AI Agent Builder

Job Description

About AUI

AUI builds Apollo, a neurosymbolic foundation model for AI agents that do consequential work, where accuracy, reliability, and visibility matter — conversations with customers, and the processes behind them. Instead of describing behavior in a prompt and hoping the model complies, Apollo compiles declared rules into a checkable program. Constraints are enforced by the system rather than requested of the model, so “this agent will never do X” is something you verify, not something you hope for. The language model works as a translation layer at either end — structuring what a person (or another agent, or a system) says into symbolic slots on the way in, and generating fluent responses on the way out.

That guarantee is what LLMs structurally can never achieve. We serve companies from startups to the Fortune 500, across regulated industries, retail, and travel — environments where an agent that improvises is a liability, and where someone has to answer for every action it takes.

Apollo-1 is a self-serve platform, and most teams can build on it themselves. Large enterprises — with deeper integrations and more involved internal processes — can use a hand getting their agents all the way to production. Our commercial team also needs proof-of-concept agents built for prospects, fast. That’s this team.

The role

You’re joining our deployment team, serving two constituencies.

Internal. Our product and commercial teams need working agents — proofs of concept for prospects, reference builds, agents that pressure-test the platform before customers find its edges. You’ll build them, often on short notice, often for a business you learned about today.

External. The most complex enterprise deployments, where the agent has to hold a real company’s rules across hundreds of scenarios and act on live systems. You’ll be in the room with their operators and subject-matter experts, not reading a requirements doc someone else wrote.

An agent starts life as a real business: a bank’s dispute policy, an airline’s rebooking rules, a retailer’s returns matrix. Someone has to sit inside that mess, understand how the business actually works, and turn it into a program the platform can enforce and certify. Then train it, grade it, try to break it, and sign off that it’s ready to act on a customer’s behalf.

The loop closes back into the product. What you solve by hand for one enterprise should become something the next customer can do themselves, and you’ll be one of the loudest sources of truth about what the platform can’t do automatically.

What you’ll do

\n
• Sit with partner teams to extract how their business actually operates — including the rules nobody wrote down and the exceptions everyone knows about
\n
• Author agent programs: translate policy, process, and edge cases into structured, unambiguous declarations the platform can enforce
\n
• Run training rounds — design scenarios, simulate conversations, grade outputs, and iterate until behavior holds
\n
• Own certification: prove a program does what it claims before it ships, and triage it when it regresses
\n
• Find the contradictions. Real businesses run on rules that conflict. You’ll be the one who notices the refund window doesn’t square with the cancellation clause, and who drives the decision on which one wins
\n
• Build proofs of concept alongside our commercial team, and support technical discovery with large prospects — you’re the person in the room who can answer “can it really do that?”
\n
• Turn one-off solutions into reusable patterns, so the self-serve product absorbs what the deployment team learns
\n
• Feed the field back to product: what’s slow, what’s missing, what breaks at scale
\n

Who we’re looking for

We care more about how you think than where you’ve worked, with one hard requirement.

-Technically fluent — not a coder. You’ve worked inside a technological product environment and you understand how systems fit together: what an API is doing, where data moves, why an integration breaks, what engineers mean when they say something is hard. You use coding agents — Cursor, Claude Code, or equivalent — as a normal part of how you work, not as something you tried once. You won’t be shipping production code here, but you will be building, and the tools are the same.

-Opinionated about AI. You’re curious about it end to end: roughly how the models actually work, what they’re doing to industries and to work itself, where this is all going. We don’t need you to hold the right opinion. We need you to hold one, and to be able to defend it and change it.

-Curious about how businesses run. You find it genuinely interesting that an airline’s rebooking logic looks nothing like a pharmacy’s intake flow. You’ve noticed that industries have structure, and you like learning a new one from scratch.

-Independent. You’re given a partner, a pile of documents, and an outcome. Nobody is going to hand you a ticket queue. If you need the work broken down for you, you’ll be unhappy here.

-Precise and stubborn. This work rewards people who will handle the ten-thousandth case as carefully as the tenth, and who push back rather than write “handle appropriately” to make a hard question go away. Most people can’t do this. The ones who can are extremely valuable.

-You’ve failed at something real. A company, a project, a venture that didn’t work. We’d rather hire someone who has been responsible for an outcome and lost than someone whose record is unblemished because nothing was ever on the line.

-A clear writer. Most of this job is written. If you can’t state a rule in a sentence that can only be read one way, the agent can’t follow it.

No degree required. It isn’t a proxy for anything we care about.

How we work: healthy ego

Our culture runs on something we call healthy ego, and it has two halves. Both are required.

The first half is ego. High confidence that you can learn and succeed at anything — paired with actually asking for what you need to get there, and then working until the result is excellent rather than finished. We also expect you to want things to succeed for your own sake, not only the company’s. AUI is a place to push yourself and evolve fast. No one will push you and no one will carry you, but everyone will work with you. We sharpen each other by challenging each other, and by the simple pressure of keeping up with the team.

The second half is what makes it healthy: your success is never built on top of someone else’s. You name your own mistakes before anyone else has to, and you go fix them — by yourself or with the team.

We’re flat. Merit decides what you own, not title or tenure.

Backgrounds that tend to work

The common thread is time spent close to a technical product, plus real fluency in how some corner of the business world operates.

\n
• Solutions, implementation, or technical account management at a software company
\n
• Support escalations, Tier-2, or support quality — the people who own the cases nobody else could close
\n
• Product operations, technical program management, or ops inside a product organization
\n
• Operations in a rule-dense industry — claims, reservations, banking, healthcare intake, logistics — paired with time in a product or technology environment
\n
• Founders and operators of things that didn’t make it
\n

What this is not

\n
• Not an engineering role. You won’t own production code.
\n
• Not a non-technical role either. If you can’t hold your own in a conversation about how a system works, this will be painful.
\n
• Not data annotation or gig AI-training work. This is a salaried role where you own outcomes, not task throughput.
\n
• Not a support queue. You’re building the thing, not staffing it.
\n

Logistics

We’re in Williamsburg, Brooklyn, in the office together five days a week. That isn’t a policy we’re apologetic about — this work runs on overhearing each other. A deployment question gets answered across a desk in thirty seconds instead of dying in a thread, and the challenge-each-other part of how we work doesn’t survive being asynchronous. If that’s not how you want to work, this isn’t the right role.

How to apply

Skip the cover letter. Send your résumé plus answers to one question:

\n
1. Tell us about something you owned that failed. What did you actually get wrong? (300 words max)
\n

The interview process includes an in-office assignment — the closest thing to the actual job we can put in front of you.