AI Harness Engineer - W2

Stealth Talent Solutions; United States; On-site; USD 150,000 - 185,000; Full time

Job summary

Stealth Talent Solutions is seeking a senior AI/ML software engineer to join a startup-like team inside a Big 4 professional services firm. The role focuses on building and operating an end-to-end agentic system, with strong emphasis on Python, production deployments, and AI-first workflows.

You will work across orchestration, tool design, context engineering, and observability, all while collaborating with a small, autonomous group and traveling up to 25% to engage with customers.

Qualifications

• Hands-on coding today; still writing code.
• Strong Python experience.
• Built and shipped an agentic system to production.
• Supported one of these in production.
• AI-first in your workflow with defined method.
• Creative and adaptable across frameworks.
• Enterprise or regulated-environment experience.
• Able to travel roughly 25%

Responsibilities

• Build and own the agent loop: orchestration, state management, step budgets, termination conditions
• Design the tool layer: tool schemas, argument validation, dispatch, and error surfaces a model can act on
• Solve context and tool-selection problems at scale, including what happens when an agent has 25 tools and starts choosing badly
• Own context engineering: what the model sees, when, and at what cost
• Instrument observability before you ship, not after, so you can diagnose a production incident from logs and dashboards alone
• Build for provider reality: rate limits, retries, circuit breakers, failover across models
• Take work through the full lifecycle and support it in production
• Deploy alongside customers, understand their workflow and pain points, and bring that back into the product

Skills

Python

Hands-on coding

Agentic systems

Production deployment

Observability instrumentation

AI-first workflow

Enterprise/regulatory experience

Travel readiness

Tools

LangGraph

LangChain

MCP

Rust

Go

Job description

Travel: Roughly 25%, up to one week per month on site or with customers

Start: Immediate

Team: Small team, new group. Not an isolated seat.

About the client

A Big 4 professional services firm. This role sits inside their AI and Digital organization, a group of roughly 60 engineers, designers and cloud operations people that builds AI services used across a 40,000-person firm. The group evaluates emerging AI platforms, builds proofs of concept, and takes selected applications into production. Primary cloud is Azure with some Google Cloud, and they use Anthropic and OpenAI models with no vendor lock.

About the initiative

A newly formed group standing up to build the harnesses that take AI agents from prototype to deployed product. It is structured deliberately like a startup inside the firm: a small team owning a product end to end, real technical autonomy, no architecture council.

Here is the honest version of what you are signing up for. Nobody knows yet what you will build. It will be one thing tomorrow and something else the week after, driven by what the firm needs and how fast the models move. You will not be handed a roadmap. If that sounds unsettling, this is the wrong role. If it sounds like the interesting part, keep reading.

What you will do

• Build and own the agent loop: orchestration, state management, step budgets, termination conditions
• Design the tool layer: tool schemas, argument validation, dispatch, and error surfaces a model can act on
• Solve context and tool-selection problems at scale, including what happens when an agent has 25 tools and starts choosing badly
• Own context engineering: what the model sees, when, and at what cost
• Instrument observability before you ship, not after, so you can diagnose a production incident from logs and dashboards alone
• Build for provider reality: rate limits, retries, circuit breakers, failover across models
• Take work through the full lifecycle and support it in production
• Deploy alongside customers, understand their workflow and pain points, and bring that back into the product

Required

• Still hands-on in the code, today. If the honest answer to "are you writing code right now" is no, this is not the role
• Python. The core language here
• You have personally built and shipped an agentic system to production. You can walk through the use case, the architecture, the execution path, and what broke
• You have supported one of these in production. You know what fails first under load, and you have built the instrumentation that let you find out
• AI-first in your own workflow. You work through coding agents daily and have a real method: how you plan, how you parallelize sessions, how you review what gets generated, how you keep quality gates in place
• Creative and adaptable. Breadth beats depth in any single framework. You have dropped into unfamiliar problems and territory and made them work
• Enterprise or regulated-environment experience. You have shipped inside security review, compliance constraints, and change control
• Able to travel roughly 25%

Nice to have

• Full stack. Useful, not decisive. If you can build a React or Next.js app with a coding agent without being a front-end expert, that is enough
• LangGraph, LangChain, MCP, or any specific agent framework
• Rust or Go

Process

Two interviews. A 30-minute technical conversation with the engineering lead, then a round with the partner who owns the initiative. There is real urgency on this.