# AI-Native JD Benchmark — ENGINEERING (Knowledge Base for the Gem)

**Purpose:** Ground the JD AI-Native Check Gem on *real* engineering JDs from frontier AI
companies, so its gap analysis and benchmark refs quote actual phrasing.

**Source:** Public ATS scrape (Greenhouse, Lever, Ashby) of 34 AI-native companies, May 29
2026. This file covers **1,421 engineering + technical-research roles** (SWE + AI/ML research).
Top hirers: OpenAI (365), Anthropic (192), xAI (86), Scale AI (76), Cohere (72), Harvey (71),
Mistral (70), Perplexity (41), Sierra (41), Glean (39).

> **How the Gem should use this:** When the JD is an engineering/research role, benchmark it
> against the patterns and quotes below. In BENCHMARK REFS, lift or adapt real phrasings and
> cite the company. Flag where the JD lacks the markers in "What makes it AI-native."

---

## Shared spine — what makes ANY JD AI-native

- **Outcome-framed**, not task lists — own outcomes end-to-end, define success metrics.
- **AI as leverage the person wields** — expected to build with / automate using AI, not fear it.
- **AI fluency as baseline**, stated concretely (agents, evals, model behavior), not "nice to have."
- **Comfort with ambiguity / 0-to-1 / "build role"** — invent the playbook, not execute one.
- **Broader scope, smaller team** — founding, high-autonomy, end-to-end ownership.
- **Judgment, taste, ownership** over years-of-tool-X.
- **Velocity** — ship fast, deployment velocity, reduce toil.

A JD missing 3+ of these lags the AI-native standard.

---

## Engineering-specific markers (what's different here)

- **AI in the loop changes the metrics** — e.g. owning productivity metrics (DORA/SPACE)
  *"when AI agents are in the loop."*
- **Builds agentic systems**, not just services — shipping agents, evals, human-in-the-loop.
- **Frontier context as a draw** — working at the edge of model capability, open-source research.
- **Infra framed as automation** — make toil visible, cap it, automate provisioning.
- **Forward-Deployed Engineer** is a new archetype — owns production outcomes inside customer accounts.

## Top engineering archetypes (2026)

Software Engineer · Member of Technical Staff · Staff Software Engineer · Engineering Manager ·
Research Engineer · Data Scientist · **Forward Deployed Engineer** · Researcher · Security
Engineer · Applied AI Architect · Applied AI Engineer.

---

## Real phrasings to lift or adapt (quoted from actual JDs)

- *"Experience defining engineering productivity metrics (DORA, SPACE…)—and a point of view on how these metrics evolve when AI agents are in the loop."* — Anthropic, PM Developer Productivity *(applies to eng leadership too)*
- *"This role is high-autonomy: you'll define the architecture, write the integrations, and represent the platform."* — Anthropic, IT Systems Engineer
- *"A strong disposition to thrive in ambiguity, taking initiative to create clarity and forward progress."* — Anthropic, Analytics Data Engineer
- *"You'll ship agentic capabilities on an open, extensible stack, with the craft and care required for enterprise trust."* — Glean, Application Security Engineer
- *"Drive the team's shift from manual operations to systemic, automated, scalable infrastructure — making toil visible, capping it, and prioritizing work that reduces it."* — Together AI, EM Site Reliability
- *"Deep intuition for statistics, neural networks, and how data quality influences training outcomes."* — xAI, Data Engineer
- *"Embed robust security controls into our CI/CD pipelines… without compromising deployment velocity."* — Mistral, CyberSecurity Engineer
- *"Be a founding member of our team… shape the future of AI in the public sector."* — Scale AI, DevOps Engineer

---

## Comp anchors (USD total, frontier-lab engineering)

Observed band across these roles: **~$104k to ~$850k** (855 figures). Senior/staff IC and
research-adjacent eng cluster high — Anthropic examples: Software Engineer ranges from
~$275k up to ~$485k+ for model-performance roles. Use as a reality check: AI-native eng roles
are scoped with high autonomy and priced at a premium vs. generic SWE postings.

---

## Honest caveat

Frontier-lab eng comp and scope skew US/SF–NYC. Apply the "everything has changed" framing
carefully to non-US or non-frontier roles, which still read more conventionally.