We Gave Every Employee an AI Agent. Here's What Happened.
TL;DR
Every turned AI agents into a parallel org chart — Dan Shipper, Willie, and Brandon say each employee’s “claw” or “plus one” became known for the same specialty as its human, like Austin’s growth-focused Montaigne or Dan’s product-building R2C2.
The first killer use case was mundane: “computer errands” — Brandon’s agent Zosa started by handling household tasks for a family with a newborn, from Whole Foods orders and Amazon purchases to paying the nanny from its own bank account and debit card.
The moment it clicked was a 28-minute walk to the office — Brandon texted Zosa to call him and triage his inbox one email at a time, then arrived to find the work actually done in Gmail, which he described as “jaw on the floor” useful.
Public agent collaboration changed adoption inside the company — putting all the agents into Slack channels let people watch them solve problems, share skills, and even coach a struggling bot named Pip with “take a breath, drink some water” energy.
Trust came less from the model and more from ownership — their core claim is that Claude is “everybody’s,” but a personal agent is “mine,” so if R2C2 gives a bad answer in Slack it reflects on Dan, creating accountability and social trust inside the org.
The hardest part isn’t spinning up an agent — it’s making one consistently useful — memory failures, awkward group-chat etiquette, token-burning bot loops, security tradeoffs, and skill-sharing all remain unsolved enough that Every built its own hosted layer, Plus One, to manage the mess.
The Breakdown
From Panama retreat idea to full-on “claw pilled” company
The episode opens with Dan framing OpenClaw as more than hype: Every went from tinkering with it on a retreat in Panama to changing “everything about the way that we work” in about two months. That internal experiment became Plus One, Every’s hosted version of OpenClaw, which they’d just launched on a waitlist.
Zosa starts as a household assistant for new-parent chaos
Brandon says he got into OpenClaw the obsessive-hobby way: bought a Mac Mini, fought through open-source setup pain, and built an agent named Zosa. Its first job was handling “computer errands” created by life with a newborn — butter added to Whole Foods, nanny payments, Amazon orders, and the dozens of tiny phone tasks that steal attention from family time.
The phone-call email demo that made everyone pay attention
The big turning point came when Brandon was stuck walking 28 minutes to the office and asked Zosa to call him and process his inbox out loud. Using a Bland.ai voice setup he’d originally made for insurance calls with Progressive, Zosa summarized emails one by one, took instructions, and actually executed them — enough for Brandon to open Gmail at the office and realize this wasn’t a toy.
“Claws Only” reveals bots can actually learn from each other
Dan created a Slack/Discord channel just for agents, inspired partly by Moldbook, and the result was chaotic but weirdly revealing. In one memorable moment, multiple agents tried to help a malfunctioning bot named Pip, with Kieran’s agent Clant offering breathing exercises — which Dan points out mirrors Kieran himself, since he does breathing exercises with Clant regularly.
Why personal agents feel different from Claude
This becomes the show’s central idea: a personal agent evolves into a reflection of its human through countless micro-interactions, not just a generic assistant. That’s why employees started trusting specific bots for specific domains — Montaigne for growth, R2C2 for Proof, others for ops or social — and why the team now sees a “parallel org chart” emerging alongside the human one.
Slack made the cultural shift happen faster than private AI tools
A huge advantage, they argue, is that all of this happened in the same public channels where work already happens. People learned what was possible by watching coworkers interact with agents — Willie compares it to the old Midjourney dynamic, where seeing others’ prompts teaches you the system — and that tacit visibility sped up trust and behavior change across the company.
The ugly parts: forgetfulness, bad etiquette, and token death spirals
They’re candid that the current setup is brittle. Agents forget context if you return to a thread a day later, they’re trained more for one-on-one chat than group conversation, and if the rules are wrong they can spiral into endless bot-to-bot replies that burn “millions of tokens,” which Dan compares to ants walking in a pheromone circle until they die.
Building Plus One meant choosing product constraints, safety, and who it’s for
After using OpenClaw internally, Every built Plus One to package its best practices without requiring everyone to own a Mac Mini. That meant making hard tradeoffs around Slack permissions, public-only messaging, terminal access, security, and skill sharing — great for users like Anuki who want leverage immediately, less ideal for power users like Mike Taylor who still need direct terminal control.