From three-person startups hitting $500K ARR to Meta’s CEO avatar project, AI agents are reshaping what a “team” looks like.

A recent Fortune story on Fathom AI shows what that looks like. The Austin team started this year with three people and $300 of their own money. Three months in, they were at $300,000 ARR. One client, Tiger Aesthetics, hadn’t opened a single new account in all of 2024. After adopting Fathom, they opened 225 in one quarter.

The founders lean on 12 agents baked into daily operations. One runs customer success for a national sales force. Another scans the competitive landscape every few hours. The CEO comes out of sales, not software, yet with this structure he was able to walk into the field on day one with his own automated system and operate like a full team.

Fortune also covered KNOWIDEA, a three-person company with a similar shape. The CEO, Yatharth Sejpal, is 23 and has never written code. In six months, his team signed six enterprise customers and hit $500K ARR, with a strategic investment at a $15M valuation.

Taken together, these cases tell a simple story. A handful of people are using AI agents as real teammates, not as side projects, and the result is the kind of impact and efficiency that used to require whole departments and big budgets.

Now look at the other end of the spectrum.

Recent reporting on Meta describes a project to build a highly realistic AI version of Mark Zuckerberg that can sit in for him with employees. The company is feeding this system with his public remarks, his way of speaking, and his current thinking on strategy, so that spending time with it feels as close as possible to talking directly to the founder. Alongside that, there is a separate “CEO agent” idea focused on helping him pull up information quickly and support his work running a $1.6 trillion company.

Inside the company, staff are being pushed closer to AI as well. People are encouraged to design their own agents to automate internal work, including by using open tooling. Product managers are being asked to go through an internal AI “baseline”, with technical design questions and a section the company calls “vibe coding.” Some see this as an honest attempt to build skills. Others look at the same exercises and quietly wonder: is this about helping me grow, or about deciding who is expendable.

One giant company trying to capture a single leader’s judgment and broadcast it across thousands of people. Three-person teams building whole companies out of software agents. Different contexts, same ingredient: systems that understand some slice of your world and go do the work.

The question is what happens when that ingredient moves into the hands of everyone else.

We started Emergent to help founders and business owners turn ideas into working software by leaning on agents behind the scenes. If you have ever launched anything, you know shipping is not the finish line. It is the starting gun. What wears you down is everything that comes after: the customer threads sitting in your inbox, the prospects you meant to follow up with, the internal tasks that never quite make it out of your head, and the personal chores that stack on top. Most people I talk to live with a low‑level sense that something important has slipped, they just do not know what it is yet.

That is why, over the past year, I keep circling a simple thought about the applicability of agent AI for everyone. If an agent can help a public‑company CEO keep up with their job, and it can help a three–person team behave like a much larger organization, why can’t it do the same for an ordinary Tuesday? It does not need a glossy interface. It needs to sit in WhatsApp, Telegram, iMessage, and quietly help you do what you said you would do.

Building an agent like that for everyday use forces you to make different choices than you would for an internal CEO avatar. The customers we speak to aren’t interested in questions about models or benchmarks. They ask, very directly, if it is going to wipe their inbox, charge the wrong ticket to their card, or send a strange message to a client. Those worries are not hypothetical. They come from stories they have seen and heard.

If you want people to leave an agent switched on, you have to respond to that honestly. Our way of doing that is to draw clear boundaries around what the agent can do on its own. On one side of that line sit the tasks it can just get on with: drafting responses, tidying your inbox, pulling together context for a meeting, keeping your contacts in step with your email and calendar. On the other side are the changes that actually commit you to something. For those, it should check in: are you sure you want to send this, update that record, book this trip. Only after a long run of getting the basics right should it be allowed to take on more, and only because you explicitly chose to widen the fence.

Seen this way, the notable shift is not that a few tech CEOs are playing with AI replicas of themselves. It is that the same kind of personalized help is starting to look like something anyone could use. The logic behind a “Zuckerberg agent” is not that far from the logic behind an agent that lives next to your family group in a chat app, keeps track of your work, and taps you before you drop the ball. The real difference is whose day it touches.

Fathom and KNOWIDEA show how little headcount you need once agents are part of the team. Meta shows how far a large company will go to bottle and scale one person’s presence. Emergent is trying to bring those ideas to the founder in Bangalore, the salesperson in Austin, the consultant in Berlin. People who sit between those extremes and want to do right by their customers without losing their life to the job.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.