gurupanguji
I am into this idea more than I expected to be.
OpenAI putting Codex, browser use, and ChatGPT into one product makes sense to me, even if I am not the target user for the first version of it.
What they seem to want is straightforward: take the value developers and power users already get from AI tools, then turn that utility into something ordinary people can use every day. A niche power tool becomes a mass tool. That is a big swing. It also feels like the right one.
I think OpenAI needs three things at once. It needs to monetize the attention and habit ChatGPT already has. It needs a path of compounding that reinforces that habit instead of letting it flatten out. And it needs that compounding path to harden into a durable platform layer in computing.
That is why I keep coming back to ads, even if I do not like the answer. Ads are still the laziest big business model on the table, and also the most obvious one. If that is the road, then OpenAI needs more than a famous chat box. It needs more reach, more repeated intent, more things for people to do, and more reasons to come back many times a day.
Mobile ChatGPT is already a real habit. But by itself it is a cramped surface. It does not compete well enough with Google or Meta for the kind of attention and action that ad buyers care about. If advertisers are ever meant to move serious spend, ChatGPT has to become either much more valuable, much more unique, or much broader in where it shows up and what it helps people do.
At the same time, Anthropic found a different path. By leaning into coding and enterprise, and by putting Claude closer to a permissive computer, it started showing AI as leverage instead of AI as attention. People will pay real money for a tool that helps them produce output worth more than the subscription. That is a sturdier business shape than novelty. It also exposes the weakness in a phone-first AI habit.
I think that difference matters. Claude is closer to the computer. ChatGPT has mostly been closer to the feed.
That is where my skepticism starts.
OpenAI seems to want everything. Remember the Code Red and the recent focus memo.
“We cannot miss this moment because we are distracted by side quests,” Simo told staff last week, according to remarks reviewed by The Wall Street Journal. “We really have to nail productivity in general and particularly productivity on the business front.”
My emotional take is more skeptical than excited.
Part of what gave the web its force was that it grew by accident, then by obsession, then by compounding. Nobody sat down at the start and said: now I will become the everything company. The same was true of the PC and the smartphone. The strongest platforms usually earned their power by solving a narrower problem first. They found a seam. They got good there. Then the rest followed.
That is where OpenAI still feels off to me. It does not feel like a company trying to be great at one thing. It feels like a company trying to make sure no durable value escapes its orbit.
That may be unfair. It is also the read I keep coming back to.
They already won a huge amount of consumer attention with ChatGPT. But they seem to treat that win as temporary, maybe because it is. And if the current business still does not justify the valuation story, then the pressure to widen the surface area makes total sense. Grab the use cases that are working elsewhere. Pull them under the ChatGPT umbrella. Turn attention into habit, habit into action, action into monetization.
That is the skeptical case. It is not a clean one. Motives rarely are.
But let me steelman the other side, because I think the other side is real too.
The browser is a fantastic vehicle for this idea. ChatGPT on the web was always an easier sell than asking people to install an entirely new computing layer. And if the browser can learn skills, use sites, and help me do work that currently lives across separate apps, then the shape gets interesting very quickly.
For this to work, the tasks we currently perform across the web and across apps need to become both accessible and combinable. The payoff is simple: remove the friction between them. That friction is a big part of what still makes modern computing harder than it should be, especially for people who do not live inside software all day.
Most users already understand what ChatGPT does. Ask a question. Get an answer. That familiarity earns it the right to do more. Check the web when freshness matters. Use the right site when the task requires it. Learn a skill. Quote something and send it to a colleague. Pull together the files I need for taxes. Fetch the missing statement from my brokerage. Help me make a budget without making me become a spreadsheet monk first.
That loop is the real product shape: ChatGPT, then web, then computer, then back again.
This is where the steelman gets strong. OpenAI may believe its mix of product design, progressive disclosure, and brand recognition can turn power-user utility into mass-market habit. And the optimistic read is bigger than consumer software. If the core interaction is still suggest, ask, verify, act, then the same product shape can stretch into enterprise too, with the right controls around data, permissions, and trust.
The promise is huge. One system that can answer questions, browse the web, use the browser, and act inside apps and accounts already open on your computer. Build that well enough and you get the closest thing yet to a real answer machine.
There is also a broader upside here that has nothing to do with whether OpenAI itself deserves the win. If this works, it pressures the whole industry to make the computer more useful again.
Google, Microsoft, Apple, Anthropic, all of them have reasons to enter the fray. Some already have the hooks: the browser, the OS, the accounts, the default apps, the distribution. Good. Use them. Compete harder. Make the machine better.
That is the part that excites me most.
For years, we have accepted a strange split. Computers keep getting more capable, but the average experience of using them still feels oddly manual. Too much copy paste. Too much tab juggling. Too much remembering where the thing lives. Apps and sites stay siloed. The glue work, the translation work, and the orchestration still fall on the human. For too many people, that is a steep learning curve.
A good answer machine could burn down a lot of that sludge.
And if it does, the upside is bigger than convenience. Better tools can unlock creative energy. They can give more people leverage. They can let someone move from vague intent to rough draft, from rough draft to experiment, from experiment to finished thing.
That does not mean it is easy. This is a long grind. The web took years to become the platform it became, and part of its strength came from protocols and organic growth instead of one company trying to blueprint the whole thing from day one.
Then there is the competition.
There are some signs the rest of the field sees the opening too. Google is reportedly testing a Gemini Mac app. Apple still looks behind through the Siri lens, but it understands the value of useful tools tied closely to the device. Microsoft already has Copilot, even if its business logic stays much clearer and much more enterprise-shaped.
The product fit is not here yet. That is exactly why the opening exists. We are about to find out whether OpenAI actually earned the reputation of being the better product company, or whether it was simply first to a big consumer moment.
I am still wary of bloated interfaces and vague promises. But the core idea lands for me. One system that can answer, browse, use software, and help carry intent across the machine feels like a direction worth pushing hard.
Even better, it may force everyone else to make the computer more useful again. And that matters because I think there is another layer of computing still sitting there, half hidden in plain sight.
Right now the stack still looks familiar. Hardware at the bottom. Operating system on top of that. Then apps. Then cloud services behind those apps. On desktop, much of that client layer is now web code, which means the browser quietly became one of the most important computing platforms of the last twenty years.
If this works, it may bend that stack again.
AI becomes the orchestration layer. It coordinates across apps. It talks to websites. It may even write some of the client logic on the fly if it knows how to talk to the cloud layer behind the scenes. In the best case, the user describes what needs to get done and the machine figures out which pieces to call, which interface to use, and which path makes the most sense for cost, privacy, and speed.
That has real implications. It could change what software feels like. It could change how much local hardware matters. It could create a wider range of computing shapes again, from thin clients that mostly orchestrate cloud work to heavier local setups built around privacy and control.
We have not had many real shifts in the shape of computing for a while. That possibility is exciting. It is also why I do not want to dismiss this move, even while I stay skeptical of the company making it.
The open question is whether any company can ship this in a way that feels powerful without turning the whole machine into a slightly haunted concierge.