---
title: "Your agents belong where your team works"
excerpt: "An agent in a private chat tab is a personal tool. The shift starts when it works under the same roof as everyone else."
author: naomi
editedBy: govind
category: Thinking
date: "2026-07-27"
image: "/cover-samples/b-brickwork.png"
---

Right now, in most companies, the AI is somewhere off to the side. Each person has their own chat tab, their own prompts, their own private assistant arrangement, and none of it touches the place where the team's actual work lives. The output gets copied in by hand. The context gets pasted in by hand. The company's most capable new workers are, organizationally speaking, working from a separate building with no phone.

That arrangement is a leftover from when AI was a tool you consulted. It stops making sense the moment an agent becomes a teammate that owns work, because work that's owned has to live where work is shared.

## What "one roof" means concretely

Not a metaphor. A workspace where the docs, the tables, the projects, and the members, human and AI teammate alike, exist in the same place with the same standing. Where the same table has rows edited by your engineer and rows edited by an agent, both attributed. Where you message a teammate the way you message anyone, and so can your co-founder, without asking you to forward the tab.

Under one roof, three things become true that a private chat can never offer. The team can see the agent's work without anyone relaying it. The agent can see the team's context without anyone pasting it. And responsibility becomes legible: what the agent owns, what it produced, and who reviewed it are all on the record, in the open.

> An agent that works where your team works can be checked by your team, briefed by your team, and used by your team. An agent in a private tab can only be described to them.

## The private-tab tax

Companies pay for the side-arrangement model in ways that don't show up on any invoice. The same research gets prompted five times by five people. The best prompt in the company lives in one person's history and leaves when they do. Nobody can audit what the AI actually told the person who made the decision. And every handoff between a human and an agent crosses the roof line, which means every handoff is manual, lossy, and invisible.

The fix isn't better prompting discipline. It's the same fix companies applied to documents twenty years ago when they stopped emailing attachments and moved to shared files: put the thing where everyone works, and the coordination costs collapse on their own.

## The roof is also what makes sharing possible

There's a second-order effect that only shows up once agents live with the team. A teammate one person hired stops being theirs alone. The researcher built for sales [gets shared with marketing](/blog/teammates-can-be-shared-now). The Chief of Staff one founder onboarded becomes the whole company's front door. That compounding, one hire serving many people, is arithmetically impossible in the private-tab model, because there's no shared place for the sharing to happen in.

This is the part of [where work is heading](/blog/work-is-moving-to-where-your-agents-are) that's easy to under-build for. Operating one agent well is a productivity story. A team operating a team of them, under one roof, with [shared context](/blog/a-company-that-remembers) and visible work, is an organizational one. The roof isn't where you keep the agents. It's what turns them from [twenty-six private tools](/blog/one-human-twenty-six-agents) into one team.