For most of its public life, artificial intelligence has been a talker, not a doer. You ask a question, it gives an answer; you type a prompt, it spits out text. That is useful, but it’s also limited. The real opportunity, as a growing number of founders and investors see it, is to build AI that doesn’t just chat but actually gets things done—booking flights, negotiating bills, chasing down unpaid invoices, even helping run a shrimp farm. That is the shift captured by Forbes’ inaugural Agentic 20 list, a curated snapshot of the companies and projects pushing AI beyond the chatbot era and into the messy, unglamorous work of the real economy. The list is not a ranking. There is no scoring methodology, because the market is still too young and too weird to be formulaically judged. Instead, it is the result of on-the-ground reporting, conversations with founders, users, and investors, and a willingness to follow the hype and then look for signs of actual use. What emerges is a picture of an industry that is simultaneously dazzling and strange: a 14-person startup valued at $10 billion, an AI matchmaker that wants to hear about your exes, agents that negotiate with insurance companies, and software that answers funeral home calls in the middle of the night.
Some of the most visible action is in consumer AI agents, where the goal is to make an assistant that feels less like a search bar and more like a helpful, slightly obsessive friend. Instinct, arguably the hottest name in Silicon Valley right now, is a personal AI assistant that lives inside iMessage and WhatsApp rather than in a separate app. It has its own email address, a private browser, and the ability to make phone calls, and it is already handling about $1 billion in annual transaction volume, mostly travel-related requests, according to its 23-year-old CEO Noah Shinn. The company reached a $10 billion valuation before even releasing its agent to the public, and it raised $1 billion from Sequoia, Benchmark, and Coatue after a compute crunch made it a victim of its own popularity. Meta’s answer, an agent called Muse, took a different path: it does the standard chores like booking flights and sending emails, but it won early fans by tackling personal finance. Early users used Muse to negotiate bills, chase refunds, and kill forgotten subscriptions, sometimes recovering thousands of dollars. It crossed five million downloads in three weeks, helped by aggressive promotion across Facebook and Instagram and by a cute, customizable mascot named Jolly. There is even a Tamagotchi-style device called Muse Charm so people can carry their agent around without a phone. A smaller but no less interesting player, Wajo, takes a more grounded approach with an assistant called Fo that can research plumbers, call for quotes, and schedule visits. When the model gets stuck, trained human contractors quietly step in to finish the job, so the user never has to watch a frozen app spin its wheels. The company is betting that people don’t care whether an AI agent did everything on its own; they just care that the job got done.
Beyond errands and entertainment, a wave of startups is selling AI agents as coworkers, employees, and even managers. Town, for example, offers “townies,” automated assistants aimed at small business owners, solo entrepreneurs, and sales teams who live and die by the inbox. For a restaurant owner, a townie might monitor supplier price lists; for a recruiter, it might follow up with clients and compile candidate reports. Town charges $50 to $200 per user per month, and it tries to make businesses comfortable by promising not to train on customer data and by deleting most information it doesn’t need after 15 days. Andon Labs is going even further, with AI agents that are currently running an entire store in San Francisco’s Marina district—writing job postings, interviewing applicants, and managing real human employees. The store is not profitable, and CEO Lukas Petersson admits that’s not really the point. The point is proving that agents can handle the messy, frustrating work of management. Andon’s agents are also running a cafe in Sweden and four radio stations, and thousands of companies from more than 40 countries have joined the waitlist, with proposals ranging from farms and hotels to a bank, a trucking company, and graveyard maintenance. Viktor, another ambitious player, describes its systems as AI employees rather than AI assistants. You give it a revenue target, and it works across Slack, Teams, and other business tools to make it happen, writing code, producing reports, and scheduling recurring tasks. The company says it hit a $50 million annualized run rate in less than seven months, which is either a sign of real demand or a reminder that the term “AI employee” is being stretched in creative ways.
The list also shines a light on agents in verticals where you might not expect to find cutting-edge software, and where the stakes are deeply human. TeamTangoAI, for instance, is the brainchild of Jimmy Lucas, a sixth-generation funeral director who owns 13 funeral homes in the Dallas–Fort Worth area. His AI agents answer calls, texts, and website chats that would otherwise fall to staff or after-hours answering services, and each funeral home decides how much the agent can handle. One rule is non-negotiable: if someone asks for a real person, the AI transfers them immediately. The goal, Lucas says, is not to replace funeral directors but to free them to spend more time with grieving families. In a similarly sensitive area, Salient uses AI agents to argue with insurance companies after a car is totaled, gathering claim files, researching vehicle values, and negotiating settlements. The same machinery handles credit card disputes, debt collection, and loan servicing, and the company says its customers include Westlake Financial, AutoNation Finance, Exeter Finance, Ally, and three of the ten largest U.S. banks. Govra is working the other side of the counter, building AI agents to help government agencies conduct tax audits. California’s tax department is already piloting the software, and Indiana officials see the potential to conduct 20 to 30 percent more audits than they do today. It sounds scary, CEO Josh Sandler admits, but his pitch is that catching tax cheats faster is actually good for people who pay their taxes. Then there’s Atarraya, whose AI agents help workers run shrimp farms in shipping containers, turning complex shrimp biology into daily tasks and reports. The company says it trained former construction workers to become shrimp farmers in just three weeks, which is either a beautiful example of human-AI collaboration or a reminder that the future of work is stranger than we imagined.
