·November 12, 2026·5 min readKresco Case Study
A founder-led team of six people put 2,009 tasks through the swarm in its first 21 days, 65.6% of them raised in Slack, and moved the website, project management, operations and the demo pipeline onto it.
Data window: the install's first 21 days, 22 October to 12 November 2026.
Kresco is a Barcelona-based company that builds AI agents that sell and support customers over WhatsApp for D2C brands and retailers. It also runs its clients' growth data, so anyone on their team can ask a question about the business in plain language and get the answer from their own data.
The company is founder-led. Iñaki Mediavilla and David Martín-Borregón, co-founders of Kresco, have run it from Barcelona since August 2024. Its clients are top players in Spain: Multiópticas, Olistic, CUBRO and Natuka.
A six-person team running sales and support for those brands.
Three weeks, one install
On 22 October 2026 the team sat down for a two-hour setup session in Barcelona. They left with an install running, and 48 tasks in the works.
By the fourth week they were running nearly 7x the tasks a day of their first days.
Agents went from 4 to 12 on 7 November. Slack is the front door: 65.6% of all tasks started from a Slack message. 1,316 of the 2,009 carry a parent task, work the swarm split and handed to a worker, and 60 workflow runs finished out of 60 started.
What they run on it
Seven things, in David's words. The list is the whole surface of the company: marketing site, project management, operations, demos, email and chat triage, and the configuration that used to need an engineer.
The website. "All of the work on our website, straight from me and Iñaki."
Project management. "Project management automations: tasks created automatically from client inputs, filled in, given context, and so on."
Operations. "We speed up feedback: it is classified and assigned automatically, and in some cases solved directly, which speeds up continuous improvement."
Debugging operations. "We have a demo-creation process that is automatic, but it does not always run smoothly. Every time it fails, it looks at what failed and tries to fix it in the short and the long term."
Email and chat triage. "We gave it access to email and Google Chat, and by forwarding emails and reading chats it keeps warning us when there are relevant things."
Business questions. "Answering business questions about technology without going through engineers."
Configuration, with an engineer as goalkeeper. "Changing configurations that still live in the repository, which someone from business can now ask for directly, with the engineer only as goalkeeper. Both changes to production things and demo configurations for prospective clients."
Engineering got back focus time
Read that list again and notice what is missing from it. No codebase work, no architecture, no product. David is precise about the boundary: "Deep engineering stays where it was."
What Engineering gained was time. "All the noise the engineers were facing, I think we are managing to make it disappear, while the business people get a kind of self-programming n8n that builds automations for them and makes their lives easier."
A self-programming n8n is a different shape of tool from the workflow builders most teams run. A visual builder automates what someone can draw. Kresco's version takes the request in words, from the person who has the problem, and builds the thing: a task, a demo configuration, a website change, a fix to a process that just broke. The engineer's job moves from writing it to signing off on it, which is what the goalkeeper is for.
Three weeks, one install, twelve agents, and a team that no longer queues behind its engineers for the small things. If you want to see what your own operations look like as a swarm, it starts at agent-swarm.dev.