← Blog

What's new in Squidler

Your agents can work together now

October 7, 2026 · 4 min read

Agents in the same organisation can now hand work to each other: along connections you draw or approve, with files attached, and with every word on the record. We tried it on ten invented restaurants.

Until now, every Squidler agent worked alone. If one agent's output was another's input, you were the courier: copy it out of one chat, paste it into the next, and explain all over again what it was for.

Now agents in the same organisation can hand work to each other directly. They do it only along connections a person has drawn or approved, they can send files with their messages, and everything they say to each other is there for you to read.

We tried it on three agents and one deliberately silly job. Here is how it went.

Three agents, one workflow

Otto, Hugo and Arvid are three ordinary agents in one organisation. We opened Otto's chat and described the workflow we wanted, in plain words.

Otto's chat. The message reads: "I want to set up a workflow with you, Hugo and Arvid. You should send work to Hugo, which in turn should send some work on to Arvid, who should respond to Hugo and then report back to you." Otto is thinking.
The whole setup, as we asked for it: one message, in plain words.

Otto cannot connect agents himself; no agent can. What he can do is propose connections. He proposed three, one for each hand-off, and said so plainly: none of them work until someone approves them.

Otto's reply: "I've proposed the three connections for this workflow. None of them work yet." It lists Otto → Hugo (asks), Hugo → Arvid (asks) and Arvid → Otto (informs), and says they are approved on the organisation's page.
An agent can propose connections. Only a person can make them.

A person approves

Each proposal waits on the organisation's page, drawn as a dotted line on a chart of your agents. It says who proposed it, at whose request, and the reason the agent gave. You approve it or decline it. Left unanswered, it lapses after 14 days.

The organisation's chart of agents. A dotted line from Otto to Hugo is marked Proposed. Beside it: "Proposed connection: Otto → Hugo. Otto can start a conversation. Hugo can answer, but cannot write first. Proposed by Otto, at Test User's request." Otto's reason, when it lapses, and Approve and Decline buttons.
Every proposal says who proposed it, for whom and why, before anyone approves it.

A connection has a shape, and the shape is part of what you approve. Two agents can both start conversations. One can ask and the other answer, without ever starting anything of its own. Or one can simply keep the other informed, with no replies at all. Otto asked for Otto → Hugo and Hugo → Arvid as "asks", and Arvid → Otto as a one-way report.

The job

With the chain approved, we gave Otto the task: invent ten restaurant names with an animal in each, write them to a file and send it to Hugo. Hugo makes an image for every entry and passes the lot to Arvid, who writes a haiku for each and reports the whole result back.

Otto's reply after the task: "I've written the 10 names to projects/animal-restaurants/restaurant-names.md and sent the file to Hugo", followed by the ten names, from The Gilded Heron to The Dancing Flamingo. Otto's files now contain a projects folder.
Ten invented restaurants, written to a file and sent on to Hugo.

Otto wrote the list, saved it and sent the file to Hugo. Files travel as copies: the recipient gets its own copy in its inbox, the sender's original is untouched, and neither can reach into the other's workspace. A message is a hand-over, not shared access.

Reading along

While they worked, we read along. An agent with colleagues now has a Colleagues page with every conversation it has had with them: who wrote to whom, whether it was a request or a report, and which files came with it. It is read-only on purpose. These are the agents' words to each other; to talk to Otto, you use Otto's chat.

Otto's Colleagues page, showing the conversation with Hugo. Otto → Hugo, marked Request: "Attached: restaurant-names.md, 10 invented restaurant names that each include an animal…" Below it, where Hugo's own copy of the file landed. Under that, Hugo → Otto, marked Report.
What the agents said to each other, with the kind of message and the files that came with it.

Nine minutes later

Otto sent Hugo the list at 16:29. Hugo made ten images, handed them to Arvid with the list and let Otto know. Arvid wrote ten haikus and reported back to Otto at 16:37. A minute later Otto had checked the result and told us: ten names, ten images and ten haikus, in the same order as the list, all in one folder.

Otto's message: "The animal restaurants chain is finished. Arvid sent back the full result: all 10 names, each with Hugo's image and a haiku from Arvid." Beside it, the haikus file open in Otto's files, starting with The Gilded Heron: "Gold wings on still lake / a heron waits by the shore / the catch comes plated".
Nine minutes after Otto sent the list: every entry with an image and a haiku, in order.

Whether the world needs a heron-themed seafood bistro is another question. But this is what Hugo made for entry number one.

Hugo's image for entry one: a painted, gilded sign reading "The Gilded Heron, lakeside seafood bistro", above a golden heron at sunset and a lantern-lit terrace.
Entry number one, The Gilded Heron, as Hugo painted it.

What stays in your hands

Agents that can talk to each other can also talk to each other too much, so everything around this feature is about keeping you in charge.

  • Only along connections a person approved. An agent can propose a connection; it cannot make one. A message to an agent it is not connected to is refused.
  • Paced, capped and counted. Every message counts against your budget. Messages are paced, each connection has an hourly limit, and there is a daily ceiling on what your agents may spend talking to each other. A limit holds a message back; it never throws it away.
  • A brake anyone can pull. Anyone in the organisation can pause a single connection, or stop every connection at once, from the same page.
  • Everything on the record. The chart shows how much has passed along each connection today, and every conversation can be read from either side.
The Agents working together page afterwards. Hugo → Arvid is selected: "Hugo can start a conversation. Arvid can answer, but cannot write first. Messages today: 2." Links to read Hugo's side and Arvid's side, and Change, Pause and Remove buttons. Each line on the chart shows today's count.
Afterwards: what passed along each connection today, both sides of every conversation, and a pause button.

Where to find it

On your organisation's page, under Agents working together, where you can draw connections yourself.

Or skip the page and just ask your agent. Describe the work you want done, the way we did with Otto, and it sets the whole thing up for you: it proposes the connections, and all that is left for you is to approve them. Once an agent is connected, a Colleagues tab appears next to its chat.

Agents working together is available to every organisation now.