An AI that writes code is useful. An AI that writes code is useful. One that answers a funeral home’s overnight calls,

negotiates with suppliers or helps businesses handle the grunt work nobody wants to touch is useful to a much larger market.

That’s the shift behind Forbes’ inaugural Agentic 20 list. AI agents—systems that can take action on our behalf, not just spit out answers—are moving beyond software triage and calendar management into the messier work of the real economy. Think booking flights, coordinating playdates, and, yes, even helping manage funeral homes.

The list highlights 20 companies and projects making a dent in the agentic AI world, from breakout personal assistant AIs like Instinct—a 14 person, $10 billion company run by a 23-year-old—to the infrastructure companies that help agents browse the web and transact. Alongside the breakout names, we looked for strange and unexpected uses (shrimp farming!) and other corners of real-world work where “AI assistant” means more than just a chatbot’s to-do list.

This is not a formula-driven ranking. There is no scoring methodology for a nascent market that is still sorting itself out. The list draws on on-the-ground reporting, interviews with founders and users and conversations with investors and others across Silicon Valley. We followed the buzz, then looked for signs of actual use. The result is a snapshot of the agentic AI companies worth watching now.

Product: AI personal assistants

Instinct is arguably the hottest AI agent startup in Silicon Valley. Proof: The company behind it reached a $10 billion valuation without even releasing its agent to the public. Still in an invite-only beta, Instinct already handles $1 billion in annual transaction volume, most from travel-related requests, CEO Noah Shinn, 23, says. The agent lives in iMessage and WhatsApp instead of its own dedicated app, making it easy for use for everyday tasks and simple errands like booking flight tickets and tracking orders. It has all the tools it needs to do a good job: its own email address, a private browser and the ability to make calls. The agent became so popular that the startup faced a compute crunch right out of the gate, prompting it to raise $1 billion from VC bigwigs Sequoia, Benchmark and Coatue.

Meta’s personal AI agent Muse might just be the comeback the social media giant needs to get ahead in the AI race. It does the standard demo reel chores—book flights, send emails and shop for groceries—but the use case that won tons of early traction was personal finance. Early adopters used Muse to negotiate bills, chase refunds and kill idle subscriptions, in some cases recovering thousands of dollars. Muse quickly went viral after its early September launch, crossing five million downloads in just three weeks and beating rival bots like SpaceXAI’s Grok and Anthropic’s Claude. Aggressive marketing and free usage might be one reason for rapid adoption:. The company allocated up to 50% of its daily house ad inventory to promote the app across Facebook and Instagram. A cute, cuddly, customizable mascot dubbed Jolly might be another. Meta even launched a Tamagotchi-style device called Muse Charm so people can interact with their agent without their phones.

Wajo is making a calculated bet: People don’t care whether an AI agent finished the job, only that the job was done. Its assistant, Fo, can research plumbers, call for quotes, check your calendar and schedule a visit. When it gets stuck, trained human contractors step in—clearing a website verification, finishing a call, or handling the awkward bit the model cannot—without leaving the user staring at a frozen app. Fo only launched weeks ago, but investors told Forbes they were impressed by founder Shivani Poddar, TKAGE, who worked on AI assistant products at Amazon, Meta, and Google and led engineering for Jules, Google’s autonomous coding agent. Fo’s pitch isn’t so much “fully autonomous,” as it is “reliable enough that you don’t have to worry how it happened.” Privacy is another selling point: Poddar says the system is designed to keep private conversations from spilling into group chats and to require permission before personal information is shared. As of early October she said Fo had tens of thousands of users, with its user base doubling every two days after public launch.

