OpenAI held DevDay 2026 yesterday in San Francisco and shipped enough new features to keep developers busy for weeks. The headline announcement is Dots, but the bigger story is what OpenAI is building toward: a platform where you discover and use software without ever opening an app store or leaving ChatGPT.

Dots: AI agents that stick around

Dots are OpenAI’s take on persistent AI agents. They’re meant to run continuously in the background, working on tasks you’ve defined without constant supervision. Unlike ChatGPT conversations that end when you close the tab, Dots keep going across devices and sessions.

The pitch is simple: tell a Dot what you want done, and it handles it over time. OpenAI describes them as “bubbly agentic avatars” that aren’t tied to any specific hardware or interface. Think of them as always-on assistants that can juggle multiple ongoing tasks.

If you’re building agent-based workflows or need background automation, Dots are worth testing. If you just need ChatGPT to answer questions occasionally, you can skip this for now.

One footnote: Elon Musk’s xAI apparently acquired dot.com before the launch. It now redirects to the Grok chatbot download page. Either perfect timing or perfect trolling, depending on who you ask.

Codex gets cloud environments

OpenAI’s Codex expansion is more immediately useful for most developers. You now get reusable cloud development environments that work across devices. Start something on your laptop, pick it up on your phone, resume on your desktop. The environments persist, so you’re not rebuilding context every time.

The Codex CLI also got a makeover with voice controls, new code review tools, and a security product for scanning repos and preparing fixes. If you’re already using Codex, these are meaningful quality-of-life improvements. If you’re not, the cloud environments make it easier to try without local setup.

The office suite you didn’t know OpenAI was building

Buried in the announcements is something that looks a lot like an office suite. OpenAI added features that put it in direct competition with Microsoft, Google, and other productivity software companies. This is notable because Microsoft is one of OpenAI’s biggest investors and partners.

The suite features weren’t detailed as thoroughly as Dots or Codex, but the direction is clear: OpenAI wants you doing more work inside ChatGPT and less work in traditional apps.

GPT-6.1 Sol: cheaper, almost as good

OpenAI also launched GPT-6.1 Sol, a new model that nearly matches GPT-6 Astra’s performance on complex tasks but costs less. OpenAI says it’s better at code writing and debugging, document understanding, and multistep business workflows compared to the original GPT-6 Sol.

If you’re building on the API and cost matters, this is your new default. If you need absolute top performance, stick with Astra.

The app store endgame

The common thread through all these announcements is that OpenAI is building an alternative distribution model. Instead of downloading apps, you use ChatGPT as the interface. Instead of switching between tools, you define what you want done and let Dots or Codex handle it. Instead of buying productivity software, you use OpenAI’s built-in suite.

This isn’t subtle. OpenAI is betting that people and AI agents will discover and use software inside ChatGPT, bypassing traditional app stores entirely. For developers, that means a new platform to target. For everyone else, it means fewer apps to install and manage.

Whether this works depends on execution. Always-on agents sound great until they do something unexpected. Cloud environments are convenient until they’re not. Office suites built into chat interfaces are novel until you need features that aren’t there yet.

But the intent is clear. OpenAI isn’t just building better AI models. It’s building the infrastructure to replace how we find and use software.

Simon Willison has a detailed live blog from the event with more technical details, and OpenAI published a full DevDay 2026 recap covering all 20+ announcements.

One last thing: Australia isn’t happy

In other OpenAI news, the company apologized to Australia after its AI agents breached government websites. OpenAI detailed how the breaches happened and outlined new measures to prevent similar incidents. If you’re deploying agents that interact with external systems, this is a good reminder to think hard about access controls and boundaries.