Join us to changewho does the work.

We build AI agents that run entire functions inside real companies. Come build the next ones with us.

8 open roles

More than thirty agents in production, real clients across Spain, and a product growing exponentially. The platform underneath them already works, and the job is making it better as it grows: how the pieces fit, how the database scales, and what is worth changing before we multiply by ten. You bring the proposal, we decide it together, and you build it with the team.

What you'll do

  • The architecture new agents get built on.
  • The database: the model, how it grows, how it stays isolated per client.
  • The decisions that are expensive to undo, made early and written down.
  • Turning what the business needs into something the team can build.

What we look for

  • You have designed systems that grew and kept working.
  • Strong with databases, comfortable in Python.
  • You explain a design as easily to a founder as to an engineer.
  • You still write code, and it shows.
Three steps, about ten minutes.

An agent here is not a prompt. It is a deterministic engine, a model layer on top, its tools, its memory and the human gates in between. You work on both sides: the agents themselves, and the platform they run on, which is our own harness, the orchestration between agents, and what it takes for dozens of them to work at once without stepping on each other.

What you'll do

  • New agents end to end: schema, tools, API, dashboard and the evaluations that prove they work.
  • The harness: the tool-calling loop, memory, retries, timeouts and cost limits.
  • Multi-agent orchestration. Agents that hand work to each other and stay coordinated.
  • Model choice. Which model, at which step, at what cost, and how fast we can swap it.

What we look for

  • 3+ years of Python in production, and SQL that does not scare you.
  • You have worked with agent harnesses: Claude Code, Codex, agent SDKs, LangGraph, or one you built.
  • You have built on an LLM API and know exactly where it lied to you.
  • You write the test for the boring case before the demo of the exciting one.
Three steps, about ten minutes.

You work on the platform underneath the agents: multi-tenant data isolation, the job scheduler, the cost ledger, the shared API. It is the piece the rest depends on, which is why good work here is felt the same day, across the catalogue.

What you'll do

  • The multi-tenant core. Isolation is a correctness property here, not a feature.
  • Scheduling and background execution for agents that run overnight, unattended.
  • Per-call cost attribution, so every euro spent on a model has a name on it.
  • The performance work nobody assigns you, because you are the one who sees it.

What we look for

  • 5+ years of Python in production, and Postgres well beyond writing selects.
  • You have run services with real traffic, on call, and came out of an incident with a better design.
  • You care about the boring guarantees: idempotency, retries, audit trails.
  • You would rather delete a subsystem than defend it.
Three steps, about ten minutes.

Every new agent brings a new surface, and that surface is where a client decides whether to trust what the agent just did. You look after the front end of the product, and the standard it is held to.

What you'll do

  • The screens where clients approve, correct and reject what agents propose.
  • A design system that stays coherent while the catalogue grows underneath it.
  • Light and dark, Spanish and English, without a second pass.
  • Speed on an old laptop in a back office, not on the machine you develop on.

What we look for

  • 4+ years of React and TypeScript in a real product, with opinions about state you can defend.
  • You care whether a table of two thousand rows still feels instant.
  • You can take a rough sketch or a rough sentence and return something better.
  • You want to answer for the front end, not for a lane inside somebody else's.
Three steps, about ten minutes.

Our agents touch accounting, payroll and tax, so being right counts for more here than being impressive. You strengthen the safety net that lets everybody else move fast, and your call on whether something is ready carries weight.

What you'll do

  • The tests and evaluations an agent has to pass before a client sees it.
  • Extra care where money is involved, because a mistake ends up in somebody's books.
  • Release checks, in both languages, before anything goes out.
  • Chasing what breaks in the field until it is fixed for good.

What we look for

  • 2+ years testing a real product, by hand and with code.
  • Enough Python or TypeScript to write the test yourself.
  • Naturally suspicious: you go looking for the case nobody saw at first glance.
  • You can say something is not ready, say it early, and say it without a fight.
Three steps, about ten minutes.

Our clients run serious operations on systems that were never designed to talk to each other: legacy databases, shared mailboxes, photographs of paper. You go in, map that reality, and come back with something the agents can work with.

What you'll do

  • Onboarding: data sources, first configuration, first month of close supervision.
  • The bespoke work. Some clients get an agent built for them, and you write that brief.
  • Training the people who approve what the agent proposes every morning.
  • Turning what you see in the field into changes engineering actually makes.

What we look for

  • Technical enough to write a query, human enough to train a room of twelve.
  • You read a client well: what they need, what they say they need, and what they will not say.
  • You answer fast and come back with options instead of questions.
  • Native Spanish and a driving licence. Some of this happens on a factory floor.
Three steps, about ten minutes.

This is a hunter's seat. You fill your own pipeline, you ask for the meeting, you run the demo and you close. Proven product, a market that has barely been touched, and nothing between you and the signature. What you sell shows up in the company the same week.

