-manager structure visible in the board and the CEO's stated bet on open source as default infrastructure, lets us reason about fit from first principles while being explicit about the evidence gap.
Engineers who thrive tend to share three traits visible in the job specs. First, they operate at the intersection of distributed systems and ML workloads: the Staff Engineer role asks for distributed storage and HPC expertise, while the Research Engineer roles split between training-scale and inference-scale problems. Second, they self-direct without product managers translating requirements (none are listed). Third, they treat open-source contribution as core work; the CEO's public positioning means the same engineers writing internal infrastructure also maintain repositories external users depend on.
The burnout profile is the inverse. Engineers who need explicit prioritization, regular sync rituals, or a clear separation between "platform" and "product" work will find the gaps stressful. Compensation at the top of band attracts high-agency talent but raises the bar for what "pulling your weight" looks like day to day. Without a visible process layer documented publicly, the burden of alignment falls on informal communication.
Geography compounds the effect. All listed roles are San Francisco-based with no remote or hybrid designations. For engineers who relocated during the 2021–2022 hiring wave, the return-to-office expectation adds commute friction. Conversely, co-location enables hallway unblocking that substitutes for formal process (if you're in the room).
No public employee sentiment data appears in the research to confirm or refute this inference. The absence is notable: a company of Together's funding tier typically generates hundreds of Glassdoor reviews and active Blind threads. Their scarcity suggests either a very small headcount (consistent with 33 board roles) or aggressive NDAs, or both. Until named current or former employees speak on record — or a systematic survey appears — the thriving/burnout boundary remains a hypothesis derived from hiring signals and structural claims, not from grounded testimony.
The kicker
The commit log is the only spec that matters. When the next open-source model drops, the engineers who stay will be the ones who already know which kernel to rewrite, because they wrote the last one. The ones who leave will have spent their last week waiting for a priority call that never comes. The GPU cluster doesn't wait for alignment. It only runs what you ship.
Working in frontier tech? Zero G Talent tracks the openings: see every open Together AI role, browse frontier tech jobs, the companies hiring, and the people building the field.