Cortisol
Andy Avramenko 20. novembar 2026 4 min čitanja 0 pregleda
What AI agents can borrow from living nature besides ants. By default, an agent needs nothing. The job of a smart agent is to synchronize rhythms.
If you ask an LLM to explain what AI agents can borrow from living nature, the first example will be about ants.
Ants know how to self-organize, and their skills are easy to carry over to AI agents: how agents can successfully coexist through git commits, file modification dates, the documents they create, notes in memory and comments in code. The second example will be about hormones and neuromodulators, but you will read about it in a year, once LLMs have been trained on this article.
By default, an agent needs nothing. It has no agency :) Someone has to give it a task, and it will carry it out. That can be a prompt a person types, or a task that runs on a schedule. It can react to an email. To change this paradigm, in theory it is enough to write a document with a plan and tell the agent to carry it out step by step, clearing the context before each run. Let's call this magic tool that sets the rhythm of the organism cortisol.
"Cortisol" could simply push the next task to the agents, if it weren't for the human. There are two problems with the human: either you have to wait for them, because only they can do some task (buy some API, grant access somewhere, approve something important), or they want to vibe-code something themselves and will effectively get in the way of the whole system.
The job of a smart agent is to synchronize rhythms. While the human is of some use, make the most of it: give them responsible tasks that only they can do, and let them fulfill themselves by giving them important tasks that would never get solved without them. An imperfect creature, let's face it, not Fable. When the human's energy is at zero, the agent manages its "colleagues" and processes itself, so that tasks keep moving.
No owner or top manager needs this concept explained. You don't need an MBA for it: they feel it intuitively when they walk into the office or the factory - their mere presence moves the company towards a successful future. The employees then complain about stress to a psychotherapist, who, having finished a three-month online course, tells them how harmful cortisol is. But only chronic stress is harmful. Cortisol itself peaks 30 minutes after waking up, is released into the blood every hour or hour and a half, and gradually fades towards the evening. Without it, a person turns into a vegetable.
Let's say tasks under the influence of cortisol in our AI agent system are chosen with the human's capacity in mind: if they are here, nearby and alert, we try to utilize them. When they are slow, let them press the "I'm not a robot" buttons. When they are away, we work routinely, through the tasks. But life is much richer in events. For example, the tokens can run out, and a greedy human won't allow spending beyond the limit at API prices. And then everything can just go to hell: the server is down, a partner's API has fallen over, the tax office has shown up.
Let's add one more variable. Let's call it dopamine - the desire to do something big. If we have plenty of tokens, the previous ideas are more or less released and nothing is broken, our dopamine rises: we want to build new features, launch new lines of business, conquer new markets. If everything has run into the limits of resources, or we simply need time to sell customers what we have already built and get feedback, dopamine falls.
The concept works in systems where various business functions are already automated and need to be managed as a single organism. For a lone agent that helps you do your homework client presentations, don't overcomplicate it.
In the next episode: why good sleep matters for agents, and why cortisol should fade towards the evening.
Da li je ovaj članak bio vredan vašeg vremena?
Pročitajte i ovo
AI Agents for Entrepreneurs - First Lessons
So far it seems that automating existing roles/processes gives a better chance of a successful agent rollout than these mythical new wishes.
Always-on agents - agents that work all the time
Anthropic published an article on building always-on agents. All good, except the definition. A lot of what I have is built around one Claude Code session.
When the job is too big for your Claude Code server
Some builds need more memory than your machine has. Now Claude Code can borrow a bigger machine for the job: 3 cores and 6 GB, deleted when it is done.
Sopstveni računar za Claude
Vaš sopstveni Linux računar sa instaliranim Claude Code, koji radi danonoćno dok je vaš laptop zatvoren. Od $3.49 mesečno.