OpenAI Dots Took Over Our Company

Summary · 4 min read instead of 34 min

Williams is head of platform at Every and led engineering on the Every Agent, a company agent that launched the week of the episode. Shipper says Williams has worked in engineering management at startups and larger companies and has founded his own companies. Every first ran personal OpenClaw agents, then moved to one shared agent.

Every moved from everyone running a personal OpenClaw agent to one shared company agent, because personal agents were hard to maintain and faded without investment. Williams argues work should run on a company agent that learns the team's workflows, while personal agents stay personal. He expects company agents to split into team-level agents over time. He says AI has made engineering management less manual but still a people job. Both men expect agents that proactively surface information to matter next, and Every is building that into its agent.

Minute by minute

1. 0:00

   Dots use cases across Every. Shipper describes how Every's roughly 30 people use OpenAI's persistent agents, called dots. Kieran Klassen uses one to scan school emails, and Ariel Shipper's dot caught access requests in Slack. Randy Koonsman worked on the launch video by phone with his dot. Shipper used his to track Slack replies to his model-testing results without getting pulled into Slack.

2. 2:40

   Dots are promising but permissions are rough. Shipper recommends dots, especially for people already in the ChatGPT ecosystem. He finds the permission structure a mystery: he often approves a task several times and the dot still says it can't do it. He expects persistent agents to be the default way of using AI in a year, but without cute characters like his dot, Boo.

3. 5:53

   How OpenClaw spread through Every. Williams recalls a contributing writer, Ash, pushing him to try the project, and Brandon wiring one up first. It spread at Think Week in January, when people showed each other what their agents could do. After a model drop, the agents worked better and began talking to each other in Slack. He says seeing colleagues use AI was the big gain, because before that people only saw outputs.

4. 9:51

   Personal agents suit tinkerers, not operators. Williams says personal setups work well for builders who like tinkering. For operators they brought security worries, complexity and constant breakage, and an agent that gets no investment fades. Every concluded it was better to put that energy into one agent the whole company benefits from.

5. 11:05

   The claw platform failed and personal agents died. Every built a one-click platform for non-technical colleagues to make their own claws. Williams says OpenClaw moved fast and was vibe-coded, so it was hard to build a platform on. The claws died, and Shipper's R2C2 went offline when a Mac Mini lost power and he never restarted it. Well-designed company agents, such as Victor, and Austin's growth agent convinced the team that a single company agent made a big difference.

6. 15:12

   Builders want complexity, operators want it to work. Williams says AI products have a large surface area, so builders can't map every user flow as they could with a SaaS app. He says builders want to turn knobs, while people running a business want something that works. For project management and operations, Every preferred a solid, shared agent where someone has already thought through each skill.

7. 16:30

   Company agents for work, personal agents at home. Shipper expects to use a dot in Slack as a portal to his Codex, alongside a company agent. Williams sees a firm split: one agent per company, because it learns workflows and everyone improves it, and a personal agent at home for him and his family. He says a personal agent brought into a company lacks context and would need about a year to get up to speed.

8. 19:43

   Engineering management leverage comes from visibility. Williams says engineering management feels the most similar to his earlier jobs. A manager's leverage comes from seeing the field, and he uses Shipper's Tend feed to consolidate Slack, email and customer reports. During the launch it showed him marketing conversations and meetings he would not otherwise have caught, so he knew whom to guide.

9. 23:49

   Managing is still a people job. Williams says his token usage fell sharply as he moved back into management, because the job is mainly removing bottlenecks through conversation. Agents now handle summaries, routing and ops work. He describes agents watching dashboards every five minutes for anomalies and acting as the first on-call responder.

10. 27:04

    Company agents will split into team agents. Williams says personal agents were too hard to maintain, and expects the company agent to subdivide into team agents, as Every already has with a couple of them. These may later split further into something like a shadow org chart. Shipper says he still might be right, because technology bets are usually early rather than wrong. Williams adds that AI transformation usually comes from a CEO or senior leader, not line engineers.

11. 28:57

    Separate agents for separate roles. Williams uses GrokBot for several coaches, including one for lap swimming, and also has a therapist. He says the coach model maps to how people deal with humans. He keeps a transactional task as a thread in Codex, and a relationship-like one gets its own bot. Shipper says one agent per area marks a lasting interest, and notes you can't have multiple dots.

12. 31:40

    Proactive agents will replace feeds. Shipper says the Every Agent includes frontier alerts that explain what happened and what it means for you. He says feeds used to require a huge corporation and now anyone can have one. Williams says he wants an agent to take the brunt of X and pass on only the useful items, and Shipper says his dot already does this for Slack.

Published Oct 7, 2026.