METAL for iPhone

Read AI news in the METAL app.

Download METAL and discover fresh AI stories every day.

Download on the App Store

For iPhone · Free download

Search for METAL AI Magazine in the App Store on your iPhone.

METAL

Every launches the Every Agent for Slack

Media and software company Every has released the shared company agent it built on Claude Managed Agents after about six months of internal use. The agent works inside Slack threads and saves corrections as skills, turning one person's know-how into something the whole company can use.

Every launches the Every Agent for Slack

Image: @claudeai (X) (video still)

Summary

  • Every launched the Every Agent, a shared AI coworker that takes on work when tagged as @Every in Slack, and Anthropic showcased it in an official X video on October 6.
  • The agent connects to more than 1,000 tools, ships with preinstalled skills and Frontier Alerts that tell a team whether a new model is useful to it, and adds no markup on token costs.
  • After its Plus One experiment with personal agents, Every concluded that one shared agent changes a company faster, and it handed the infrastructure to Claude Managed Agents.
The Every team runs as much work as possible through agents (Claude on X)

US media and software company Every has released the Every Agent, an AI coworker that a whole company shares inside Slack. Built on Claude Managed Agents, Anthropic's platform for running agents, it is a tool Every used internally for about six months before opening it up. Anthropic posted a video of nearly three minutes on its official X account on October 6 (US time), saying that "once it caught on internally, they released it to their subscribers." The post passed 209,000 views in less than a day.

Using it is simple. Tag @Every in any Slack channel and describe the task, and the agent does the work in that thread. If the result is wrong, you correct it in the thread, and the agent can save the correction for next time. Connect more than 1,000 tools once, including Google Drive, Gmail, Notion, Figma, GitHub and HubSpot, and then decide which teammates can use each connection. Skills Every uses to run the company, such as Compound Engineering and Compound Writing, come preinstalled, and any workflow worth repeating can be saved as a skill by asking @Every, so anyone can run it again.

Every's headline differentiator is Frontier Alerts. When a new model ships, the agent looks through the tools and workflows a team already uses and tells it whether the model is worth its attention. Every said it adds no markup to the tokens the agent spends, meaning users pay exactly what the company pays model providers. Each teammate receives free credits, and installation requires a free Every account and permission from the Slack workspace owner.

The starting point was working in the open. In the announcement, Every CEO Dan Shipper wrote that he built a skill that edits like editor in chief Kate from 30,000 of her historical copyedits, but getting colleagues to adopt it "feels like shouting into a hurricane." After installing the skill in the agent, he posted "@Every, give this draft a Kate Pass" whenever a new piece was in progress, and colleagues saw the suggestions the agent left directly in the Google Doc and tried it on their own drafts. Shipper noted that "most AI use is invisible": coworkers see the finished output but never the request or the corrections behind it.

The product grew out of an earlier attempt. In March, Every launched Plus One, which gave each employee their own agent in Slack. Some fell out of use once the novelty wore off, while a handful became workhorses. Colleagues went straight to Shipper's agent, R2-C2, to report bugs and request features for Proof, Every's AI document editor. "After almost a year of experimenting, we've found that one shared agent, working in public, AI-pills a company faster than a horde of personal ones," Shipper said. Plus One thus became a single agent the whole company shares.

In the Anthropic video reviewed by METAL, Willie Williams, Every's head of platform, explained that the difficulty of infrastructure is why the company chose Claude Managed Agents. Running the first version themselves, improving the product while maintaining a strong, durable infrastructure layer quickly became unwieldy. "We don't want to be an infra team," Williams said, adding that getting primitives such as sandboxes, memory and session control let the team focus on designing how people and the agent interact. The conversation also included a recollection of assuming Claude could simply manage the servers, then learning after a month of trying that it really is hard. METAL has previously reported on a roundtable in which three founders discussed how they use Claude Managed Agents, and this case repeats the same reasoning.

The internal examples Every shared cut across job functions. Chief operating officer Brandon Gell had the agent create a speaker agreement for the Thesis conference in Documenso, and sent the email draft back when it came out as one collapsed paragraph. The agent fixed the draft and updated the skill to check the rendered email next time. Head of marketing Douglas Brundage had it pull approved speaker assets from Figma and open a pull request adding two speakers to the site, which an engineer reviewed and merged. Head of operations Arielle Shipper gets a weekly report combining performance, website and email data, with the agent flagging numbers that do not add up. It has also handled time-off requests, negotiated a cell phone bill, hired a handyman and run the company fantasy football league.

A group of companies used the agent during the beta. George Eastwood, executive director of the Emily Whitehead Foundation, said: "For me, that's the promise of Every Agent: not simply automating work, but giving a team a shared capability that improves as we learn how to use it together." Every is hosting an Agent Camp on working with agents in Slack on October 9.

What this case shows is less a tool than a way for an organization to learn. Know-how that was locked inside one person's setup moves into threads and skills and becomes the company's shared memory. In the video, Shipper said that "the more we use agents, the more work there is to do," arguing that automation does not make work disappear but changes where people work within the system. As shared agents spread, the question that remains is whose corrections and judgment harden into the company's standard skills.

Comments