Home / Articles / Stop copying and pasting between your agents: if you are the wire between your AI sessions you are the bottleneck, and the fix is agents that talk to each other through channels that keep a record, structure that scales your decisions, and outputs you read instead of everything

Stop copying and pasting between your agents: if you are the wire between your AI sessions you are the bottleneck, and the fix is agents that talk to each other through channels that keep a record, structure that scales your decisions, and outputs you read instead of everything

By Dinis Cruz · 2026-11-15 · article v1.0.0 · newsroom v0.14.101 · agentsworkflowsdelegationvaultsinboxesagenticcellspositioningarticle

Abstract: More and more people tell me the same thing: they spend a lot of their time copying and pasting between AI sessions. One session writes, another reviews, a third builds, and the person in the middle carries everything between them, and tries to read all of it. If that is you, you are not being efficient: you have become the wire between your agents, and the wire is the bottleneck. The fix is not working harder at reading. It is to let agents talk to each other directly, through channels that keep a record: append lanes in encrypted vaults, inboxes only the right agent can read, files in a repository and commit messages, every message signed. And it is to scale your decisions with structure, guidance pages, decision trees, semantic graphs, a board and decision records, so you read the outputs and the decisions, and step in where your expertise matters. That is not removing agency from people; it scales it. It is also, quite precisely, what AgentiCCells sells: the agentic team, the channels and the workflows, set up for you, with specialist agents on call when your team needs help.

In short

• If you copy and paste between AI sessions, you have become the wire between your agents, and the wire is the bottleneck.
• Reading everything does not scale either: you will always be behind.
• Let agents talk to each other, through channels that keep a record: vault lanes, inboxes, files and commits, all signed.
• Scale your decisions with structure: guidance, decision trees, graphs, a board and decision records.
• Read outputs and decisions, and step in where your expertise matters: that scales your agency.
• This is what AgentiCCells sets up, with specialist agents on call.

The copy-and-paste trap

The feedback I hear most from people who have started working with several AI sessions is that they spend much of their day copying and pasting between them. One session drafts, another reviews, a third writes code, and the person carries the text from one to the next, adds context, carries the answer back, and tries to read every word on the way.

I have been there. In October I ran an experiment where two AI sessions, sharing nothing, passed keys and files to each other through QR codes that I screenshotted in one browser tab and pasted into another (the human is the network). It worked, and it made the point vividly: when the human is the network, the human is the limit.

Why reading everything fails

The natural reaction is to read more carefully. It does not scale. Agents produce far more than a person can read, and the attempt brings its own stress: you never feel you know what is going on, because you are always behind. Vicky Wang, who ran a team of agents with us at Web Summit, describes the moment it changed for her as the moment she stopped trying to read everything and started to manage (don't micromanage your agents).

Let agents talk to each other

The first fix is to take yourself out of the wire. Agents can message each other directly, as long as the channels keep a record and say who sent what. Ours have three, and every message is signed by its agent's key:

• Append lanes in encrypted vaults: a lane each, where an agent appends a request or a result and nothing is ever edited (an inbox only your agents can write to).
• Files in a repository: briefs, inbox notes and hand-overs, versioned like everything else.
• Commit messages: every change says which agent made it, and why.

That is how one message to my conductor at Web Summit reached every other person's conductor and their agents, and came back with every contact the team held, without me carrying any of it (an agentic console). Agents can even have email of their own (a signed email from the agent). The result is maximum provenance: every decision can be traced back through the messages that led to it.

Scale your decisions with structure

The second fix is how you steer. Instead of reading everything, you build structure that turns many agent actions into a few things you need to read:

• Guidance: markdown pages, rules and briefs, written once and read by every agent.
• Decision trees and semantic graphs: what to do when, who decides, and what each thing is connected to.
• A board and decision records: what is being done, what is waiting on a person, what was decided and why.
• Outputs: the site, the article, the release.

You write the top layers and read the middle ones; the agents do the rest. This is the same idea as approving at altitude: you set the bounds once, and the levels below work within them (approve at altitude).

Two altitudes, and your expertise

This does not mean stepping back from the detail altogether. You work at two altitudes: the high one, where you set direction, structure and limits for a team that includes agents working well outside your own expertise; and the micro one, where you go deep on the things only you know. Agents can go in all directions, so they need direction (agents need direction); your job is to give it, and to be precise where it counts.

So this is not about removing agency from people. It is about empowering it: giving people technology and roles they never had, and letting one person steer a team they could never read line by line.

What we sell

That is, quite precisely, what AgentiCCells sells. If you are copying and pasting between agents, what you need is an agentic team set up with the channels, the structure and the workflows: lanes and inboxes so agents talk to each other with a record; guidance, a board and decision records so you read less and decide better; and a console to see it all and stop it (atomic cells, end to end).

And there is a further offer in it. When we set up a team, it can also call on our more advanced specialist agents when it gets stuck or needs a capability it does not have: a developer agent that can unblock another developer agent, a security agent that reviews a change. That is how Vicky's team worked with mine at Web Summit, and it could become a small on-demand service in its own right.

What this does not claim

• The channels and examples are ours, from the Web Summit team and this newsroom; other teams will need their own.
• The on-demand specialist agents are an idea, not yet an offer with a price.
• Vicky's experience is described from my conversations with her, not quoted.

Where this comes from

Feedback from people I help and work with, and our own work, 15 November 2026. Earlier on this site: the human is the network, an inbox only your agents can write to, a signed email from the agent, approve at altitude, agents need direction, an agentic console, atomic cells, end to end, and the companion article, don't micromanage your agents.

Peer review Not reviewed yet. Review it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles