The eight FDE archetypes, FDE Hiring Playbook

AI lab / foundation model companies

e.g. OpenAI, Anthropic, Google DeepMind, Mistral, Cohere, xAI

Job to be done: Turn raw model APIs into production deployments at Fortune 500 / government accounts. Own discovery, eval design, system build, rollout. Increasingly: build agentic systems.
Tech depth: Full SWE + LLM application expertise. Python, TypeScript, prompt engineering, agent development, evals.
Travel intensity: Hybrid 3 days/week + 25–50% travel.
Day-to-day mix: 40–50% coding, 20–30% customer discovery, 15–20% internal product feedback.

Concrete example

OpenAI + Morgan Stanley: 98% advisor adoption.
OpenAI + Klarna: 400+ policies parameterized.
Mistral + Ericsson: Codestral fine-tune for legacy code translation.

Data lab / RLHF / talent platform companies

e.g. Scale AI, Surge AI, Mercor, Turing, Invisible

Job to be done: Build data pipelines, RL environments, and eval harnesses for AI labs as customers. Less ‘enterprise deployment,’ more ‘embedded with research customers.’
Tech depth: ML researchers and ML platform engineers at AI labs (less C-suite than segment A).
Travel intensity: Mostly hybrid SF/NY. Customer base is concentrated, so less travel.
Day-to-day mix: ~60% coding (data pipelines, eval infra), ~20% customer collaboration, ~20% internal platform.

Concrete example

Note on market shift:Meta's mid-2025 ~$14.8B partial acquisition of Scale (Alexandr Wang to Meta) put many Scale alumni into circulation, they are an active sourcing pool through 2026.

Horizontal application platforms

e.g. Glean, Cursor, Cognition

Job to be done: Build the ‘last mile’ of the deployment into the customer environment, bespoke connectors, agents, workflows. Glean’s ‘Founding FDE’ posting is unusually founder-flavored: ‘Your primary job is 0-to-1 product creation … What you build becomes what Glean ships.’
Tech depth: Heavy full-stack. Cursor: ‘Strong in Python and TypeScript/JavaScript … handled production reliability before (metrics, alerts, incident response).’ 5+ years typical.
Travel intensity: Mostly remote/hybrid. Cursor explicitly remote-friendly.
Day-to-day mix: Heaviest coding share of any segment. Less travel; deep customer integration work.

Vertical agent companies

e.g. Sierra, Decagon, Harvey, Hebbia

Job to be done: Configure / build agents for specific customer workflows on the customer’s platform. Less ‘novel system design,’ more ‘agent design + integration + evals.’ Sierra’s deliverable is a working agent with guardrails on Agent OS, more configuration than custom-app builder.
Tech depth: Full-stack. Hebbia: 5+ years at venture-backed startup or top tech firm; APIs, data pipelines, frontend when needed.
Travel intensity: High in-office. Hebbia: 5 days a week in NYC/SF, embedded onsite with major buyside customers. Sierra: in-person SF.
Day-to-day mix: ~40% coding, 30–40% customer scoping/agent design, 20% feedback to product. Heavier on AOPs (agent operating procedures), agent skills, and guardrail design than novel infrastructure building.

Enterprise SaaS / B2B horizontal adopters

e.g. Ramp, Salesforce, Rippling, Sourcegraph

Job to be done: Make complex enterprise SaaS work for biggest customers; integration depth into their data/systems; close enterprise deals.
Tech depth: Full SWE, with rising LLM literacy. Less RLHF/fine-tune than AI labs.
Travel intensity: Ramp runs ~16 FDEs in pods, 7 of whom are former founders (Leo Mehr, Ramp Builders, August 5, 2025). Salesforce launched Agentforce FDE program April 2025: pods of 1 deployment strategist + 2 FDEs, full-time on one client for ~3 months (‘Momentum’ program). Sourcegraph requires 80% travel Mon–Thu for onsite customer engagements.
Day-to-day mix: ~50% travel including travel / on-site requirements.

