Frontier Work Observatory4 sources verified
PUBLIC HIRING INTELLIGENCE29 JUL 2026 · 18:31 PDT

The frontier is
hiring in public.

A source-grounded observatory tracking how frontier organizations turn research priorities into teams—without confusing a fresh posting, a repost, and a newly observed role.

Explore the signal Inspect the method
3,413live roles
198
2,063
403
749
Snapshot integrity: source-linked
00 / PULSE

985 roles carry source-published dates inside the 30-day window.

198+37 / 30d2,063+615 / 30d403+106 / 30d749+227 / 30d
01 / SIGNALS

Where the organizations converge—and where they do not.

Hiring is an organizational error signal: teams add capacity where capability, reliability, or execution still falls short. Read together, the roles form a map of unresolved constraints.

Convergence: agents, evaluations, RL systems, safety infrastructure, and compute. Divergence: SpaceXAI emphasizes model velocity; Anthropic, empirical safety and research tooling; OpenAI, agents and product-grounded post-training; SpaceX, operational AI and physical compute.

02 / OPPORTUNITY LANES

Choose the proof you want to be hired for.

These are not opaque recommendations. Each lane connects public demand to the concrete artifact a strong candidate can demonstrate.

THESIS

Capability is becoming an environment-design problem.

EVIDENCE ON THE HIRING SURFACE

Teams are hiring for graders, harnesses, computer use, long-horizon tasks, and the systems that turn fuzzy behavior into measurable signal.

PORTFOLIO PROOF

Ship a benchmark with baselines, variance analysis, adversarial cases, a failure taxonomy, and a dashboard that makes regressions hard to miss.

4 source-linked roles in this view

Software Engineer — Evals

Palo Alto, CA

MARKET SIGNALOwn instruments for model capabilities and behavior, including datasets, grading schemes, and failure diagnosis.

PROOF TO SHOWA trusted eval product that converts taste, quality, truthfulness, and safety into concrete criteria.

Staff Software Engineer — Environments Infrastructure

San Francisco / New York

MARKET SIGNALInfrastructure for the environments in which agents learn and are evaluated.

PROOF TO SHOWA reproducible task environment with isolation, replay, observability, and deterministic scoring.

Research Engineer — Computer Use

SF / NYC / Seattle

MARKET SIGNALResearch and engineering for agents that reliably operate software interfaces.

PROOF TO SHOWA computer-use benchmark with task completion, intervention rate, recovery, and unsafe-action metrics.

Research Engineer / Scientist — Personal AGI Model Experience

San Francisco, CA

MARKET SIGNALPost-training and model behavior for a more useful, coherent personal intelligence.

PROOF TO SHOWLong-horizon behavioral experiments with rubrics, preference data, and product-grounded outcome measures.

03 / RECONCILIATION

One brand is not always one hiring system.

SpaceXAI branding and the SpaceX careers surface overlap in public language, but their ATS boards remain distinct. The model preserves that truth instead of forcing a convenient merge.

PUBLIC BRANDSpaceXAISpaceX
SOURCE SYSTEM Greenhouse / xAI Greenhouse / SpaceX
OBSERVATORY ENTITY spacexai · retained spacex · retained
01first_published

What the ATS says

02first_seen

What this system observed

03content_hash

What materially changed

04closed_at

What disappeared and when

04 / METHOD

Credibility is a product feature.

A snapshot is evidence, not history. This first release marks its temporal boundary plainly and separates measured facts from interpretation.

Source identity4 / 4

Every organization retains its own ATS and canonical job ID.

Claim traceability100%

Every curated role and count links to a primary public source.

Historical certaintyDay 0

Closures and true first-seen events require future daily captures.

Fit-score opacity0

No unexplained match percentage is presented as fact.

01

Capture

Query official public ATS feeds and record the timestamp, source identity, and provider fields without silently rewriting them.

02

Reconcile

Normalize titles and locations while retaining canonical job IDs, brand relationships, and near-duplicate groups.

03

Evaluate

Test extraction, event classification, duplicate precision, link health, and the completeness of every evidence chain.

04

Interpret

Derive organizational signals with explicit uncertainty. Never replace eligibility checks or candidate evidence with a mystery score.

05 / APPLICATION VALUE

Build proof, not claims.

xAI asks candidates to describe exceptional work. Anthropic asks to see an LLM project with complex behavior or quantitative experiments. Frontier eval roles ask for the ability to move from a fuzzy problem to a reliable measurement system.

The observatory is strongest when it becomes that evidence: reproducible ingestion, temporal reasoning, evaluated reconciliation, calibrated uncertainty, and a product people can actually use.

Audit the source layer