What you'll do

  • Your own pipeline. Outbound, network, events, whatever gets a meeting in the calendar.
  • The full cycle: first contact, discovery, demo, pilot, contract, first renewal.
  • Running the demo yourself. You will know the agents well enough to break them on purpose.
  • Pricing against the real alternative, which is usually an extra hire or an outside firm.

What we look for

  • 3+ years closing B2B software in Spain, and you can name the deals and the ticket size.
  • You have hit quota two years running and can show it.
  • Technical enough to demo without an engineer beside you.
  • You would rather qualify out on Monday than nurse a deal for a quarter.
Three steps, about ten minutes.

We are opening the United States, and the first client there gets closed by somebody already in it. No local playbook, no local brand, nobody who has heard of us: pure hunting, with a product that already works behind you. San Francisco is where we intend to land, and you are likely to be the reason we do.

What you'll do

  • The first American clients, from cold outreach to signature.
  • What has to change about the product for this market, argued straight to engineering.
  • The case for where the office should be, and eventually who staffs it.
  • What a first hire carries in a market: partners, events, the first references.

What we look for

  • 5+ years selling B2B software in the United States, with quota attainment you can show.
  • You have opened a territory, or been early in a company that did.
  • You are comfortable being the only person in the country for a while.
  • You are ready to be measured on closed revenue rather than activity.
Three steps, about ten minutes.
Barcelona today, San Francisco and Madrid next. Permanent, full-time contract on every role.
Who we are

Every company still does the invoices by hand.

Vermoon was born in Chicago and is built in Barcelona. We make AI agents for other companies. Not assistants and not chatbots: software that takes a whole function, runs it end to end, and reports back on what it did.

The agents that matter don't answer questions. They close the month.Finance

More than thirty are in production today, working every day for real companies across Spain, in Spanish and in English, on systems that were never designed to talk to each other.

They don't summarise the invoice. They chase it until somebody pays it.Collections

They keep double-entry books, reconcile banks, plan production, count stock, run campaigns, screen candidates and keep the records an inspector will ask for. When a company needs something that is not in the catalogue, we build that one too.

They don't suggest a stock count. They do it, before the shop opens.Operations

The people doing that work by hand are capable of far more, and they spend their week on the part a system should have handled. It is the most expensive habit in any company, and nobody has time to fix it while it is happening.

They don't draft the compliance record. They file it, and it survives the inspection.Food safety

We are small, we are growing fast, and everyone here is responsible for something that matters from their first month. The catalogue grows by about one agent a month, and the next ones belong to whoever joins next.

They don't recommend a campaign. They run it, and they report what it cost.Marketing
What we actually do30+ agents in production
Whole functions

An agent here takes over something that used to be somebody's week

Closing the month. Chasing the late payment. Counting what is in the warehouse. Filing the record an inspector will ask for. It runs on a schedule, without being asked twice.

Human gates

People approve the decisions that deserve a person

Every agent stops at gates we design into it, shows its work, and waits. The client approves, corrects or rejects, and the agent learns where its edge is.

Audit trail

Everything it touches leaves an audit trail

These agents move money, payroll and compliance records. Every action is logged, every number is traceable back to the document it came from, and every cost has a name on it.

The platform

And underneath them, the layer that makes many agents work as one

Our own harness, agents that hand work to each other, a shared cost ledger and evaluations that let us swap models in a command. That layer barely exists anywhere yet, and it is where the next few years of this company are.

How we work

Seven things we actually mean.

These are the ones we hire on, promote on, and part ways over.

Our valuesBarcelona, 2026

Extreme ownership

Responsabilidad extrema

We take responsibility for our work, for what it produces, and for the company around it. If you spot something broken, or something that could simply be better, you have the room to fix it and people around you who will help, and nobody asks whether it was your job. That starts in your first week.

The client is the standard

El listón lo pone el cliente

Every interaction a client has with us is the product, not the packaging around it. Quality compounds: a client who trusts what an agent did last month lets it do more this month. That is the whole growth model, and it is built one careful interaction at a time.

Fast and exact, never one without the other

Rápido y preciso

An idea on Monday ships on Thursday, and it ships finished. Startup pace is not permission to leave rough edges: whatever goes out has to feel effortless to the person on the other side, who never asked how it was built and should never have to care.

The best argument wins

Gana el mejor argumento

Ideas get judged on the argument, not on who made it or how long they have been here. Bring yours in your first week and expect a real answer, in the room, from whoever disagrees.

Build on the frontier

Siempre en la frontera

New models, new tools, new hardware. Trying something unproven is part of the job here, not a favour you have to ask for, and the time to do it is already in the week.

Correctness is the craft

Bien hecho, siempre

Our agents touch money, payroll and compliance records. Being right matters more here than being impressive, and we build so that being right is provable rather than promised.

Low ego, high standards

Buena gente, buen nivel

Direct about the work, warm with the people. We are demanding with each other and easy to be around, and we never confuse the two.

What we build with

We run on what shipped this year, not what was safe five years ago.

Nobody here is protecting a stack. Frontier models and open weights, hosted and local, bought and built in house, whatever is right for the job and the cost per run. If something better lands in October, the person closest to that domain moves to it in November.

In production and on the benchUpdated monthly
Frontier models
Claude Opus 5Claude Sonnet 5Claude Fable 5Claude Haiku 4.5Anthropic APIGemini 3.1 ProGemini 3.7 FlashGrok 4.6GPT-5.6Computer useExtended thinking
Open weights
Qwen 3.8 27BKimi K3Gemma 3 27BLlama 4Mistral LargeDeepSeek V3gpt-oss 120B4-bit quantizationLoRA fine-tunes
Agent tooling
Claude CodeCodexMCP serversTool callingStructured outputsPrompt cachingBatch APIsVector searchSub-agent orchestration
Built in house
Our own agent harnessInternal eval suiteLuna, our orchestratorPer-call cost ledgerInternal MCP serversAgent-to-agent messagingAgent action logs
Inference and hardware
NVIDIA DGXCUDAvLLMOllamaLocal inferenceOn-prem deploymentsFine-tuning rigs
Product stack
Next.js 15React 19TypeScriptTailwindFastAPIPython 3.12SupabasePostgreSQLVercelRailwayPlaywrightpytest
What we lean on every dayIn the stackOn the bench, being tested now
Where we work

Born in Chicago. Built in Barcelona. Two more on the map.

The company started in Chicago and grew up in Barcelona, which is home today. San Francisco is next, and Madrid is opening alongside it: both are being worked on at once, and the United States is where we are pushing hardest. The people who join now are the ones who will be asked to open them, spend a season there, or run a domain across both time zones.

Vermoon on the mapBorn in Chicago · Barcelona open · 2 planned
San FranciscoNEXTChicagoWHERE WE WERE BORNMadridPLANNEDBarcelonaHEADQUARTERS
Where we were bornOpen todayPlanned, and the people joining now are the ones who open them
ChicagoWhere it started

Vermoon was born here, and the first version of what is now the agent catalogue was written here. No office today, but it is the reason the United States is on the plan rather than an ambition somebody has to argue for.

BarcelonaHeadquarters

Two offices. Poblenou, fifteen minutes from the beach and eight from Glòries, one floor, a lot of screens and very good coffee within forty metres, plus a second one in Sant Cugat. Hybrid, three days on site, and seven of the eight open roles start here.

San FranciscoNext

The next one, and our first outside Spain. It is the reason one of the open roles is already based in the United States. Travel starts before the office does, and a stretch based there is on the table for anyone who wants it.

MadridPlanned

Our second Spanish office, opening alongside San Francisco as the deployment side of the company grows. Several roles will start in Barcelona and move, and whoever wants to open it will be asked rather than made to argue for it.

Applying

Three steps, ten minutes, and a person reads every one.

We are not a traditional company, and this is not a traditional application. We want to see what you have built: the side projects, the thing you made because it annoyed you, what you understand about how systems actually fit together. The CV comes along as collateral. What you can build, and what you would be like to build alongside, is the whole decision.

The applicationAbout 10 minutes
01

Upload your work

Your portfolio and your side projects first, your CV alongside them. A repo, a product, a dashboard, a shop you run, a campaign, a set of books you closed. Half finished counts. This is the part we open first, and the CV is the part we open last.

02

Answer four questions

What you built that you are proudest of and what made it hard, what value you would bring here, where you want to be based, and which languages you work in. Short answers are the right answers: we are reading for how you think, not for how well you write a cover letter.

03

Review and send

You see exactly what we will see before it goes. No cover letter, and nothing else to chase afterwards.

We hire on what you have built and how you are inside a team that is growing fast, never on a history of job titles. This is a small company that intends to stay a close one.
What happens next3 weeks end to end
01

We read it, within 5 days

A person goes through what you built. You get a yes or a no with the specific reason behind it, never silence.

02

We talk about the domain

Half an hour on what your domain does today, what is weak in it, and what you would change first. No panel loop, no culture screen.

03

A day with us

Spend a day with the team, see a real problem up close and watch how decisions actually get made here. It is as much your call as ours: you leave knowing exactly what you would be joining.

04

An answer, either way

An offer with the band, the scope and the start date in writing, or a no with the reason behind it.

No take-home puzzle that gets thrown away, at any step.

The promise: never more than three weeks from your first upload to a decision. If we go quiet on you, write again and you get an answer that day.

Don't see the right role?

We are selective, and we always make room for someone exceptional. Tell us which function you would take over, and what its agents should be doing that they are not.

Careers