ƒunc(it)
Takes your idea. Makes it work.
We build agents: programs that read, decide and act inside your tools. A team of two engineers who build them and keep them running in production.
What we do
It always starts with understanding the business.
Only then do we decide what the agent does — and it rarely looks like what anyone pictured at the start. A few shapes we've delivered or know how to deliver:
- Answer incoming requests and book the appointments
- Handle a case end to end, and hand it to a person when it gets stuck
- Follow up, qualify, and keep up to date what nobody has time to keep up to date
- Watch a system and act before anyone calls you
What makes an agent hold up in production
- The harness and the toolsWhat the agent is allowed to do. That's the main engineering work, well before the prompt.
- Access to dataConnecting your system, deciding what to expose and what not to. Agents fail on data, not on models.
- SecurityInjection, exfiltration, execution scope, sensitive data. The harness is the first vulnerability.
- Bench testingSynthetic and real cases, replayed before every change of model or prompt.
- EvaluationObserving is seeing. Evaluating is judging: scoring quality, comparing two versions, knowing whether a change improves or degrades.
- EconomicsDoes it earn more than it costs? Cost per task, model trade-offs, portability from one provider to another.
- Handing off to a personWhen the agent hands over, to whom, and with what context.
- Progressive rolloutIn observation first, then on a fraction of traffic. We switch it on when the numbers say so.
An agent doesn't live on its own: it needs an application, an API, an infrastructure. We deliver the whole chain, not just the model.
Already done, and kept running for three years
Around ten agents in production on WhatsApp and Instagram, in Latin America, for hotels, clinics and hair salons. Three unrelated businesses — it's the method that carries over, not the recipe.
Try it
The team
Karim Benhammou
Product Engineer · LLM agents in production
- Go, distributed systems, APIs
- LLM agents in production, API security
- Flutter iOS and Android · AWS and Hetzner deployment
- French, English, Spanish
Cédric Zalewski
Product Engineer · full-stack
- TypeScript, React, Node, Tailwind
- Product and UX, observability, incident response
- Very high volumes — a platform serving millions of users
- French, English
Both went through École 42 in Paris, both practise AI-assisted development daily. The team shows up whole: every architecture decision goes through two sets of eyes — what holds in production, and what teams actually adopt.
How it goes
- Scoping.
A conversation about your business: what repeats, what gets stuck, and what it costs you today.
- Phase 1, to judge on evidence.
A bounded scope, delivered to production on your side. You decide what's next with a result in front of you, not a promise.
- Rollout.
We extend to the next task, measure, and hand over to your teams.
Let's talk about your business, and the first agent to build.
Tell us what repeats at your company. An email, or thirty minutes on a call at a time that suits you.