Founding / Principal Engineer, Trust & Control Infrastructure for AI Agents

Stealth AI Trust; San Francisco (CA); Hybrid; USD 190,000 - 260,000; Full time

Benefits offered by this job

Medical benefits

Dental benefits

Vision benefits

401(k)

Job summary

Stealth AI Trust in San Francisco is building a founder‑funded AI trust infrastructure company. As a founding engineer, you will work directly with the founder to create a production stack and standards for agent compliance. The role is hybrid, with 1–2 days in the office to start.

We expect deep experience in distributed systems, data infrastructure, and in production AI‑native systems. You will own critical SDKs and services used by other engineers and their agents, and you’ll help scale from

Qualifications

• Minimum six years building distributed systems or data infrastructure.
• Experience with agentic tool-calling models or MCP servers is a plus.
• Fluent in Python and TypeScript.

Responsibilities

• Framework integration with multiple frameworks within the first quarter.
• Verification path for end-to-end agent decision verification.
• Control surface for constraining, correcting, and signing off on agent decisions.
• Guarantees that costs are minimal and latency is low, measured and published.
• Design and build the production stack and tooling from prototype to scalable platform.

Skills

Python

TypeScript

Job description

Stealth · San Francisco · Hybrid, 1-2 days a week in person to start · Q4 2026

Join the Founding Engineer Team at a founder-funded AI trust infrastructure company. Working prototype in review and test, development partners next, pre-seed/seed financing in progress, founder has gone founding-to-IPO before.

THE PROBLEM

Agents are being handed real authority. They approve refunds, move records between systems, change production configuration, and negotiate with other agents. Today no one can prove what an agent saw, what it decided, or what it did, and watching an agent is not the same as verifying it.

What an agent is shown decides what it does. That makes an agent's context and its decisions an attack surface nobody owns a control point over. Humans and systems need to constrain, correct, and sign off on agent decisions, before and after the fact, not just observe them. And none of this can depend on one framework or one model: a fleet built on LangGraph, CrewAI, AutoGen, the OpenAI Agents SDK, the Anthropic Agent SDK, Google ADK, and MCP needs one independent standard underneath all of them.

We are building that standard: the way any agentic deployment, from a team's first ten agents to an enterprise's ten thousand, gains control over and insight into what its agents do, and puts them into production quickly, with trust and confidence.

How we do it stays under wraps until the development-partner phase. What it solves is above.

WHAT YOU WILL BUILD

• The framework integration layer. Our product running with at least two of the frameworks above, with identical guarantees, within your first quarter.
• The verification path. A design partner's security team can verify an agent's decisions end-to-end without us in the room.
• The control surface. How a human or a policy engine constrains, corrects, and signs off on an agent's decision, and how that holds when the agent is fast and the human is not.
• The performance envelope. Guarantees that cost agents milliseconds, not seconds. Measured, and published.
• The production stack. A working prototype exists and is in review and test. You review it, test it, and design the world‑class production stack from it: your choices of languages, services, testing, agent tooling, and on‑call become how we work.

WHAT YOU BRING

Background

• Six or more years building distributed systems or data infrastructure in production; you have owned a service other teams depended on. In the last two years part of that surface has been agentic: tool‑calling models, agent runtimes, or MCP servers that other engineers' agents consume.
• Real time in at least one of: data infrastructure (databases, event streams, historians, ledgers), security or integrity engineering, observability, tracing, and audit systems, or developer platforms. Lineage from a company whose product was the record other systems trusted counts double; so does a team that shipped an agent product to real users and lived with what it did.
• Founding or pre‑scale stage, or scaled a team from a handful to dozens, and you can say what you would do differently now that agents do a share of the engineering.

Systems experience (must have)

