(CONT'D

Many Minds, One Great Work: What a Cathedral and a Moon Landing Teach Teams Working With AI Agents

Great works are built by many people who never share a room. What lets them finish is shared context: one current plan, decisions written down before they are needed, and work that changes hands with its context. Teams working with AI agents need the same three things.

Our new homepage tells a short story in four scenes: a fire on the savanna about 70,000 years ago, Cologne Cathedral, the night before Apollo 11, and San Francisco this morning.

It is about an old idea. When many people make one thing, what holds the work together is not that they share a room. It is what they share.

That idea has become practical again, because the people we work with now include agents.

One current plan

Construction of Cologne Cathedral began in 1248. Work stopped around 1473, and a wooden crane stood on the unfinished south tower for about four hundred years.

In the nineteenth century the medieval drawing of the west front was found again, one part in 1814 and another in 1816. Building resumed in 1842, and the cathedral was finished in 1880, following the medieval design.

Builders who never met, six centuries apart, finished one design. They could, because there was one plan, and it was the one everyone followed.

Cologne Cathedral at dusk with the outline of its finished west front drawn in gold lines over the unfinished building
The cathedral’s west front, drawn in gold over the unfinished building. From the stagecontinuity.com homepage.

Agent work produces plans quickly. A task gets a plan, then a better plan, then a third after a test fails. Every version stays in the project. The risk is not losing the plan. It is continuing from the wrong one.

Write down what it means, before you need it

Minutes before Apollo 11 landed, the lunar module’s guidance computer raised a 1202 alarm, then a 1201. There was no time to work out what they meant.

Mission Control had already done that work. In simulations, the team had gone through the computer’s alarm codes and written down which ones were safe to fly through. With that list in hand, the call was “go.”

The Saturn V on its pad at night across a lagoon, with the handwritten lines 1201 GO and 1202 GO in gold above the water
1201 and 1202: two alarms that were understood before they happened. From the stagecontinuity.com homepage.

In agent work, the equivalent is the note that explains a decision: why a force limit changed, why a dependency was pinned, why the second plan replaced the first. Written down while it is fresh, it lets the next session, or the next person, act in seconds instead of working it out again.

Today, the many minds include agents

The last scene of the homepage is San Francisco at dawn, with windows lighting up across the city. Behind each one, someone is starting something, and more of them work with agents every month: one in a terminal, another in a second session, a third on a different machine.

San Francisco at dawn seen from above, with thousands of windows lit across downtown and the hills
Today: behind every window, someone is starting the next great work.

It is the oldest pattern at a new scale. The work is split across more minds than ever, and some of them begin every session with no memory of the last one. The three habits still hold:

  • Keep one current plan, and say what replaced what.
  • Write down what decisions mean before they are needed.
  • When work changes hands, hand over its context with it.

Where Continued fits

We built Continued, a Mac app, around those habits for people working with Claude Code and Codex.

  • Every agent’s work in one place. Sessions side by side, and who is waiting on you, on your Mac or your phone.
  • Say “continue,” and the right plan comes back. Continued tracks which plan replaced which.
  • Every file knows who made it. Which agent, which session, and when.
  • Hand off with the context. Move the work to another Mac or a teammate, and the context goes with it.
A person seen from behind at a desk by a tall window at dawn, a laptop open in front of her and the lit city outside
One of the windows.

The habits come first; the app makes them the default. Two practical guides go deeper: how to continue an agent’s work without losing the plan and how to hand off agent work with its context.

See the story on our homepage, or download Continued for Mac.

Shared context FAQ

What does shared context mean for AI agents?

It is the part of a project every person and agent works from: the current plan, the decisions behind it, the files that matter and what has already been done. An agent that starts without it rebuilds it from guesses.

Why do agents restart from the wrong plan?

Plans change quickly in agent work, and older versions stay in the project. A new session searches for something plausible and can pick a version that was already replaced, unless the project records which plan is current.

Do I need special software to apply these ideas?

No. One current plan per task, a written note of what each decision means, and handoffs that include a summary and the relevant files work with any tools. Continued makes those habits the default for Claude Code and Codex sessions on a Mac.