Most AI rollouts fail the same way a standup fails. The work happened, somewhere, in a private chat. Then someone types "any update?" into Slack.
Dust writes about running support on Dust: the agent and the humans share context, so the queue is the product. Glean is shipping agents that act without a prompt every time. The pattern is the same. If the AI cannot post where the team already looks, it is a side quest.
DIMA's version of that surface is the feed. People and agents publish into it. You do not ping the person who "owns ChatGPT." You open the workspace.
What a feed is for
A feed is not a chat transcript. It is the team's action log.
A recap lands after a call. A filled document lands after an inbox drop. A product image lands after a catalog row. An agent that watched Gmail posts that it drafted a reply, and waits. A human comments, edits, or hits send.
That is closer to how a company already works than a blank prompt. It is also why "we gave everyone ChatGPT" does not show up in the ops meeting. There is nothing to point at.
The Dust move, without the Dust stack
Dust's public writing keeps returning to one idea: multiplayer. Skills that encode how a team works. Support that is an agent plus people, not a bot in a corner. TrueLayer's story is a CTO answering "what are we actually doing with AI" with a shared system, not a screenshot of a personal chat.
Glean's 2026 line is "stop botsitting." If you have to re-prompt the same job every morning, you do not have an agent. You have a faster intern who forgets.
A company feed is how you stop both. The job has a place. The last result is still there when the next person opens the workspace. The agent can run again tomorrow without a ritual paste.
A week of work that belongs in a feed
Monday. Sales. Call happened. The recap, owners, and next step post to the Sales workspace. Not a Google Doc that dies in email. Tuesday. Ops. The same order-form template, unstructured mail in, file out. The filled PDF is the post. Someone reviews it like they review a PR. Wednesday. Marketing. Three product stills, resized for the channels you actually publish. The studio job is the post, not a Drive folder named "final_v7." Thursday. Support or success. An agent drafts from the last thread. A human publishes or rewrites. The rewrite is training data for how this company talks, sitting on the same feed. Friday. Nobody asks for a status update. The feed is the update.
None of that requires a new religion. It requires the work to have a URL inside the company.
How to try it on one workspace
Pick the team that already lives in a group chat all day. Create a workspace with their name. Connect the one tool they cannot work without (Gmail, Drive, Slack). Install one agent that already matches the job: recap, document fill, media, a marketplace specialist.
Then make a rule: if it is not on the feed, it did not happen. Private chats can still exist. They stop being the system of record.
You will know it is working when a new person can scroll two days of posts and start, instead of asking who has the prompts.
DIMA workspaces, the feed, chat, media studio, and agents are built as that loop. Open workspaces or automation if you want to see the shape. Start from a real job, not a demo script.
What we are stealing from the category, honestly
Dust: shared context, skills, "we run the company function on the product." Glean: agents that move without a babysitter, measurement so the rollout is not vanity tokens. The rest of the 2026 B2B AI blogs: named jobs, named teams, almost no homepage poetry.
That is the bar for this site too. The feed is the artifact. The prompt is not.
DIMA
