Every Is Half Agent Now

Source: original

‘AI & I’: Agents work among us

Today, we’re releasing a new episode of our podcast AI& I. Dan Shipper sits down with Every’s COO Brandon Gell and head of platform Willie Williams to discuss the good, bad, and weird of how daily operations change when everyone at your company has an agent.

A “parallel organization chart,” in which each AI worker has a name, manager, and job description, allows your company to move faster than it ever could with humans alone. It also raises a host of new questions about how work can—and should—get done.

Watch on X or YouTube , or listen on Spotify or Apple Podcasts. You can also read the transcript.

Here are the highlights:

• We’re writing the etiquette in real time. Each person at Every has a dedicated OpenClaw AI assistant, or Plus One , trained to assist with or fully handle parts of our jobs. R2-C2, for example, reports to Dan and is responsible for collecting flagged bugs and generating pull requests for Proof , Every’s collaborative document editor for agents and humans. So when do we turn to Dan versus R2-C2 for Proof-related troubleshooting? Brandon’s rule of thumb: If an established process or tool needs to be used or fixed, ask a Plus One. R2-C2 knows all about Proof, and Dan’s a busy guy—bug reports and questions about how to use the app or report a bug should always go to the agent.
• Agents gain credibility by doing. The fastest way to get other people to trust and use your Plus One is to have it execute tasks in public. Austin Tedesco is Every’s head of growth, and Montaigne, his Plus One, essentially co-runs the department. Austin asks Montaigne to generate campaign scorecards, analyze metrics for growth insights, and handle all sorts of other complex tasks. Watching Montaigne pull off these requests proves its capabilities to the team—and inspires others to push their Plus Ones to achieve more, too.

Austin Tedesco asks Montaigne to analyze YouTube keywords for ‘AI & I’ (All screenshots courtesy of the Every Slack workspace unless indicated otherwise.)

• Everyone is a manager now. Agent sidekicks force each of us to change our approach to getting work done. To get the most out of a Plus One, you need to actively manage it—onboard it, delegate tasks to it, evaluate its performance, and give guidance so mistakes aren’t repeated. For anyone who hasn’t had a direct report before, “there’s an education that has to happen,” Brandon says.

Miss an episode? Catch up on Dan’s recent conversations with LinkedIn cofounder Reid Hoffman ; the team that built Claude Code, Cat Wu andBoris Cherny ; Vercel cofounder Guillermo Rauch ; podcaster Dwarkesh Patel ; and others, and learn how they use AI to think, create, and relate.

Signal

Anthropic’s most capable model is coming—just not to youThe news: Anthropic has built Mythos , a powerful new model, but does not plan to make it public. Instead, access is going exclusively to Project Glasswing , a coalition of big technology companies including Apple, Google, and Microsoft, giving them time to patch bugs the model will expose.

The context: Mythos scores 93.9 percent on SWE-bench Verified , up from 80.8 percent for Opus 4.6, an unprecedented 13-point jump that means it “crushes any programming task—and that includes finding security vulnerabilities in software,” says Every engineer Nityesh Agarwal. Mythos found zero-day bugs in every major OS and browser, without human guidance.

“With a jump like this, you can point Mythos at any codebase, tell it to build a feature, and it’ll just do it,” Nityesh Agarwal says.

Why it matters: This is the first time a frontier lab has opted not to release a model publicly. Glasswing is Anthropic’s bet that the window between “this exists” and “this is everywhere” can be used to harden the world’s software before Mythos—or a similar model from a rival lab—wreaks havoc.

Steal this workflow

A directory for agents

At Every, the parallel organizational chart for our agents built itself organically. So we went back and catalogued how who our Plus Ones reported to, what repertoire of skills each one had, and how we were interacting with them.