Job description

ABOUT MINERVA Minerva builds AI for marketing leaders. We work together with marketers to automate and accelerate their workflows with AI agents. Our agents augment the end-to-end workflow of the marketer by automating analysis, campaign planning, activation, measurement and reporting. Customers like the NBA, Ramp, Wander, Hard Rock Stadium Group and Luxury Presence use Minerva to develop a better understanding of their customers and introduce AI specifically built to understand the customer into their commercial organizations. Our campaigns outperform other tools because we built the entire stack: from identity to enrichment to data integration as well as agents to action and report on the analysis. We've raised $20M from The General Partnership, 8VC, Lingotto, NBA Investments, Topology Ventures, Future Positive, Background Capital and many others. Our team brings together operators and investors from Citadel, Dentsu, Bridgewater, Meta Superintelligence, Growthloop, Amazon and Lazard, alongside engineers and researchers from Berkeley, MIT, Stanford and Cambridge. OUR STACK - Agents: Claude across the company (in Slack, through the API, with skills and MCP), plus Mastra for the agents we build ourselves and Temporal for long-running work. - Code: Python and TypeScript, scheduled jobs and webhooks. - The systems your agents work across: Slack, Google Workspace, Notion, and the CRM, recruiting, HR, finance and compliance tools a growing company runs on. THE ROLE We build agents for marketers. This seat builds them for us. Every company our size runs on work nobody sees. Someone copies a new hire's details into six tools. Someone remembers which contract renews next month. Someone answers the same security questionnaire for the fourth time. Most companies hire a coordinator for that work. We're hiring an engineer. You're a forward-deployed engineer for Minerva itself. Other companies call this an AI operations engineer or an internal AI engineer. Your customers are every team here: operations, finance, people, recruiting, sales and the founders. You sit with them, find the work they'd most like to stop doing, and replace it with a system that runs without them and leaves an audit trail. You report to our COO and work closely with engineering, which owns anything that touches production. WHAT YOU'LL DO - Own the company's knowledge agent. One agent over our docs, Slack, calls and tickets, with Slack as the front end. It answers with sources, and people trust it. - Automate the processes that run the company. Hiring, onboarding and offboarding, access provisioning, vendor reviews, customer security questionnaires, contract renewals and billing. Start where the hours or the risk are highest. - Run the skill library. Every team gets Claude connected to the tools it uses and a set of skills, each with a named owner on that team, an eval and the right permissions. What one person figures out, everyone gets. - Own the source of truth. Decide where each kind of information lives, move it there and keep it findable by people and by agents. When an agent spots something out of date, it proposes the fix and a person approves it. - Build the plumbing under our operating dashboards: the metrics, controls and board reporting that today get stitched together by hand from five systems. - Put the guardrails in first. Least-privilege access, a human approval before any agent writes to a system of record, and audit trails that double as compliance evidence. - Sit with your users. Watch how a recruiter, a seller or our finance lead actually works, then automate the part they hate. - Measure adoption. Which systems get used, by whom and at what cost, including what we spend on models and which agents can touch what. Retire what nobody opens, and write the runbook for what stays. WHAT YOU HAVE - You've shipped agents or automations that coworkers use every week, and you can show the usage: who stopped doing what, and how many hours came back. - Production Python or TypeScript, with LLMs wired into real APIs. You're comfortable with OAuth scopes, webhooks and scheduled jobs, and with the ways each of them fails. - You know how a company runs. You can talk to a recruiter, a controller and a security auditor in their own terms, and you fix the process before you automate it. - A controls instinct. Before you ship, you've decided who approves what and what happens when the agent is wrong. - You're current. You know what shipped in AI tooling this month, and you can show us your own setup. - You write clearly. People follow your runbooks without asking you. - You're in New York City or willing to relocate. YOU MIGHT BE A FIT IF YOU ARE… - A founder or founding engineer who automated your own company's back office because someone had to. - A chief of staff or business operations lead who learned to build and stopped writing specs for engineers. - A solutions or forward deployed engineer who wants the whole company as your customer. - A GTM or internal tools engineer who automated a sales team and wants scope beyond the sales stack. - Or you don't fit any of these neatly, but you've automated your own job at least once. Apply. NICE TO HAVES - You've rolled out Claude or another model platform to a whole company: access, permissions, usage and cost. - You've been through a SOC 2 or similar audit on the side that produces the evidence. - Hands-on time with the systems a startup runs on: HRIS, ATS, CRM, billing and spend tools. WHAT THIS ISN'T - An IT helpdesk. You'll automate the tickets away, not answer them. - Product engineering. Our customer-facing agents and production infrastructure belong to the engineering team. - An AI strategy role. You ship systems people use. Nobody here needs a deck about transformation. - The owner of the work you automate. Our compliance lead still owns the security answers, and whoever runs sales ops owns the CRM. You build the systems that do the work for them. WHAT SUCCESS LOOKS LIKE - First 90 days: the knowledge agent is live in Slack and used daily. A new hire gets set up in every tool in under an hour, and the access record writes itself. At least three workflows run without a person in the loop and are still in use four weeks later. - First 6 months: security questionnaires go out the same day without a founder writing them. Contract renewals, notice windows and price increases surface before they cost us money. Every team has a system it would fight to keep. - The number you own: hours of manual work retired each month, measured against a baseline you set in your first month. COMPENSATION $211,000–$236,000 base salary, depending on experience. Meaningful equity. Medical, dental and vision. HOW TO APPLY Skip the cover letter. Show us one thing you built that other people used: a demo, a repo, or a few paragraphs on what the process looked like before, what you built, how many people used it and what broke. Applications without one won't be reviewed. WHERE WE WORK We work in person from our office at 300 Kent Avenue in Williamsburg, Brooklyn. BENEFITS Every full-time role includes equity; medical, dental and vision coverage; unlimited PTO; a monthly fitness stipend; and an annual company offsite. We'd rather pay people well than buy perks nobody uses. PAY TRANSPARENCY The salary range in this posting is our good-faith range for the role in New York City at the time of posting. Where an offer lands depends on the level and experience we see in the interview process. APPLYING Apply on this page. We don't take applications by email, and we don't accept unsolicited submissions from recruiting agencies. If you need an accommodation at any point in the process, write to hiring@minerva.io. WORK AUTHORIZATION You'll need to be authorized to work in the United States. We consider visa sponsorship case by case, so tell us on the application if you'll need it. EQUAL OPPORTUNITY Minerva is an equal opportunity employer. We make hiring decisions without regard to race, color, religion, creed, sex, pregnancy, sexual orientation, gender identity or expression, national origin, citizenship status, age, disability, marital status, veteran status or any other characteristic protected by federal, New York State or New York City law. RECRUITMENT FRAUD Our hiring runs through Ashby and @minerva.io email addresses. We will never ask you for money, bank details or other financial information during the process. If a message claiming to be from Minerva looks wrong, forward it to hiring@minerva.io.

