What OpenAI will unveil today

Edition 125 | October 6, 2025

If you’re not training like a model while training your models, are you really even grinding?

Welcome back to Building AI Agents, your biweekly guide to everything new in the field of agentic AI!

In today’s issue…

• OpenAI to unveil new features at today’s DevDay

• A science agent company raises $300 million

• Perplexity and Opera release agent-powered browsers

• CrewAI’s new Agent Management Platform for enterprises

• AI agents are automating real work

…and more

🔍 SPOTLIGHT

Today is OpenAI’s annual early Christmas for AI developers, and everyone’s wondering what the presents will be.

The company’s third DevDay will kick off at 10 am Pacific Time with a keynote by its CEO Sam Altman and conclude at 4:15 pm with a fireside chat between Altman and designer Jony Ive of Apple fame, who recently joined OpenAI when it acquired the company he founded.

The top question on every AI agent builder’s mind, of course, is what new tools and features OpenAI will be releasing. Should we expect it to be an earth-shaking event, a nothingburger, or something in-between?

The first OpenAI DevDay was held in 2023 and saw the company, riding high on the explosive success of ChatGPT, release a new flagship model—GPT-4 Turbo—as well as the GPTs feature that allowed each user to build their own custom agent-like tools and a “store” to share them through. It also introduced the Assistants API, OpenAI’s first step towards a dedicated interface for building agents.

2024’s DevDay was more low-key but still significant: it featured the release of the Realtime API for building conversational agents that interact with users through speech, and prompt caching, which reduces costs when repeatedly feeding in similar prompts—something particularly relevant for agents.

How important did these innovations turn out to be in the long run? The much-hyped GPTs feature was a flop—barely anyone uses it anymore. GPT-4 Turbo was a step forward from GPT-4, but nothing revolutionary.

The Assistants API, on the other hand, evolved into the agent-oriented Responses API, which is now the preferred method for interfacing with OpenAI’s models, reflecting the dominance of agents as the main application for LLMs. The Responses API has similarly grown in importance with the rise of conversational agents. Prompt caching, despite its boring name and premise, turned out to be one of the main innovations that made the agent boom possible.

So what possibly-revolutionary features do we expect to see today? The rumors hint at three. First, an AI-powered browser, an extremely hot trend these days—see Perplexity and Opera’s announcements later in the newsletter. Second, improvements (some might say badly-needed improvements) to the GPT store, making it more useful for agent users, builders, and sellers. And third, updates on the the mysterious AI-powered device OpenAI has reportedly been working on—the reason it paid $6.5 billion to bring Jony Ive on board.

So if you’re reading this before DevDay ends, tune in and see what goodies are under OpenAI’s Christmas tree, and keep an open mind. If the last two years are anything to go by, their importance might not be obvious today. Not every new feature will be a hit, but the ones that are could change the future of AI agents.

Always keep learning and building!

—Michael

🤖 LEARN TO BUILD AGENTS

Build your first AI Agent, even if you’ve never coded.

Inside the community, Michael and I walk you through every step, from the foundations to building your first agent and turning your ideas into something people actually use.

Join now to lock in your lifetime price before it increases at 85 members!

Not sure yet? Join The Building AI Agents Community free for 7 days.