OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux

I have spent enough time in the crypto and AI sectors to recognize the smell of a paradigm shift. It usually starts with the people closest to the source code changing their vocabulary. Recently, Tibo Sottiaux, the Head of Product for ChatGPT at OpenAI, started talking about a transition from passive tools to active agents. It is a subtle distinction, but for those of us building in this space, it changes everything about our roadmap.

The End of the Chat Interface

For the last two years, we have been obsessed with the 'box.' You type something in, the machine spits something out. It is a linear, conversational loop that mimics human interaction but lacks human agency. OpenAI is signaling that this phase is nearing its expiration date. They are pivoting toward what they call 'Dots'—a internal focus on connectivity and reasoning that goes beyond just predicting the next word in a sentence.

For a founder, this means the chat window is no longer the product. If you are still pitching a 'ChatGPT for X,' you are already behind. The real value is moving toward the background. We are looking at a future where AI does not wait for a prompt; it observes a workflow, identifies a bottleneck, and clears it without being asked. This is the shift from a tool to an employee.

Why Agents Will Own the Web

One of the most provocative claims coming out of the OpenAI product camp is that agents will soon dominate internet traffic. Right now, the web is designed for humans. We click, we scroll, we authenticate. But when you deploy an agent that can navigate a browser, the math changes. An agent does not need a flashy UI. It needs a high-rate API and a clear set of permissions.

If agents become the primary consumers of web content, the way we build websites has to change. We have spent a decade optimizing for SEO and human attention spans. If the majority of your 'users' are actually autonomous scripts running on behalf of a human, your conversion metrics, your layout, and your monetization models are all broken. Builders need to start thinking about 'Machine-Readable UX.' If an LLM cannot parse your value proposition in milliseconds, you do not exist.

The Reasoning Gap

Sottiaux pointed out something that I have been preaching to my own dev teams: builders are still overestimating the 'intelligence' and underestimating the 'reasoning.' Most current AI implementations are just fancy autocomplete. They are probabilistic, not logical. The next era—OpenAI’s 'biggest bet'—is moving toward systems that can plan multiple steps ahead.

This is where the 'Dots' concept comes in. It is about connecting disparate pieces of information to form a coherent strategy. If you tell an AI today to 'start a business,' it gives you a list of steps. An agentic system with reasoning capabilities will actually go register the domain, set up the landing page, and run the first round of A/B tests. The leap from advice to action is the gap OpenAI is trying to close.

What Builders are Still Getting Wrong

There is a recurring mistake I see founders making right now. They are treating AI as a feature rather than a foundation. They take a legacy product, slap a chatbot on the side, and call it 'AI-powered.' That is the equivalent of putting a jet engine on a horse-drawn carriage. It is heavy, it is awkward, and it does not take advantage of the tech.

• Stop building wrappers: If your product can be sherlocked by an OpenAI system update, you do not have a business; you have a temporary exploit.
• Focus on proprietary data loops: The model is a commodity. The feedback loop—how your users interact with the AI and how that improves the system—is your moat.
• Solve for agency, not answers: Users do not want answers; they want outcomes. Build systems that do the work, not just systems that talk about the work.

The Skeptical Takeaway

We have to be careful here. OpenAI has a vested interest in making us believe the 'next big thing' is right around the corner. It keeps the VC funding flowing and the talent from jumping ship to Anthropic or Google. The 'Agentic Era' sounds great in a podcast, but the technical hurdles are massive. Hallucinations are annoying in a chat box; they are catastrophic when the AI has access to your bank account or your production servers.

However, the direction of travel is clear. We are moving away from LLMs as a destination and toward LLMs as an operating system. As a founder, your job is not to build the OS. Your job is to build the essential applications that run on it. If you can create a system that acts reliably on behalf of a user, you are winning. If you are just building a prettier version of ChatGPT, you are just waiting to be disrupted by a system update.

The era of the chatbot was the opening act. The era of the agent is the main event. It is time to stop talking to the machine and start letting the machine work for us. That is the only way to build something that actually lasts in this cycle.

Read the original at Lenny's Newsletter →