More than 20 announcements in a single day. Here’s what matters if you’re building products, running operations or deciding where AI fits in your business.
The short version
OpenAI’s DevDay on 29 September was its biggest yet, with more than 20 major announcements across ChatGPT, Codex, the API and its models. Strip away the product names and three themes stand out:
- Agents that keep working. AI is moving from answering questions to taking on ongoing responsibilities.
- More intelligence for less. Faster tiers and cheaper models make ambitious workflows affordable.
- ChatGPT as a platform. With 1.2 billion weekly users, it’s becoming somewhere businesses build, not just somewhere people chat.
Agents that take work off your plate
The headline launch was Dots: always-on agents that learn what matters to you and work on your behalf in the background. Around them, OpenAI introduced new ways for people and agents to work together, including ChatGPT Space, a shared home for a team’s knowledge, Pages for documents that people and agents edit together, and team tasks that handle recurring work on a schedule or when something happens, such as a new email or Slack message.
For developers, the Agents API now supports computer use, so agents can operate software directly. It also brings Codex’s multi-agent capabilities, tool search and context compaction into your own applications, with OpenAI running the infrastructure. And with Bedrock Managed Agents, built with Amazon, the same capabilities can run entirely inside AWS.
Faster, cheaper models
GPT‑6.1 Sol is a major upgrade focused on agentic coding, computer use and professional work. OpenAI says it delivers close to the intelligence of its flagship GPT‑6 Astra at a fifth of the token price, so the same budget stretches a lot further.
For workloads where speed matters most, the new Ultrafast tier offers up to 8× faster generation in Codex, around 300 tokens per second, and up to 6× in the API.
Smarter decisions, not just smarter chat
One of the quieter announcements may prove one of the most useful. The Decisions API focuses a model on a fixed set of questions with predefined answers. You supply context as text or images, and get back answers you can use to classify content, route requests or choose an agent’s next step.
It’s a sign that the industry sees value in fast, structured decisions alongside open-ended reasoning, the same idea behind the new generation of System One models we’ve written about separately.
Building for and inside ChatGPT
OpenAI is opening up the platform it uses to build ChatGPT itself. Plugin extensions let developers add their own panels and file viewers inside ChatGPT, MCP events let plugins trigger automations when something changes in a connected app, and Sign in with ChatGPT lets people use their plan allowance across 16 partner tools.
Privacy gets a product
For regulated industries, OpenAI Private Intelligence may matter most. Zero Data Retention with private safety processing means automated safety reviews happen without OpenAI staff seeing the content, and a preview of Private Inference, using confidential computing, is due this autumn.
| Announcement | What it is | Why it matters |
|---|---|---|
| Dots | Always-on personal agents | Ongoing work handled in the background |
| Agents API with computer use | Build agents that operate software | Automate workflows without bespoke integrations |
| GPT‑6.1 Sol | Upgraded model at a fifth of flagship pricing | Ambitious workflows on the same budget |
| Ultrafast | Premium speed tier | Real-time experiences become practical |
| Decisions API | Fast answers to predefined questions | Reliable routing and classification |
| Private Intelligence | Stronger data protection controls | A clearer path for regulated sectors |
What we’re telling clients
Don’t rebuild everything. New capabilities arrive every few months. The businesses that win pick one valuable workflow, prove it, and expand from there.
Design for agents, with guardrails. Agents that act on your behalf need clear permissions, human checkpoints for sensitive steps, and a record of what they did.
Match the model to the job. Not every task needs the most powerful model. Fast, cheap decisions for routine steps, and deeper reasoning where it counts, is how AI products stay quick and affordable at scale.
Treat data protection as a feature. The new privacy options make AI viable in more sectors, but only if your architecture is built to use them.
If you’re working out what DevDay means for your roadmap, we’re happy to talk it through.
