# People and agents learn together

Made by Many, Agentic Playbook. By Andrew Flett, Made by Many.
Status: draft. Published: 2026-09-02. Last reviewed: 2026-10-05.
Canonical: https://agentic.madebymany.com/principles/people-and-agents-learn-and-grow-together

Every project teaches us something. For years we've captured this in retros, but not well: notes written from memory weeks later, filed somewhere, and the next project starting from scratch. Agents give us a way to fix this. A lesson can now change something while the work is still going: the [skills](/making) our agents work from, our shared [knowledge](/knowledge), or the [learning programme](/learning).

## Where a lesson goes

If it's about how an agent handled something, it becomes a rule in that agent's skill. If it changes what we think, it goes into the shared knowledge. If only a person could have used it, it goes into the learning programme.

```loop
{
  "title": "An agentic learning loop",
  "nodes": [
    {
      "label": "Projects",
      "note": "Any work an agent helps with."
    },
    {
      "label": "Capture the lesson",
      "note": "The agent notes it as it happens.",
      "accent": true
    },
    {
      "label": "Agentic skills",
      "note": "The agent makes the fix, a person reviews it."
    },
    {
      "label": "Human training",
      "note": "It becomes part of what we teach."
    }
  ]
}
```

## It runs while the work happens

We got this wrong to start with. Our own log grew to thirty lessons, each naming the skill it should change, and not one of those skills had changed. Writing lessons down was never the hard part. What's difficult is acting on them.

Now the agent we're working with does a lot of it. When something goes wrong, surprises us or works better than expected, it writes the lesson down there and then. That goes for a pitch or a workshop as much as a build. It says where the lesson belongs, makes small fixes itself, and files anything bigger as work to do. A lesson isn't finished until its change is made or on the list. [`mxs-learn`](/skill/mxs-learn) is the skill that does this.

## People decide which lessons are real

People do the part that needs judgement. They review every change before it lands, and they tell a real pattern from one bad afternoon. An agent doesn't rewrite how we work on the strength of a single run.

## The people grow too

Agents are only as good as the person steering them. They make someone with [taste](/principles/humans-bring-taste) much better, and everyone else faster, which isn't the same thing. Growing that taste is why we put more into the [learning programme](/learning) than anything else. It's also what lets one person with agents do work that used to need a bigger team: [designers who build](/knowledge/designers-who-build)  ship the thing itself, not a picture of it.

## Related agentic skills

- [mxs-learn](https://agentic.madebymany.com/skill/mxs-learn)