Environment Engineering

Fleet AI, Inc.; Buffalo (NY); On-site; USD 150,000 - 210,000; Full time

Job summary

Fleet AI is seeking an Environment Engineer to own end-to-end environments, from design through first graded agent rollout. You will rebuild full-stack environments faithful to real business conventions, with agent-facing tools, data, tasks, and programmatic verifiers.

You will define the env, build it, design contracts, generate data, and collaborate with experts to ship environments that produce training signals at scale.

Qualifications

• Experience building end-to-end environments for AI agents.
• Strong TypeScript and web fundamentals.
• Ability to work with domain experts and translate feedback into code.
• Ability to ship quickly while maintaining quality.

Responsibilities

• Build and own end-to-end environments for agents.
• Design tool contracts, data models, and seed pipelines.
• Review envs and data for correctness and robustness.
• Collaborate with domain experts to QA envs and tasks.
• Instrument environments for agent training and evaluation.

Skills

TypeScript

Full-stack

Web fundamentals

AI tooling

Tools

SvelteKit

React

Node.js

Express

Docker

AWS

Postgres

Job description

About Fleet AI

Fleet AI is an applied artificial intelligence company focused on human-AI collaboration at scale.

Fleet AI is an applied artificial intelligence company focused on human-AI collaboration at scale.

Our goal is to accelerate the transition to the allocation economy, a future where humans direct work rather than perform it. We build training gyms for AI agents: high-fidelity simulation environments where agents practice real tasks while humans supervise, evaluate, and steer their behavior.

About The Role

As an Environment Engineer, you own environments end-to-end, from designing the env through to the first graded agent rollout. Each environment is a full-stack rebuild of how a real business actually runs, faithful to the original right down to its conventions, quirks, and how things break in production. The bar is high: an agent that masters our env should be able to do the real job. In practice, that means full-stack rebuilds of real products, instrumented with agent-legible tool surfaces, data, tasks, and programmatic verifiers that turn every rollout into reward.

You define the env, build it, design the tool contracts, generate the data, integrate expert feedback, and take the env from "built" to "producing training signal". Your taste at every stage shapes what agents can learn.

Own a Vertical End-to-end

• Build the env: research the real product and ship the full-stack app, including the data model, UI, business logic, tool surfaces legible to agents, and seed pipelines holding realistic, resettable state across millions of rollouts. The first graded rollout proves the whole stack works.
• Own the hardest envs on the roadmap: multi-step tasks that need coherent state across many actions (like closing the books at quarter-end, or running a multi-week candidate sourcing pipeline), interactions with simulated stakeholders who push back (a customer disputing a charge, a hiring manager rejecting a candidate), real codebases with conventions and tech debt rather than toy repos, and open-ended design problems with no single right answer. Once these are built, much of the work is debugging: when an agent fails, distinguishing genuine environment issues from artifacts and redesigning to target deeper failure modes.
• Review at scale: as Fleet's env library grows, quality only holds if review keeps pace. You'll programmatically review envs, data, and tasks for correctness and robustness, advise on data, task, and grader design, and build and maintain the review standards we need to keep scaling.
• Expert feedback: work with domain experts to build, review, and QA envs, data, and tasks at volume, and ship the systems that keep that feedback loop tight.
• Leverage: direct fleets of coding agents to parallelize development and raise the ceiling on what one engineer ships.
• Platform: shape the shared abstractions and quality bars every Fleet environment inherits.

You May Be a Good Fit If You

• Are a high-agency, deeply technical full-stack builder with strong TypeScript and web fundamentals
• Have product taste. You can study a real app and know which details matter to faithfully recreate it
• Are comfortable working directly with domain experts (accountants, recruiters, IT admins) and turning their feedback into code
• Ship fast and thrive in ambiguity. You can figure out the right approach without a detailed spec
• Use AI tools daily and can effectively direct multiple coding agents in parallel
• Sweat the details that make a simulation indistinguishable from the real thing
• Think in terms of determinism, invariants, and contracts, not just features
• Have a deep appreciation for AI research and low-level infrastructure

You Will Not Be a Good Fit If You

• Thrive on routine, consistency, and long planning cycles. We ship in days, not quarters.
• Want to go deep on one narrow technical problem; this role spans the stack and crosses into research, infra, and product
• Need a clean handoff between spec, design, and engineering. Here, those are the same person.
• Aren't excited to spend serious time communicating clearly with AI agents
• Move fast without caring about what you ship. We don't trade quality for speed.

Stack

TypeScript, Svelte/SvelteKit, React, Node.js, Express, oRPC, Tailwind, SQLite/Postgres with Drizzle, Python, Docker, AWS (ECR/ECS), MCP (Model Context Protocol), Vitest, Playwright, Claude Code and multi-agent workflows.

Compensation & Structure

Full-time, on-site in SF or NYC. Compensation (salary + meaningful equity) will be extremely competitive.

Our Philosophy

We believe the most important advances in AI will come from close, thoughtful collaboration between humans and machines. We own our outcomes, move with relentless speed when we build conviction, and choose truth over comfort. We don't cut corners. We cut scope.

We encourage you to apply even if you don't meet every qualification listed. We care far more about how you build and think than about credentials alone.