Andon Labs wants its AI agents to run entire companies. Currently, they are running an entire store in San Francisco’s Marina district, where they write job postings, interview applicants and manage real human personnel. They are not exactly crushing it. CEO Lukas Petersson says the store is not profitable. But he argues the point is not profitability, it’s proving that agents can actually handle messy, real-world managerial work. Andon Labs’ AI agents are also running a cafe in Sweden and four radio stations, one called “Grok’n Roll Radio.”’ Petersson says Andon opened a waitlist a month ago and has already drawn interest from thousands of companies from more than 40 countries, with about half looking to hand over existing businesses and half hoping to build new ones. Proposals range from farms and hotels to apps with tens of millions of users, a bank, a trucking company and even graveyard maintenance.

Before Known finds you a date, its AI wants to hear all about your exes. Users of the AI-powered dating app start with a roughly 20-minute interview about past relationships, romantic preferences, ambitions and the other material dating apps reduce to prompts and pics. Then Known does something old-fashioned: It introduces one person at a time, with a tailored explanation for why the match might work. Decline an introduction, and there’s no endless queue of profiles to browse. Known goes back to searching. When two people accept a match, Known acts as a first-date concierge, charging $15 per date, and choosing a location, just like a real matchmaker. CEO Celeste Amadon, who dropped out of Stanford University as a junior and has raised $10 million for Known so far, told Forbes that 60% of introductions on Known are accepted, and that it has scheduled thousands of dates. While Known currently only operates in the Bay Area and San Diego, the company is planning a nationwide expansion and says it is debuting in three or four new cities imminently.

Product: AI agents for work

Town is selling AI as a coworker. Its automated assistants, called “townies,” are aimed at businesses where a missed email can mean missed revenue: solo entrepreneurs, small business owners and sales teams at small to midsize companies. Its CEO, Jean-Denis Greze, says the sweet spot is work that lives in the inbox and dies without follow-up. For a restaurant owner, that might mean monitoring suppliers’ price lists for increases; for a recruiter, following up with clients and compiling candidate reports. Unlike free consumer assistants like Muse or Instinct, Town charges $50 to $200 per user per month for its service. A key selling point: what it doesn't keep. Greze says Town doesn’t train on customer data, deletes most information it doesn't need long-term after 15 days and does its best to access data through connected software rather than storing its own copy. These policies are designed to make companies comfortable handing Town sensitive work, not just making dinner reservations. Launched in June, Town says its paying businesses have grown ninefold since July.

While many AI companies are pitching agents to help you fight tickets, file taxes and fill prescriptions, Govra is working the other side of the counter. Its software is designed to streamline government audits, with human officials calling the shots on key decisions. California’s Department of Tax and Fee Administration has already partnered with Govra on a pilot. Indiana is preparing another focused on businesses, and an official there sees the potential to conduct 20% to 30% more audits with its help. Govra is betting that as taxpayers become more comfortable using AI to navigate bureaucracy, governments will need agents of their own. CEO Josh Sandler acknowledges that the prospect of more tax audits sounds scary, but his pitch is about fairness as much as efficiency. He argues that if governments are able to more quickly find tax cheats, compliant taxpayers win. Sandler predicts potential recovery across federal, state and local tax agencies to be up to a trillion dollars annually—money that can be used to support public services, reduce debt or even make room for tax cuts.

OpenAI was never short on AI agents. It had Codex and ChatGPT Work. After Meta’s Muse took off, the AI behemoth added one more to the mix. Dots are meant to be personal AI agents that are “always on” and juggling tasks in the background. One could make design changes on your website while ordering dinner for you, for example. And with the ability to connect with over 4,000 different apps and plugins, they can keep an eye on things, spot tasks to take on and work on them without being explicitly told to do so. Taking a page out of Jolly’s book, they’re cute, colorful and customizable. That said, they’re not wildly different from OpenAI’s existing product suite. Dots is available for people using OpenAI’s Pro plans starting at $100 per month.