Requirements

Must-haves

Agents: Claude across the company (in Slack, through the API, with skills and MCP), plus Mastra for the agents we build ourselves and Temporal for long-running work.
Code: Python and TypeScript, scheduled jobs and webhooks.
The systems your agents work across: Slack, Google Workspace, Notion, and the CRM, recruiting, HR, finance and compliance tools a growing company runs on.
We build agents for marketers. This seat builds them for us.
You report to our COO and work closely with engineering, which owns anything that touches production.
Own the company's knowledge agent. One agent over our docs, Slack, calls and tickets, with Slack as the front end. It answers with sources, and people trust it.
Run the skill library. Every team gets Claude connected to the tools it uses and a set of skills, each with a named owner on that team, an eval and the right permissions. What one person figures out, everyone gets.
Own the source of truth. Decide where each kind of information lives, move it there and keep it findable by people and by agents. When an agent spots something out of date, it proposes the fix and a person approves it.
Build the plumbing under our operating dashboards: the metrics, controls and board reporting that today get stitched together by hand from five systems.
Put the guardrails in first. Least-privilege access, a human approval before any agent writes to a system of record, and audit trails that double as compliance evidence.
Sit with your users. Watch how a recruiter, a seller or our finance lead actually works, then automate the part they hate.
Measure adoption. Which systems get used, by whom and at what cost, including what we spend on models and which agents can touch what. Retire what nobody opens, and write the runbook for what stays.
You've shipped agents or automations that coworkers use every week, and you can show the usage: who stopped doing what, and how many hours came back.
Production Python or TypeScript, with LLMs wired into real APIs. You're comfortable with OAuth scopes, webhooks and scheduled jobs, and with the ways each of them fails.
You know how a company runs. You can talk to a recruiter, a controller and a security auditor in their own terms, and you fix the process before you automate it.
A controls instinct. Before you ship, you've decided who approves what and what happens when the agent is wrong.
You're current. You know what shipped in AI tooling this month, and you can show us your own setup.
You write clearly. People follow your runbooks without asking you.
You're in New York City or willing to relocate.
YOU MIGHT BE A FIT IF YOU ARE…
A founder or founding engineer who automated your own company's back office because someone had to.

Nice-to-haves

You've rolled out Claude or another model platform to a whole company: access, permissions, usage and cost.
You've been through a SOC 2 or similar audit on the side that produces the evidence.
Hands-on time with the systems a startup runs on: HRIS, ATS, CRM, billing and spend tools.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aiapipythontypescriptgo

Platform: ashby
Effort: quick

Benefits

Benefits detected in the listing.

Dental

Vision

Pto