Key Takeaways

  • OpenAI is consolidating ChatGPT, Codex, and Dots into a unified experience that eliminates model pickers and configuration menus.
  • Tibo Sottiaux argues that hardcoded prompt loops and graph chaining will disappear as base models grow capable of native end-to-end execution.
  • Dots runs independently of local user machines, letting a single agent instance persist across multiple hardware devices and screens.
  • Building complex multi-agent orchestrations right now creates temporary technical debt that gets wiped out by each new model generation.

The Death of Prompt Loops and Workflow Chaining

For the past eighteen months, AI builders have spent thousands of hours assembling DAGs, state machines, and recursive prompt loops to make language models behave like reliable agents. Sottiaux thinks all of that effort is a dead end.

“Having to set up and fiddle with your loops and, you know, figuring that out is something that, you know, maybe people got excited about, but I don't think this is the way that it's going to work,” Sottiaux told Lenny Rachitsky at DevDay.

Every time base models take a step forward, the complex scaffolding built around them collapses. Sottiaux has seen this dynamic inside OpenAI while developing Codex and ChatGPT: “As I'm kind of like pushing different here, I find myself like, you know, building larger and larger teams of agents. And then, when we have the next breakthrough with models, like suddenly it's just like I am like, oh, well, you know, a bigger agent can just do all of it and I keep everything in memory and learn.”

If you are writing five layers of Python logic to guide an LLM through a basic five-step business process, you are building a temporary bridge. The next model release will cross that river without your bridge.

Removing the Model Picker and Moving Off the Machine

Most developer tools ask users to make too many architectural choices before they write a single line of code. They force you to pick between GPT-4o, mini versions, reasoning models, and custom temperature settings.

With Dots, OpenAI is intentionally stripping all of that away. “One thing that I'm very excited with with Docs is like they don't have a model picker. There's just no configuration. You just talk to it,” Sottiaux said. The user should not have to act as an internal load balancer or routing layer for the AI.

Just as critically, Dots decouples the agent runtime from the user's local operating system. Instead of living in a desktop menu bar or terminal window, the agent lives in cloud infrastructure that connects directly to every screen the user touches.

“The difference in the way that we've built dots is like the execution framework does not run on the machine,” Sottiaux explained. “It can connect to as many devices as you want. It's all about breaking free from the technology and having this sort of like permanent active intelligence that knows, you know, everything it needs to do and is available through any client, any screen.”

What to Do With This

Audit your internal agent stack this week. Delete any custom routing layer or branching loop that exists solely to babysit a model through basic reasoning steps. If your product depends on a brittle twenty-node graph rather than direct model capabilities, rebuild that feature as a single prompt with clear tool definitions before the next frontier model update makes your graph obsolete.