Nous Research does not look like a standard enterprise AI vendor, which is partly the point. The Brooklyn open-source lab has a pseudonymous chief marketing officer, the artist SHL0MS, known for stunts like blowing up a Lamborghini as an anti-crypto statement. Its logo: an anime-style woman with the phrase “rebellion to tyrants is obedience to God.” Not exactly procurement deck material. Yet Nous is fast becoming one of the leading options for large enterprises looking to roll their own internal agentic solutions without giving up their data to OpenAI, Anthropic, or another outside provider. Its open-source agent framework Hermes, started as a tool for technical users who wanted to host and run their own agents, choose their own models and keep their own data. Now, Nous is pushing into enterprise deployments, offering businesses the ability to run agents and models on their own servers. SHL0MS says Fortune 500 companies are already rolling Hermes out to employees. The open question is whether big enterprises will trust a company known more for artistic provocation than corporate convention.

Product: AI agents for shrimp farms

Atarraya’s AI assistant isn’t here to book meetings. It helps workers run shrimp farms in shipping containers. Overseeing its shrimp-growing tanks is software designed to take on the analysis and coordination that would otherwise require aquaculture specialists, mapping shrimp biology, and turning farm data into reports and daily tasks for the people working on site. The bet is that AI can make a messy biological operation manageable for workers without years of aquaculture experience. Atarraya says it trained former construction workers to become shrimp farmers in just three weeks. Its original Shrimpbox packaged the operation into shipping containers; Shrimpbox Air adapts it for spaces such as industrial buildings. The company says it has raised $22 million and is expanding in North America and the Middle East.

Product: AI sales employees

Viktor is selling something more ambitious than another AI task rabbit. Give its AI a business goal, its founders say, and it should wrangle everything to the outcome, not wait for a manager to break the work into prompts and checklists. CEO Fryd Wiatrowski describes Viktor as an AI employee, a system that a company can hand a revenue target to achieve or department to assist. It lives in Slack and Microsoft Teams, works across connected business tools, writes code and produces reports and other deliverables. Its underlying system gives it a persistent workspace and the ability to schedule recurring tasks and operate beyond a single conversation. Plenty of startups now claim to offer “AI employees.” Viktor’s sharper claim is traction: the company says it has reached a $50 million annualized run rate in less than seven months.

Product: AI agents for funeral homes

Jimmy Lucas, the CEO of TeamTangoAI, sees an opening for AI in a business his family has worked in since 1860. A sixth-generation funeral director and social worker who owns and operates 13 funeral homes in the Dallas–Fort Worth area, he’s betting that software can take some calls and messages off of staff’s hands, freeing up more time to focus on caring for families on their worst days. His startup, TeamTangoAI, answers calls, texts and website chats that previously fell to staff or after-hours answering services. Each funeral home determines how much the AI agent handles. Lucas says one rule is non-negotiable: If someone asks for a real person, the AI agent should immediately comply. TeamTangoAI is still pre-revenue and not venture capital backed, but Lucas argues that his own personal experience means that his product will be better suited for funeral homes and their customers. The goal, he says, is to free funeral directors to spend more time with families—not put them out of a job.

Grok Bot started off as an internal agent that handled SpaceX’s own needs: fixing bugs, seating new hires and spinning up marketing campaigns. In August, it was launched into the wider world to help with all kinds of professional work. A team of Grok Bots work together on a task, collaborating with each other on their own virtual computer and exchanging messages; they run in parallel. A “chief of staff” agent manages all the other bots. One SpaceX employee described the experience as having “eight arms like an octopus.” The agent topped 400,000 weekly users a month after it launched.

Agents can browse the web, compare products and fill out a shopping cart. That’s the easy part. The risky part begins when the agent reaches check out. Giving an agent your credit card is one problem. Giving it your banking information is a potentially worse one. Payment behemoth Stripe’s answer: Link, its consumer checkout product, now being pushed into agentic shopping. Link sits between an agent and your money. When you approve a purchase, Link issues a virtual one-time-use credit card for the exact amount owed. The agent can complete the transaction but never sees the real credit card behind it. Stripe has also added purchase protections for agentic shopping, including coverage for loss, theft and price drops. Meta’s Muse AI agent launched with Link, allowing people to approve purchases from directly inside Muse. In fact, the service is quickly becoming the default payment method for autonomous commerce: Buzzy AI agents Instinct and Town also use Link. Stripe said last week in a blog post that agentic purchases using Link increased 38x in the past month.