Vertical AI in regulated industries

e.g. Harvey (legal), Hebbia (finance), OpenAI Life Sciences/Federal, Anthropic Federal Civilian, Commure (healthcare), Chestnut (insurance)

Job to be done: Regulatory and compliance constraints define the role. OpenAI Life Sciences FDE: ‘Define launch criteria for regulated contexts, including validation evidence, outcome metrics, and acceptance thresholds tied to production use.’ Enforce operating standards for auditability, traceability, and inspection-readiness.
Tech depth: Full SWE + domain depth + validation rigor. Eval design and acceptance criteria are first-class deliverables.
Travel intensity: 25–50%, often onsite extensively. Commure ~50%.
Day-to-day mix: Heavy compliance work. Eval suites, validation evidence, inspection readiness.

Agentic startups (overlaps with C2)

e.g. Sierra a/b/iSpec, Decagon, Harvey, Vircel

Job to be done: The ‘FDE’ is increasingly an ‘Agent Engineer’ or ‘Agent PM’ whose deliverable is a working agent (with goals, guardrails, evals, AOPs) rather than a custom application. Sierra explicitly refuses the FDE label and uses ‘Agent Engineer’ (sierra.ai/blog, December 2024). a16z calls this ‘title arbitrage’: legal engineer, GTM engineer, agent engineer, all denoting ‘AI made this work different.’
Tech depth: Full-stack engineers doing significant agent-tuning, evals, and guardrail work, different shape than research-heavy AI labs.
Travel intensity: Mostly hybrid SF/NY.
Day-to-day mix: Heavy on agent design, less on traditional integration.

Dev tools / DevEx companies

e.g. Cursor, Sourcegraph, Cognition, Replit, Vercel

Job to be done: The customer is itself an engineering team, so the FDE is an engineer-to-engineer rather than enterprise consultant. Cursor’s posting: ‘work directly with Staff/Platform/Eng leaders, going deep in code while also communicating clearly about tradeoffs.’ Workflows shipped: large-scale refactors, migration tooling, PR review loops, internal codemod platforms.
Tech depth: Strong production engineering. Production codebases included.
Travel intensity: Mostly remote/hybrid. Less customer travel than enterprise FDEs.
Day-to-day mix: Engineer-to-engineer; deep code work.

Defense / government tech

e.g. Palantir (FDE - original), Anduril (Mission Operations Engineer / Mission Software Engineer), Shield AI, Helsing

Job to be done: Embedded with military, intel, and government customers, often in austere or air-gapped environments. Palantir’s ‘Forward Deployed Software Engineer’ frames the role as ‘hands-on AI startup CTO.’ Anduril’s MOE: up to 75% travel to forward-deployed accounts, active US clearance often required.
Tech depth: Production engineering with security clearance background.
Travel intensity: Highest of all segments. Defense/gov roles add SERE training, ISOPREP, foreign-travel pre-clearance. ISOPREP, Antiterrorism Level I, APACS clearance per DoD 4500.54E. Foreign travel may require months of pre-clearance. Fully cleared FDEs are ‘the unicorns of the unicorns.’
Day-to-day mix: Field embedding > coding share.

Archetype	% coding	Travel	Mid-senior TC	Distinct artifact
A. AI Labs	40–50%	25–50%	$350–550K	Evals, agents, MCP servers
B. Data Lab	~60%	Low	$180–280K	Eval pipelines, RL envs
C1. Horizontal App	50–60%	Remote / hybrid	$200–400K	New product surfaces
C2. Vertical Agent	~40%	High in-office	$200–350K	AOPs, agent skills, guardrails
D. Enterprise SaaS	30–40%	Variable, up to 80%	$180–300K	Custom integrations
E. Regulated	35–45%	25–50%	$250–450K	Validation evidence
G. Dev Tools	60%+	Low	$200–400K	Workflow automations
H. Defense	30–40%	25–75%	$135–200K base + RSU	Mission ontology

Cross-segment quick reference