AI Agent Architect

Senior IC who decides whether a problem needs one agent, three, or none. Cares about contracts between agents, escalation paths, and the moments where probabilistic systems must defer to deterministic ones.

Indicative comp$220K – $320K base (US, staff/principal)

Ranges are indicative US base salary at senior level. Actual offers depend on company stage, equity, and candidate strength.

What this role actually owns

Decide where agents earn their keep and where deterministic code is the right answer.
Define interfaces between agents, tools, and human reviewers.
Set policy on retries, timeouts, escalation, and approval gates.
Author the technical strategy doc the team builds against.
Mentor agentic AI engineers on patterns that hold up at scale.

What we screen for

8+ years engineering, with at least one staff/principal-equivalent role.
Has architected a system with 3+ agents in production — can sketch its topology on a whiteboard.
Articulate about tradeoffs: planner-executor vs. supervisor-worker, monolithic vs. specialized agents.
Has written an RFC the rest of the team actually used.
Bonus: background in distributed systems or workflow engines.

Sample job description

A starting point you can paste into your ATS and adjust. The exact wording matters less than the rubric — the bullets above are what we'll calibrate against during search.

AI Agent Architect

Designs the system shape — which agents exist, what they own, how they hand off, and where the human stays in the loop.

You'll own:

Decide where agents earn their keep and where deterministic code is the right answer.
Define interfaces between agents, tools, and human reviewers.
Set policy on retries, timeouts, escalation, and approval gates.
Author the technical strategy doc the team builds against.

We're looking for:

8+ years engineering, with at least one staff/principal-equivalent role.
Has architected a system with 3+ agents in production — can sketch its topology on a whiteboard.
Articulate about tradeoffs: planner-executor vs. supervisor-worker, monolithic vs. specialized agents.
Has written an RFC the rest of the team actually used.
Bonus: background in distributed systems or workflow engines.