• You have built a system of record: an append‑only, event‑sourced, or ledger‑style store where history is immutable, replay is deterministic, and people outside your team relied on it as the truth. You know what that guarantee costs, where it breaks, and what "deterministic" has to mean when a model is in the loop.
• Cryptographic integrity in practice: hashing, signing, content addressing, verification by a party who was not there. You designed for key rotation and algorithm change up front rather than bolting it on.
• Data lifecycle under constraint: deletion or retention designed into a system that also had to stay verifiable (regulatory erasure, legal hold, audit). You have lived the tension between "immutable" and "must be forgotten," including for what an agent was shown, not only what it wrote.
• A developer‑facing surface you were accountable for: an SDK, library, API, or MCP server that other engineers, and their agents, build on, with versioning discipline, backward compatibility, and an upgrade path you owned.
• AI‑native in production: a model or agent is a production component in something you shipped. You have written evals, handled non‑determinism, built guardrails and human‑in‑the‑loop approval paths, traced an agent's tool calls end to end, and dealt with prompt injection. Production experience with at least one of LangGraph, CrewAI, AutoGen, the OpenAI Agents SDK, the Anthropic Agent SDK, Google ADK, or MCP.
• AI‑native in practice: you build with coding agents (Claude Code, Codex, Cursor, or the like) writing, testing, and reviewing code under your direction, and you are judged on what ships, not on what you typed.
• Fluent in Python and TypeScript.

Strong signal

• Standards and protocol work: you have shipped against an evolving spec (MCP, A2A, OpenTelemetry's GenAI conventions, an IETF draft).
• Multi‑tenant from day one: tenancy and permission boundaries for people and for the agents acting on their behalf.
• Shipped into a regulated or audited environment (finance, health, government, defense), built for an auditor who actually reads it.

The founding‑era bar

• Moxie. You take the hard path when it is the right one, you ship under uncertainty with a small team, and you own the outcome, not the ticket.
• Intelligence. You reason from first principles, you write the design down before you prompt the agents, and you change your mind on evidence.
• Ambition. You want to set the standard an industry adopts, not fill a seat. If you would choose this over a lab offer, your three sentences should say why.
• Signal awareness. You know which three things on the board change the outcome this week and which twenty are noise, and you say so out loud, including when the founder has it wrong.

And you can be in San Francisco one to two days a week to start, with the rest remote: already in the Bay Area.

HOW WE WORK

• San Francisco, in person one to two days a week to start, remote the rest. The frameworks, the model labs, and most of the developers we are building for are within a few miles of each other here. Office location shared at the first call.
• AI‑native by default. We use agents to build the thing that governs agents, and you are judged on what you ship.
• You would be one of the first three engineers. You work directly with the founder, a repeat enterprise‑software operator who has scaled teams from founding stage through IPO and acquisitions, and with the founding engineer.
• Decisions in days, not quarters. Founder‑funded through the prototype; pre‑seed/seed financing in progress.
• Stealth now. A development‑partner test phase precedes the public developer launch.

COMPENSATION

• Base salary: $190,000-260,000. This is the pay scale we reasonably expect to pay for this position, set in good faith; where you land depends on demonstrated depth.
• Founding equity, sized to experience and scope. Four-year vest, one-year cliff, monthly thereafter.
• Benefits: medical, dental, and vision coverage; 401(k); flexible paid time off.

HOW WE HIRE

• Step 1: a 30-minute conversation with the founder.
• Step 2: a 60-minute technical deep dive on something you built that had to scale. Your choice.
• Step 3: working together, or proceed. After the deep dive we decide together: a working session in San Francisco to meet and ideate on a real problem, with the prototype in front of you and under NDA, or, when the fit is already clear on both sides, straight to step 4.
• Step 4: references, then offer. The process moves as fast as fit and mutual alignment allow.
• A working‑together path. This seat is critical and it is a partnership, so there is a second way in: begin by meeting and ideating with us before either of us commits to the full‑time founding seat. Say in your application which path you want.

NOTICES

Pay scale. The base salary range above is the pay scale for this position, provided consistent with California Labor Code § 432.3 as amended by SB 642 (effective January 1, 2026): a good‑faith estimate of the salary range we reasonably expect to pay upon hire. Equity and benefits are in addition to base salary.

Salary history. Consistent with California Labor Code § 432.3, we will not ask about or rely on your salary history.

Fair chance. We will consider qualified applicants with arrest and conviction records in a manner consistent with the California Fair Chance Act (Gov. Code § 12952) and the San Francisco Fair Chance Ordinance. We do not ask about criminal history before a conditional offer.

Equal opportunity. We are an equal opportunity employer and do not discriminate on the basis of any characteristic protected by federal, California, or San Francisco law.

Accommodation. If you need a reasonable accommodation at any point in the application or interview process, tell us and we will provide it.

Work authorization. You must be authorized to work in the United States. We can support transfers of existing H-1B, O-1, or TN status; we are not initiating new visa sponsorship at this time.