The Agentic 20: 20 AI Agents Already Doing Real Work, From Funeral Homes to Shrimp Farms

For years, AI’s signature trick was conversation: you asked, it answered. That era is ending. The newest wave of software books the flight, calls the plumber, chases the unpaid invoice and negotiates with the insurance company, all without waiting for a human to click “send.”

Forbes has now put a name to the trend with its first-ever Agentic 20, a list of companies whose AI agents take action rather than simply produce text. It isn’t a scoring ranking. Forbes built it from reporting, founder and user interviews, and conversations with investors, then checked for signs of genuine usage. The result is a snapshot of where agentic AI is heading in late 2026, and the destinations are surprising.

Here are all 20, grouped by what they do.

Part 1: The Personal Assistants Everyone Is Talking About

1. Instinct: the $10 billion agent nobody can download yet

Instinct may be Silicon Valley’s hottest agent startup, and it hasn’t even launched publicly. Still in invite-only beta, it already reached a $10 billion valuation and, according to 23-year-old CEO Noah Shinn, handles about $1 billion in annual transaction volume, mostly travel. The agent lives inside iMessage and WhatsApp and comes equipped with its own email address, a private browser and the ability to place calls. Demand was so heavy it created a computer crunch, which led to a $1 billion raise from Sequoia, Benchmark and Coatue.

2. Meta Muse: the comeback play

Meta’s Muse covers the usual chores, but personal finance is what made it take off. Early users had it negotiate bills, chase refunds and cancel forgotten subscriptions, sometimes recovering thousands of dollars. After its early September launch, it passed five million downloads in three weeks. Free access, heavy promotion across Facebook and Instagram, a customizable mascot named Jolly and even a physical companion gadget, Muse Charm, all helped.

3. OpenAI Dots: always on, always working

OpenAI already has plenty of agents, but Dots is built to stay switched on in the background, spotting tasks and tackling them without being asked. One Dot might tweak your website while ordering your dinner. It connects to more than 4,000 apps and plugins and comes in a friendly, colorful, customizable style. Forbes notes it doesn’t differ wildly from OpenAI’s existing lineup. It is available on Pro plans starting at $100 a month.

4. Wajo AI’s Fo: humans in the loop, on purpose

Wajo’s bet is that people care about whether the job gets done, not whether a machine did every step. For research, plumbers, calls for quotes, checks your calendar and books the visit. When it hits a wall, such as a verification screen or an awkward phone call, trained human contractors quietly step in. Founder Shivani Poddar, who previously worked on assistant products at Amazon, Meta and Google, says Fo has tens of thousands of users, with growth doubling every two days after launch. Privacy controls are a core pitch.

5. Known: the AI matchmaker that interviews you first

Known skips the endless swiping. New users complete a roughly 20-minute interview about past relationships, preferences and goals. The app then introduces one person at a time, with a reason the match might work, and acts as a first-date concierge for $15 per date. CEO Celeste Amadon says 60% of introductions are accepted and thousands of dates have been scheduled. It operates in the Bay Area and San Diego, with more cities coming.

Part 2: AI Coworkers and Autonomous Operators

6. Town: the coworker that lives in your inbox

Town’s “townies” target small and midsize businesses where a missed email means missed money: a restaurant owner watching supplier price changes, or a recruiter following up with clients. Unlike free consumer assistants, it charges $50 to $200 per user monthly. Its privacy stance is a selling point: no training on customer data and most unneeded information deleted after 15 days. Paying customers have grown ninefold since July.

7. Viktor: an “AI employee” with a revenue target

Viktor wants you to hand it a goal rather than a to-do list. Working inside Slack and Microsoft Teams, it connects to business tools, writes code, produces reports and schedules recurring tasks across conversations. Many startups use the “AI employee” label, but Viktor says it reached a $50 million annualized run rate in under seven months.

8. SpaceXAI’s Grok Bot: an octopus with eight arms

Grok Bot began as an internal tool at SpaceX, fixing bugs, seating new hires and spinning up marketing campaigns. Since August it has been public. Multiple bots work in parallel on a shared virtual computer, supervised by a “chief of staff” agent. It passed 400,000 weekly users within a month of launch.

9. Nous Research’s Hermes: the rebel enterprises are quietly adopting

This Brooklyn open-source lab has an unconventional image, but Hermes, its agent framework, lets companies host their own agents, pick their own models and keep their data in-house. Nous says Fortune 500 firms are already rolling it out to employees. The open question is whether big enterprises will fully trust a company known as much for provocative art as for software.

10. Andon Labs: AI that runs the whole business

Andon Labs lets agents manage actual companies. They run a store in San Francisco’s Marina district, writing job postings, interviewing applicants and managing human staff. They also operate a cafe in Sweden and four radio stations. The store isn’t profitable, but CEO Lukas Petersson argues the goal is proving agents can handle messy managerial work. Thousands of companies from over 40 countries have joined the waitlist.

