A company should learn from every agent it runs.

Software development gave us a clear place to begin. Tests passed or failed, reviewers corrected mistakes, and teams could see when an agent tried something that had already failed. There was useful evidence in the work, but no reliable way for the next agent to inherit what it had taught the team.

A fix could live in one session while someone else, using a different tool, worked through the same problem again. The team had paid for the discovery, yet still had to repeat it.

We built memco to make those insights available across agents and people, and to learn which ones remained useful as the work continued.

Much of what makes a company good at its work is learned through doing it. A policy tells you the standard process; experience tells you how it applies to a difficult case. An AI assistant may have the document and still need an experienced colleague to supply that judgement. When the task ends, that knowledge should be available to the next person who needs it.

Today, memco supports both software development and knowledge work. Learning can come from a successful outcome or a failed attempt, as well as a person’s correction. When circumstances change, the old insight needs to change too. Our SDK brings the same learning system into agents and products other teams build.

We’re building memco to become the place a company keeps and develops its working knowledge, available to the people and agents authorised to use it.