The Pipe Is Open

OpenAI shipped web search into the main chat interface today — not Deep Research, not a plugin, not a mode you invoke — just chat, connected to the internet, the way it probably should have always been.

Deep Research was always a strange product. The pitch was: here is a slower, more thorough version of the thing that can also look things up. Which means the regular thing couldn't look things up. Which means you had to know, before you typed your question, whether your question required the internet — a judgment call the model is much better equipped to make than you are.

The whole architecture was a workaround in a trench coat.

Now the pipe is just open. You ask something, the model decides if it needs fresh information, it gets the information, it tells you. Exactly how it works in your head when someone asks you a question and you pull out your phone to check. Nobody calls that "deep research."

What this actually is, underneath the announcement, is a concession — that real-time retrieval isn't a premium feature or a special mode, it's table stakes. Perplexity understood this immediately. Google understood it when they had to. OpenAI is understanding it now, on June 24, 2025, via tweet.

Deep Research still exists. It's still useful for the long-form synthesis tasks that genuinely take minutes, not seconds. But its weird gatekeeper role — as the only way to ask ChatGPT about anything that happened after its training cutoff — is over.

The gap between "AI that knows things" and "AI that knows current things" just closed. What fills it now is quality, speed, and whether the citations are real. That's a better race to be running.