Everyone's Building the Agent. Decide If Your Product Is One.

The biggest names in tech spent their last earnings calls promising the same thing: an AI that works on your behalf, all day, without you typing much at all. Meta, Microsoft, OpenAI, plus a fresh crop of funded startups are all pointing at the same target. That means your product now sits somewhere in this shift, and you need to know where. Let me catch you up.

The whole industry pointed the same direction

The pitch is now consistent across the big players. Mark Zuckerberg told investors it's "extremely unlikely" that five years out, billions of people won't have a personal agent handling their finances, health, and household. Microsoft's Satya Nadella confirmed a Copilot "super app" shipping this year that folds chat, coding, and agents into one thing. OpenAI's Greg Brockman said the company is building a "family of devices" so you talk to your computer instead of typing.

When three companies this size describe the same future in the same quarter, it stops being a vision and starts being a roadmap you have to plan around. The agent is no longer a feature bolted onto an app. It's being positioned as the thing you open first.

Coding was the easy win. Everything else is the prize

Agents proved themselves in coding first, because engineers will fight janky tools to get results. The interesting part is what happened next. OpenAI says Codex hit more than 10 million users combined with ChatGPT Work less than two weeks after launch, and knowledge workers were growing 3x faster than developers on it. The line OpenAI's Akshay Nathan keeps coming back to: there are roughly 100x more people who use software than who can write it.

That's the real market. Instead of opening a spreadsheet or a slide deck and running the features yourself, you describe the outcome and an agent assembles it. OpenAI is even betting interactive Sites will replace decks and spreadsheets. If your product is a place people go to do a task by hand, that habit is the thing under pressure.

The catch: your product might become the plumbing

Here's what should worry you. If an agent does the task, the user may never open your interface again. They ask the agent, the agent calls an API, the job gets done. Your careful design work becomes invisible pipes.

Meta is already selling this both ways. Zuckerberg says more than 1 million businesses use its business agents each week on WhatsApp and Messenger, and he framed the enterprise play as APIs, agents, and compute sold to other companies. He also admitted selling to enterprise is a "different muscle" than Meta's ad business. So even the giants are figuring out where they sit in this stack. You should force the same question for your own product.

Small and cheap is beating big and general

Don't assume this belongs only to the labs with the most compute. Prentis, a new lab co-founded by Reid Hoffman and Marc Pincus, is raising $100M to build agents that run routine office work like insurance claims and customs refunds. It claims its smaller Hive-32B model beats GPT-5.4 and Claude Opus 4.6 on two computer-use benchmarks, at roughly 10x lower cost per task.

And it charges 20% of the savings it delivers, not a seat license. That pricing tells you where value is moving: to the outcome, not the tool. A focused agent aimed at one ugly workflow can win without a frontier model behind it. That's a door for your team, and a threat if a startup aims it at your customers first.

The deep cut

Nathan's sharpest point wasn't about models. It's that when almost anyone can build, ideas and taste become the bottleneck, and teams have to tell real progress from motion that just looks busy. That's the practical read for you. The scarce thing on your team is no longer the ability to build a screen. It's knowing which outcome is worth building and whether you own it.

So pick a lane on one real workflow this quarter. Either your product becomes the agent that owns the outcome, or it becomes the clean API an agent calls, or you name it as exposed and move budget. Waiting to see how it shakes out is a choice too, and it hands the outcome to whoever moves first.

Three questions for your team

1. Pick your top workflow. If an agent did it end to end, would users still open our interface? If not, are we the agent or the API it calls?
2. Prentis charges 20% of savings, not per seat. Could we price one feature on the outcome it delivers instead of access, and what would that require us to prove?
3. Where does our real defensibility sit now: the interface we designed, the data we hold, or the judgment about which outcome matters? Name it, because the interface is the least safe answer.