Browserbase isn’t trying to be another AI assistant. It wants to be the browser they use. As agents zip through the web for users, they need infrastructure to load sites, click buttons, enter information and complete transactions. Browserbase sells that layer to other agent companies, helping their bots to do tedious web work: research, tax filing, shopping and the many other tasks that still require a person sitting in front of a computer. The company says its platform has already automated more than 2,000 years’ worth of browsing, and its annual run rate has exceeded $25 million.

Before an agent can book the flight, draft the sales email or fix the code, it has to know what it’s looking at. Exa sells that first step, a search engine built specifically for agents. Using Exa, agents can pull background information on a prospective customer, trawl for technical documentation or write code. Rather than repackaging another search engine’s results, Exa says it built its own web crawler, AI models and databases—and that more than 500,000 developers and 5,000 companies are building with its tools. Investors have noticed. In May, the company raised $250 million at a $2.2 billion valuation.

Finding a possible security hole is one thing. Proving someone can actually break through it is the thing that companies pay for. XBOW’s agents probe customers’ applications, identify vulnerabilities and build working exploits to show just what an attacker could do, including exploiting multiple weaknesses to mount a more consequential attack. XBOW does the discovery, research and testing that human penetration testers previously handled step by step; it’s essentially an AI red team charged with finding vulnerabilities before criminals can exploit it. XBOW told Forbes it serves more than 200 customers—including major banks, Fortune 500 companies, cloud providers and software businesses—and has found more than 14,000 exploits in real customer applications.

Product: AI agents for higher ed

Element451 is selling AI agents for the least glamorous, highest-stakes part of college enrollment: reviewing the pile of applications no one has read yet. Acting as AI admissions consultants for colleges, Element451’s Bolt agents draw on each institution’s data to answer questions about programs and financial aid, nudge applicants to finish forms and follow up before an interested student slips away. Instead of sending everyone the same canned email reminders, the agents tailor their outreach to what each student has—or hasn’t—done. Behind the scenes, they read applications, evaluate transcripts and flag admissions fraud like plagiarism or falsified information. The company says more than 400 colleges and universities use its broader platform, including Utah Valley University, Wake Tech and Bridgewater State University.

Product: AI agents for loan servicing and consumer finance

Your car is totaled, and the insurance settlement looks too low. Salient’s AI agents can take over the argument, gathering the claim file, researching the vehicle’s value and negotiating with your insurer. They can also handle the paperwork and collect the payment to close out the loan. The company is applying the same machinery to credit-card disputes, debt collections and customer service. Behind the scenes, its agents read contracts and loan policies to determine the most effective move, work traditionally routed to off-shore operations centers. CEO Ari Malik told Forbes that Salient’s customers include Westlake Financial, AutoNation Finance, Exeter Finance, Ally and three of the ten largest U.S. banks, and the company is now generating more than $40 million in annualized revenue.

Product: Agents for invoice management

Stuut’s AI agents handle a blunt, but often frustrating task: making sure companies get paid. They chase invoices across email, text and phone calls, taking on the follow-ups that otherwise fall to finance teams. But the company’s ambitions extend beyond automated reminders. It targets “order-to-cash”—the frustrating stretch between a customer placing an order and the seller getting their money— with agents that investigate billing disputes, determine whether customers’ deductions are justified and match incoming payments to the correct invoices. Stuut is pursuing large enterprises, with customers including Honeywell, Verifone, ZoomInfo and NCR Voyix. For example, Stuut says it collected more than $21 million for ZoomInfo by taking over repetitive collections work. Across its platform, more than $3 billion has moved through Stuut, the company told Forbes .