The short answer
OpenAI’s Dev Day shipped Dots, an always-on agent, plus Spaces and Pages, cloud Codex, and the Decisions API, each aimed at a tool people already pay for. For small AI startups, the lesson is that any product someone can clone from your URL is not a moat. For buyers, the move is to keep a human review gate, test the previews on personal projects first, and treat customer experience as the asset that survives.
OpenAI’s Dev Day shipped 25 updates in one day, and four go head to head with tools I have recommended, including Grok Bot, which I praised in my last few videos. After 24 hours with the new Dots agent I have moved roughly 45 percent of my Grok Bot workflows to it. So this is partly me eating crow, and partly my read on what it means if you build a small AI product or buy from one.
What did OpenAI announce at Dev Day?
Four of the 25 updates target a tool people already pay for. Dots is an always-on agent, the answer to Grok Bot. Spaces and Pages are a shared workspace and a document built for humans and agents, the answer to Notion. Codex now keeps working in the cloud after you close the laptop, the answer to Devin. And the Decisions API, which can only answer from a fixed list, is the answer to Jev. All of it came a month after ChatGPT-6 Astra, which Dots runs on today.
What is Dots, and why did I move my Grok Bot workflows?
Until this week, ChatGPT was a pile of chats. One for email, one for the report, one for the landing page. Dots replaces the pile with one agent. You name it, mine is Mad Max, then message it, email it, add it to Slack, or call it. I tried calling it and it hung up on me. It gets its own computer in the cloud and keeps working after the laptop closes.
The difference from Grok Bot is context. Grok Bot and Meta’s Muse had to learn you from scratch. ChatGPT has two years of my memory, skill files, and conversations. Futurepedia, who had early access, sent his Dot one short message asking for three video intros, motion graphics for the best one, and a thumbnail. It wrote the intros, handed the winner to Codex for the animation, and opened another chat for the thumbnail.
Within its first 12 hours, mine flagged a loose API key my daily Claude security audit had missed. That is the shift. Your context is now your biggest asset, and your job moves from instigator to reviewer. One catch: paid plans get one Dot per account, so if you like several specialized agents it will feel cramped.
What changes when the agent works in the cloud?
Spaces is a shared home where your team, ChatGPT, and your Dot work off the same knowledge base. A Page is a document inside it, built for human and agent collaboration. If you have used Notion it will feel familiar. I am not leaving Notion yet, but I can see that $20 a month subscription becoming one I no longer need.
Grok Bot’s whole pitch was working after you close the laptop. Weeks later, Codex does the same with far more context about you. I can start a build from the ChatGPT app on my phone, play a round of golf, and come home to a working product. The Mac mini era of running agents locally out of fear is over. If you hold client data, keep running locally and test into the cloud slowly.
What is the Decisions API, and should you use it with clients?
Jev came out of Type Safe about a week and a half ago, invite-only, and went viral because it cannot write anything. It can only pick from a list you give it. OpenAI’s version is the Decisions API. You hand it context, text, or images plus a fixed set of answers, and it picks one. OpenAI suggests it for classifying content, routing requests, or choosing an agent’s next action.
In my world that means a lead is qualified or not, sits in one of five lifecycle stages, and goes to one of three reps. If you have built agentic workflows for HubSpot, you may have watched an agent invent a lifecycle stage that does not exist. The Decisions API can only say what is on your list, so you can check its work. That is a review gate you can put in front of a client.
- It is a limited preview, with broad release promised within days.
- Greg Isenberg, who likes Jev, is openly unsure which one wins.
- My verdict: test it on your own projects and keep it away from client work for now.
Why does Sign in with ChatGPT matter more than the clones?
On the same day it cloned Notion and Devin, OpenAI launched Sign in with ChatGPT with 16 partners, Notion and Devin among them. It lets you use the plan you already pay for inside other tools. OpenAI wants to be where you log in, where the work lives, and where the agent runs. Everyone else becomes a feature inside it or a partner next to it. OpenAI is well ahead in that race.
The moat question
If you are building, assume someone can hand your website URL to ChatGPT-6 Astra and have a product an inch better than yours in a day or two. Your moat is not the build. It is the customer experience around it, and how human it feels in a transactional, agentic market.
What should a small AI startup, or a business buying from one, do now?
Builders: pick the part that cannot be cloned from a URL. My specialty is customer relationships, and what most teams still fail at is a next-level customer experience. Buyers: keep the human review loop active and skip the swarm of agents spamming your customers. Everything here is a day old. Dots is one per account, the Decisions API is a preview, and the numbers are my own. Recheck all of it before you cancel anything.
If you want a second set of eyes on which of these releases belongs in your business and which should wait, I would be glad to talk it through.
Book a free callKey takeaways
- Jake Bauman ran OpenAI’s Dots agent for 24 hours and moved roughly 45 percent of his Grok Bot workflows to it, crediting the memory and skill files ChatGPT already holds from years of use.
- OpenAI’s Decisions API takes context plus a fixed set of answers and picks one, which Jake frames as a review gate for lead routing and lifecycle stages that an agent cannot corrupt with invented values.
- On the same day it launched Spaces and cloud Codex, OpenAI released Sign in with ChatGPT with 16 partners including Notion and Devin, which Jake reads as a bid to be where you log in, where the work lives, and where the agent runs.