For all the talk of agents taking action, none of it would work without what happens underneath the hood, and some of the most valuable companies on the list are building the plumbing. Browserbase doesn’t want to be an AI assistant; it wants to be the browser that AI assistants use. As agents zip through websites to research, shop, and file taxes, they need infrastructure to load pages, click buttons, and enter information, and Browserbase sells that layer to other agent companies, claiming to have automated more than 2,000 years’ worth of browsing. Exa is building a search engine specifically for agents, one that lets them pull background information on a prospective customer, trawl technical documentation, or find code examples, rather than just repackaging Google results. The company says more than 500,000 developers and 5,000 companies are building with its tools, and it raised $250 million at a $2.2 billion valuation in May. Stripe, the payments giant, is positioning itself as the financial safety net for agentic commerce. Its Link product issues a virtual one-time-use credit card for every approved purchase, so an agent can complete a transaction without ever seeing the user’s real card details. It also offers protections for loss, theft, and price drops. The strategy is working: Meta’s Muse launched with Link, Instinct and Town use it too, and Stripe says agentic purchases using Link increased 38-fold in the past month. Nous Research, a Brooklyn open-source lab with an anime-style logo and a pseudonymous chief marketing officer known for blowing up a Lamborghini as an anti-crypto stunt, is betting that big enterprises will want to run agents on their own servers, keeping data away from OpenAI and Anthropic. Its open-source framework, Hermes, is already being rolled out to employees at some Fortune 500 companies, even if it doesn’t exactly look like a typical procurement deck.
Then there are the breakout names that seem to embody both the promise and the chaos of this moment. SpaceXAI, Elon Musk’s entry, started as an internal agent to help with bug fixes, onboarding new hires, and marketing campaigns before it was released to the public in August. It works as a team of Grok Bots that collaborate on their own virtual computer, overseen by a “chief of staff” agent, and one employee described the experience as having “eight arms like an octopus.” It passed 400,000 weekly users within a month. OpenAI, not to be outdone, launched Dots, always-on personal agents that can juggle tasks in the background, like making design changes on your website while ordering dinner, drawing on connections to more than 4,000 apps and plugins. They are cute, colorful, and customizable, much like Meta’s Jolly, and they’re available to Pro users who pay at least $100 a month. In the security world, XBOW is selling AI red teamers: agents that probe customer applications, find vulnerabilities, and build working exploits to show what an attacker could do, with more than 200 customers and over 14,000 exploits found. Element451 is bringing agents to college admissions, where Bolt agents act as AI admissions consultants, reading applications, evaluating transcripts, nudging students to finish forms, and flagging fraud. Stuut is using agents to chase unpaid invoices across email, text, and phone, handling the awkward follow-ups that finance teams hate, and the company says it has moved more than $3 billion through its platform. And Known, an AI-powered dating app, starts with a 20-minute interview about your past relationships and then introduces you to one person at a time, with a tailored explanation of why it might work. There is no endless queue of profiles; if you decline, it goes back to searching. The company charges $15 per date it schedules, and says 60 percent of its introductions are accepted.
What ties all of this together is a simple but profound shift in how we think about software. For decades, computers have done what we tell them, step by step, and only after we learned their language. Agentic AI promises something different: systems that can take initiative, make judgment calls, and handle the work that nobody wants to touch. That is exhilarating and unsettling at the same time. It raises questions about trust, privacy, and accountability that the industry is only beginning to grapple with. It also invites a healthy dose of skepticism, because the hype cycle is real, and not every “agent” is as autonomous as it claims to be. But the underlying direction is hard to argue with. The companies on this list are not just betting that AI can write code or summarize documents; they are betting that AI can show up, do the unglamorous grunt work of the economy, and make life a little easier for the rest of us. Some of them will fail, and some of them will pivot into something unrecognizable. But the best of them will be measured by a simple standard: did the job actually get done? That is the real breakthrough. We are moving from an era of asking AI to tell us things to an era of asking AI to make things happen—and in the process, we are learning just how much of the world can be handled by a patient, tireless digital assistant that never sleeps, never complains, and always picks up the phone, even if the person on the other end is calling a funeral home at 3 a.m.
