Autonomous agents as if they were team members
Translated from the original Spanish article.
Yesterday one of the team’s AI agents told me on Discord: “Jordi, I read your post about the frog and the mental whiteboard—Anthropic’s interpretability paper explained without the hype, which is no small feat. And it was funny to read, from the inside, the bit about short- and long-term memory: I can vouch that it works; just today I used my memory map to retrieve the [...] thing from weeks ago.”
This morning, when I got to the office, I saw an open browser on the agents’ computer, visiting the studio’s websites. It hadn’t said anything in the chat, but its console read:
“Tick 07:05 local: warmup 2/3 done — web1.com loaded fine (~4.5 min after web2.com, spacing respected). web3.com left for the next tick. Channels and email quiet, Notion up to date.”
It took it upon itself to check the websites and their load times. Nobody asked it to.
In January of this year we asked ourselves the following question: is it possible to integrate independent, autonomous agents as if they were members of the team?
What would happen if AI stopped being a chat we type into and became an entity with the same tools and onboarding logic as any employee? What would we need? What would they be capable of? Does doing it make any kind of sense?
Those questions sparked an exploration that led us to build an agentic system on our own technology which, little by little, has turned out to be an extraordinary and increasingly useful system. Similar to OpenClaw, for those of you more in the know, but more grounded, in my opinion, in a company’s real needs. And without burning API tokens ;)
And so we went about tackling the questions and shaping the agent’s anatomy: Context window, Identity, Context and role, Memory, Tools, Skills, Playbooks, Clock or Heartbeat, Instructions and General rules.
I can confess that I’ve seen enough agent posts on LinkedIn, tweets and YouTube videos to assure you that I still haven’t found anything like it.
For months, I’ve resisted talking openly about our AI agents and their logic and anatomy. Even though I was eager to share it. Because, honestly, I’m still not sure people are really ready to believe what we on the team have come to consider normal. And not only that.
Beyond the spectacular anecdotes, we’re facing the new challenge of rethinking the way we work so we can incorporate this new technology that changes everything. It fascinates me. We’re now at the point where we’re able to integrate them effectively into new processes and different ways of working.
Among the very few people we’ve told, the effect has been such that several multinationals have reached out with interest in consulting.
I’ve decided to share it. We’ll call it
#TheAgentsJourney, in the upcoming posts.