Case study · Inside BL Diseno
Our best case study is our own studio.
Before we propose an agent to you, we use it ourselves. BL Diseno runs every day on an AI agent system built in-house: shared memory, automatic reports, prospecting, research with verification. This is what we’ve learned, mistakes included.
Talk about your caseIn short
BL Diseno runs on its own AI agent system: a knowledge base that every agent reads and updates, a daily report sent to a phone, a prospecting pipeline that produces tailored profiles and demos, and research agents that check their sources. What makes it reliable isn’t the AI model but the rules: proof, limits and human control.
The systems
Shared memory
All the studio’s context (projects, decisions, tasks) lives in a knowledge base that each agent reads when it starts and updates when it finishes. No session starts from scratch, and every decision has a date and an author.
The morning report
Every day, an agent reads the email and calendar, matches what has come in against open tasks and sends a prioritised summary to a phone that takes thirty seconds to read.
Automated prospecting
A seven-step pipeline collects businesses from public sources, removes duplicates, spots those without a website and prepares tailored profiles, demos and messages. On its first run it turned 226 rows into 155 unique profiles, with one demo per profile.
Verified research
For legal or technical questions, one agent searches official sources and another checks that every quotation appears word for word in the original text. In a tax law session, all 50 quotations were found verbatim in the consolidated text of the BOE, Spain’s official state gazette.
A task runner with limits
An agent picks from the task list the ones it can do alone, with exclusions written into the code: nothing that involves sending, paying, deleting or deciding on someone’s behalf.
A weekly auditor
Every week, another agent reviews the knowledge base, spots anything disorganised or unlinked and checks the health of the whole system.
The rules that make it reliable
Measured or declared
Every closed task states how it was checked: “measured” if someone looked at the result (a file, a message, a dated output), “declared” if it was closed on someone’s word. A system that says “done” without proof doesn’t count.
Four tests to be “in service”
An automation only counts as in production if it’s in the register, leaves an observable, dated output, has a plan for when it fails and is covered by the health check. We learned that a scheduled task’s “completed successfully” status proves nothing.
Archive before deleting
Nothing is removed without a full copy and without checking that no link sent to a third party points to it. The rule came from one of our own mistakes: a link shared with a client stopped working.
The human decides
Agents propose rules; a person approves them. Irreversible actions always require explicit confirmation.
What it means for your project
The same rules apply to the systems we build for you: written limits, traceability, verifiable outputs and a person in charge. And we know the typical failures of AI agents because we’ve been through them first, in-house.
It also explains how we can cover so much ground with a single point of contact: much of the preparation, research and verification is done by agents, and human time goes into understanding your problem and making decisions.
Frequently asked questions
What is an AI agent system?
A set of programs built on language models that work in coordination: each has a task (summarising, researching, verifying, organising), they share a common memory and follow rules on what they can do alone and what needs human sign-off.
Can you build a system like this for my company?
Yes, scaled to your size. We usually start with a single piece with a clear return (the daily report, email sorting or an internal search tool) and add the others once the first has proved its value.
Do the agents act without supervision?
Only within limits written into the code. Anything irreversible (sending to a third party, paying, deleting, deciding on someone’s behalf) requires human confirmation, and every action is logged.
Keep reading
AI agency
AI studio in Barcelona: AI agents, process automation and AI audits for businesses. Custom projects, fixed price up front and everything in your name.
AI agents
AI agents that answer your clients, qualify leads, draft replies and search your documents, on the web, by email or WhatsApp. With human oversight and GDPR.
About BL Diseno
BL Diseno is Laszlo Brusa’s technical studio in Barcelona: websites, hosting, automation and AI agents for businesses, with a single point of contact.
How to implement AI in a business: 7 realistic steps for SMEs and professional firms
A realistic path to implementing AI in an SME or professional firm: find repetitive work, measure it, pick one pilot, check GDPR, build, measure and scale.
Let’s talk
Tell us what you need. We reply within one business day.