HSLabHSLab · open AI lab
PT
← All lines
Research line 01EH · Running

Harness for hybrid teams

Question

What has to exist around a model so humans and agents can work on the same team, with trust?

Thesis

The model sets the ceiling of intelligence. The harness — context, tools, loop and governance — decides how much of that ceiling reaches real work.

12 experiments0 articlesagent-builder (OrganIA platform) · Squad Híbrido

The most mature line in the lab. It comes from building the OrganIA platform (agent-builder) and Squad Híbrido: agents working alongside people, with shared memory, live dashboards and governance.

What I am investigating

  • Harness engineering: why production failures are rarely the model’s fault.
  • Context engineering: context as a budget, not a warehouse.
  • Memory: each person’s second brain and the company brain, without copying data.
  • Interface: the agent’s answer as an operable dashboard (Live Apps), and configuring the platform by talking (No Brain Mode).
  • Evolution under governance: the agent that improves with use without changing on its own.
Experiments in this lineEH-01When an agent fails, the model is rarely the cause17.3 → 8.4 minRunning16.10.2026EH-02Organizing context beats switching models76%Running08.10.2026EH-03Setting up the whole platform by chat214Running20.09.2026EH-04The agent’s answer becomes a dashboard21Running01.10.2026EH-05Matching the same customer across systems0.90Running05.09.2026EH-06One memory per person, shared by every agent13%Running12.10.2026EH-07The agent suggests improvements and changes only when approved94%Running16.10.2026EH-08AI risk, privacy and cost in one place1 clickRunning31.10.2026EH-09An agent that uses systems with no integration13Running20.07.2026EH-10A bot that joins Meet and records the meetingSilence.Done07.10.2026EH-11What a better model really costsUS$ 28Done06.08.2026EH-12The internal agent serving the websiteAG-UI · Widget · TurnstileRunning01.11.2026
On the radar