Agent Engineer

Overjoy

New York (NY)

On-site

USD 120,000 - 160,000

Full time

8 days ago

Application generator

Don’t send a generic resume — generate a resume and cover letter tailored to this exact role.

Get past ATS filters

Benefits offered by this job

Competitive salary + meaningful equity

Health, dental, and vision insurance (

Free lunch & dinner around the office

Generous budget for laptops, monitors,

Job summary

Overjoy is seeking an Agent Engineer to own the agents at the center of our platform—systems that find retailers, run outreach, manage pipelines, and capture orders for our brands. You will ensure agents work in production, not just in demos, with strong emphasis on reliability, observability, and cost.

You’ll join a small team, work directly with the founders, and deploy agents for real brands within weeks.

Qualifications

You've built production agent systems you can point to and discuss what broke and what you did about it.
You have deep hands-on experience with prompt engineering, context optimization, tool design, and multi-agent orchestration.
You've debugged agents in production and know how to make them predictable at scale.
You have a strong backend foundation. Agents live in queues, databases, and APIs.
You treat evals as part of the product, not a checkbox at the end.
You want real ownership and are comfortable deciding how an agent should behave.

Responsibilities

Own the agents behind our services, focusing on outreach, pipeline, closing, reorder, and capture from PDFs and emails.
Design how a brand is seen by an agent through knowledge bases, CRM, emails, catalog, and rules.
Develop long-running agent workflows that survive retries and edge cases.
Build an evaluation loop using usage data and production monitoring to improve performance.
Shape human-in-the-loop aspects and determine where automation vs. human input occurs.
Ship agent capabilities integrated into the CRM, inbox, and product back-end.

Skills

Agent systems

Prompt engineering

Context optimization

Multi-agent orchestration

Backend development

Tools

Pydantic AI

Temporal

Logfire

OpenAI

Claude

Job description

About Us

Overjoy is an AI-native sales company for SMBs. We began as a sales platform for consumer brands, then expanded to running the full sales process end-to-end and selling the outcome.

Every consumer brand needs a wholesale department: someone to find the right retailers, reach out, close the account, and keep it reordering. For most brands that department is too expensive to hire. Overjoy runs it for them with people, AI agents, and the software we spent years building for exactly this job.

Over 800 brands like Nature's Path, Milk Bar, and Simply Gum use Overjoy to grow wholesale revenue.

We're backed by Bessemer, OSV, Floodgate and founders of Thumbtack, CareRev, Ando, and Unsupervised — and 6 of our own customers invested after using the product.

The Role

We're hiring an Agent Engineer to own the agents at the center of Overjoy: the systems that find retailers, run outreach, work the pipeline, and capture orders for the brands we serve.

The central problem of the role is making agents hold up in production, not just in demos. Every agent at Overjoy does work a customer is paying for. When an outreach agent sends a weak email, a brand loses a retailer. When a reorder agent stalls, an order doesn't get placed. Reliability, capability, and observability are the whole job.

You'll join a small team, work directly with the founders, and have agents running for real brands within your first weeks. There is real market pull for our services offering, and every new brand we take on means more agents running around the clock.

Stack: Python · FastAPI · PostgreSQL · Pydantic AI · Temporal · Logfire · frontier and open models (OpenAI, Claude, open-source)

What You'll Own

The agents behind our services. Outreach, pipeline and closing, reorder and customer engagement, and order capture from PDFs and email threads. Their capability, their reliability, and their cost.
Context engineering. How an agent sees a brand: the brand knowledge base, the CRM record, email history, product catalog, and the rules the brand has set. What goes into the window, what stays out, and how it stays current.
Long-running agent workflows. Multi-step agents on Temporal that survive retries, partial failures, rate limits, and the edge cases real retailers throw at them.
Evals and the improvement loop. Offline evals built from our own usage data, online monitoring in production, and the loop that turns a bad agent output into a fixed one. Our reply rates have improved by mining our own data; you'll make that systematic.
Human-in-the-loop design. Where an agent acts on its own, where it drafts for our services team, and how work hands off in both directions without anything falling through.
Shipping agents into the product. Integrating agent capabilities into the CRM, inbox, and automations alongside our product and backend engineers.

You're a Fit If

You've built production agent systems you can point to, and you can talk about what broke and what you did about it.
You have deep hands-on experience with prompt engineering, context optimization, tool design, and multi-agent orchestration.
You've debugged agents in production and know how to make them predictable at scale.
You have a strong backend foundation. Agents don't live in isolation; they live in queues, databases, and APIs.
You treat evals as part of the product, not a checkbox at the end.
You want real ownership and are comfortable being the person who decides how an agent should behave.

Bonus Points

You've built agents for sales, support, or another domain where the output is judged by a customer rather than a benchmark.
You've fine-tuned a model or built retrieval for a specific domain.
You've worked with Pydantic AI, Temporal, or LLM observability tooling (Logfire, Braintrust, LangSmith) in production.
You've turned agentic coding into a team multiplier: shared harnesses, review loops, and quality gates, not just personal productivity.
You've worked on email deliverability, CRM data models, or B2B commerce.

Why Overjoy?

Build for a real problem in a wildly underserved market. Brand founders with multiple exits tell us, "I wish I had Overjoy when I launched my first brand." Wholesale has always been the most expensive channel for a small brand to staff. We're making it the cheapest.

Agents that do the work, not features that help. We spent years building sales tools for people. It turns out agents learn and use those tools better and faster than any person could. Our customers were limited by time and headcount; now they get an AI team instead of one overworked founder. You'll build the agents that find retailers, close accounts, and keep them reordering, and you'll see them run every day.

You'll shape how we build. We're past searching for product-market fit and are building the foundations under it. You'll define how we build, evaluate, and ship agents, and those standards will compound for years.

Our customers love us. We work with hundreds of small and mid-sized brands, startups in their own right, and build closely with them for their very specific needs. When you solve pain points nobody else has touched, the love is loud and frequent.

We're built to last. We run a capital-efficient business with a clear path to profitability, so you're building on a foundation, not a countdown.

Why You Shouldn't Join

This isn't a 9-to-5 job. We're doing hard things with a small team, and it's intense. We won't sugarcoat that.

The autonomy cuts both ways. You'll decide how an agent should behave, make the call, and own being right or wrong about it in front of a paying customer. That can be exciting, but also stressful, and it's not for anyone.

You'll wear a lot of hats. One day you're rewriting an agent's context pipeline. The next you're reading a brand's inbox to figure out why a reply went wrong, and shipping an eval so it never happens again. If you prefer a narrow, clearly-bounded role, this might not be the right environment.

Perks & Benefits