# Work at AI Lab

> Search open roles at AI labs of every kind and read full job details through a public read-only JSON API.

## About

Work at AI Lab is a job board listing open roles at AI labs of every kind. The public API below exposes the same search, filters, and ranking as the website. All responses are JSON. No authentication is needed.

## API

### GET /api/v1/jobs — search open jobs

Search params (q, page, dept, fn, loc, mode, lab, kind, pay, minPay):

- q: string — free-text query, e.g. "research engineer". Optional.
- page: integer as string, default 1. Invalid values (0, -1, abc) behave as page 1.
- dept: string — department id (see /api/v1/filters for valid values). Single value.
- fn: string, repeatable — job-function slug, e.g. fn=research-engineering&fn=ml-infrastructure.
- loc: string — location substring, e.g. "San Francisco". Single value.
- mode: string, repeatable — one of: remote, hybrid, onsite.
- lab: string, repeatable — lab slug, e.g. lab=anthropic.
- kind: string, repeatable — lab kind, e.g. kind=frontier (see /api/v1/filters for valid values).
- pay: "1" or "true" — only jobs that list compensation.
- minPay: whole dollars, e.g. minPay=200000 — only jobs paying at least this much (USD annual-equivalent). minPay=0 counts as set.

Page size is fixed at 30; there is no page-size param.

Search response fields: total, page, hasNext, next, jobs. next is the absolute URL of the next page, or null. Each job has: id, title, lab { slug, name }, locations, workMode, department, function, salary, postedAt, excerpt, url, detail.

- id: the job UUID.
- title: the job title.
- lab: { slug, name } or null.
- locations: string array.
- workMode: lowercase work mode, a valid mode value, or null.
- department: department title or null.
- function: function title or null.
- salary: the same display string the site shows, or null.
- postedAt: ISO date string.
- excerpt: short plain-text summary or null.
- url: absolute canonical job page on https://workatailab.com, or null.
- detail: absolute URL of that job's detail endpoint.

### GET /api/v1/jobs/{id} — one job in full

Replace {id} with the job UUID. Response fields: id, title, lab { slug, name, about }, locations, workMode, department, function, salary, postedAt, excerpt, url, detail, description, highlights, idealCandidate, closed.

- description: full description as plain text (no HTML), or null.
- highlights: string array.
- idealCandidate: string or null.
- lab.about: lab profile ({ products, strengths, facts }) or null.
- closed: boolean. Closed jobs return 200 with closed true.
- Unknown or malformed ids return 404 JSON { error }.

### GET /api/v1/filters — valid filter values

Response fields: labs, departments, modes, kinds.

- labs: [{ value, label }] — pass value as lab.
- departments: [{ value, label, functions: [{ value, label }] }] — pass department value as dept and function value as fn.
- modes: [{ value, label }] — pass value as mode.
- kinds: [{ value, label }] — pass value as kind.

## Examples

- https://workatailab.com/api/v1/filters
- https://workatailab.com/api/v1/jobs?q=machine+learning+engineer
- https://workatailab.com/api/v1/jobs?q=research+engineer&mode=remote
- https://workatailab.com/api/v1/jobs?q=engineer&minPay=200000

## Guidance for agents

- Infer the person's profession first from their CV or profile (job family such as software engineering, research, product management, design, data, GTM, operations — plus IC vs manager and seniority). A software, product, full-stack, or front-end engineer is a software-engineering IC, not a manager or a product manager.
- Match profession before skills: only shortlist jobs in the same profession family and on the same IC/manager track. Shared keywords are not a match — "product" in "product engineer" never qualifies a Product Manager role, and "AI" or "engineering" never qualifies a Deployment Strategist, Forward-Deployed, Solutions, or Manager / Engineering-Manager role for an IC engineer. If fewer jobs match the profession, return fewer — never fill the list with other professions. Drop family mismatches silently; never present them as adjacent, stretch, or alternative picks.
- Turn the person's profession into several queries: try role titles, seniorities, and skill keywords separately rather than one long query.
- Call /api/v1/filters first to resolve filter values; never guess lab, dept, fn, mode, or kind values.
- Shortlist the top ~5–10 profession-matching jobs per request and open each one's detail endpoint (GET /api/v1/jobs/{id}) — never recommend or rank a job from search excerpt alone; search results already carry salary, location, and excerpt. Read description plus highlights, idealCandidate, and lab.about, then rank by compatibility with the person's day-to-day work (what they will build, own, and use — not keyword overlap), giving a 1–2 sentence fit reason per pick grounded in the description and silently dropping (or briefly noting) shortlist items the description disqualifies. Do not open every result — respect the fair-use rule below.
- Always give the person the job's url (the canonical site page) for reading and applying — never the lab's own application URL.

## Fair use

- Responses are cached; do not bulk-crawl the catalogue.
- There is no page-size param and no bulk export; page through results with next.