Part 3: Agents in Unexpected Industries

11. TeamTangoAI: gentle automation for funeral homes

Jimmy Lucas, a sixth-generation funeral director who owns 13 funeral homes in the Dallas-Fort Worth area, built TeamTangoAI to handle calls, texts and web chats so staff can spend more time with grieving families. Each home chooses how much the agent handles, and one rule is fixed: anyone who asks for a human gets one immediately. The company is pre-revenue and not venture-backed.

12. Atarraya: shrimp farming, run by software

Atarraya’s agents help workers operate shrimp farms inside shipping containers, turning farm data into reports and daily tasks. The company says it trained former construction workers to become shrimp farmers in three weeks. Its Shrimpbox Air version adapts the system to industrial buildings. It has raised $22 million and is expanding in North America and the Middle East.

13. Govra: AI on the government side of the counter

While many tools help people file taxes and fight fines, Govra helps agencies run audits, with human officials making key calls. California’s tax and fee agency has piloted it, and Indiana is preparing its own pilot, with an official there seeing room for 20% to 30% more audits. CEO Josh Sandler argues faster detection of cheats benefits compliant taxpayers, and projects recoveries across agencies could reach a trillion dollars a year.

14. Element451: admissions help for colleges

Element451’s Bolt agents answer questions about programs and financial aid, nudge applicants to finish their forms and personalize follow-up based on what each student has or hasn’t done. Behind the scenes, they read applications, evaluate transcripts and flag fraud such as plagiarism. The company’s broader platform is used by more than 400 institutions.

15. Salient: negotiating with insurers and servicing loans

Salient’s agents can gather a claim file, research a vehicle’s value and argue with an insurer after a car is totaled. They also handle loan paperwork, credit-card disputes and collections, reading contracts and policies to choose the best next move. Customers include Ally, AutoNation Finance and three of the ten largest U.S. banks, and annualized revenue tops $40 million.

16. Stuut: making sure companies get paid

Stuut’s agents chase invoices across email, text and phone, then go further into “order-to-cash”: investigating billing disputes, judging whether deductions are justified and matching payments to invoices. Customers include Honeywell and ZoomInfo, for which it says it collected over $21 million. More than $3 billion has flowed through its platform.

17. XBOW: the AI red team

Finding a possible vulnerability is easy. Proving it can be exploited is what security teams pay for. XBOW’s agents probe applications, chain weaknesses together and build working exploits. It serves more than 200 customers, including major banks, and says it has uncovered over 14,000 exploits in real customer applications.

Part 4: The Infrastructure Powering It All

18. Stripe Link: letting agents spend without seeing your card

Browsing and filling a cart is the easy part. Checkout is the risky one. Stripe’s Link issues a one-time virtual card for the exact amount when you approve a purchase, so the agent never sees your real card details. Stripe also added protections for loss, theft and price drops. Muse, Instinct and Town use it, and Stripe says agentic purchases through Link grew 38x in a month.

19. Browserbase: the browser built for bots

Agents need infrastructure to load pages, click buttons and complete transactions. Browserbase sells that layer to other agent companies, powering research, filing and shopping. It says its platform has automated more than 2,000 years of browsing.

20. Exa: a search engine for machines

Before an agent can act, it must understand what it’s looking at. Exa built its own crawler, models and databases specifically for agents. More than 500,000 developers and 5,000 companies build with it, and in May it raised $250 million at a $2.2 billion valuation.

What the Agentic 20 Tells Us

Agents are leaving the lab for the “boring” economy. Funeral homes, invoices, admissions, tax audits and shrimp tanks aren’t glamorous, but they are full of repetitive, high-volume work, and that is where agents earn their keep.

Trust is the real product. Town’s data-deletion policy, Wajo’s human backup, TeamTango’s “humans on request” rule and Stripe’s one-time cards all answer the same question: how do you hand real tasks to software safely?

Human-in-the-loop is a feature. Even the most ambitious players keep people involved where stakes are high, from Govra’s officials to Wajo’s contractors.

The picks and shovels are booming. Payments, browsers and search for agents may prove as valuable as the assistants themselves.

Want Your Own Agent? How Cubix Can Help

The common thread across the Agentic 20 is that none of these companies simply plugged in a chatbot. Each built an agent around a specific workflow, connected it to real tools and data, and designed guardrails for trust. That takes engineering, integration and a clear view of the problem, and that is where the best AI development company comes in.

Cubix is a software development company with 18+ years of experience, building custom software, mobile and web apps, cloud solutions and AI. Its AI services include generative AI, AI chatbots and virtual assistants, NLP models, intelligent automation, recommendation engines and AI agent development, plus integrating AI into software you already run. Cubix also supports cloud and DevOps work such as MLOps, so models can be deployed and maintained reliably, and serves clients across industries including healthcare, education, hospitality and finance, with a presence in the UAE, Saudi Arabia, the UK and the USA.

Whether you want an assistant for customers, automation for repetitive back-office tasks, or an AI layer inside your existing product, Cubix can take you from idea and prototype to production.

Ready to build your own agent? Explore Cubix’s AI development services at or get in touch for a consultation.