A year ago, the conversation around artificial intelligence was loud with predictions that programmers and white-collar workers would soon be replaced by machines, yet the technology still fumbled the simplest everyday tasks. It couldn’t reliably book a dinner reservation, and even the biggest tech companies were struggling to build an assistant that could actually use your apps without creating more work for you. Apple quietly delayed the Siri upgrade that would have allowed it to take action inside apps, and software chief Craig Federighi gave a brutally honest explanation: “This just doesn’t work reliably enough to be an Apple product.” Amazon’s upgraded Alexa+ did manage to book a reservation when a reporter tested it in July 2025, but there was a catch—the request was for Friday, and the agent booked Thursday. That kind of near-miss captured the frustration of the moment. AI could generate poetry, write code, and pass law exams, but it couldn’t be trusted to get a simple date right. Now, though, a new generation of assistants is finally getting those chores done well enough that handing off an errand doesn’t feel like taking on a second job babysitting the AI. Forbes catalogued some of the most interesting in its inaugural Agentic 20 list, and the shift in feel is unmistakable. The technology has moved from being a clever toy to something closer to a reliable personal assistant, at least for certain tasks. It’s not perfect, and there are still plenty of ways it can trip over itself, but the direction is clear, and the momentum is real.
At the center of this shift is a tiny, invitation-only startup called Instinct, which has become an obsession for Silicon Valley’s venture capital class. Investor Sarah Guo, whose firm Conviction was an early backer, described using it for restaurant reservations, marathon entries, and camping bookings. The whole process was simple: you connect your accounts, send a text request, and the assistant handles the rest. That simplicity caught fire. In early August, a $75 million Series A valued Instinct at more than $500 million. By August 26, a $250 million Series B had pushed that valuation to $2.5 billion. Then, on September 28, the company announced another $1 billion from Sequoia Capital, Benchmark, and Coatue, bringing its valuation to $10 billion. All of this with just 14 employees. The founder, 23-year-old Noah Shinn, says the startup earned its reputation by doing something AI has long promised but rarely delivered: actually getting things done right. “The first time I really started to get excited about the product we were building, I had a very simple realization: Instinct just works,” he told Forbes, borrowing Apple’s iconic marketing line. “Our goal was to build a personal assistant, and Instinct behaves in a way we’ve been waiting for AI to deliver on since its inception.” Shinn says Instinct’s user base is growing an average of 10% daily, despite the invitation-only beta, with 80% retention among users who connect accounts like Gmail. The numbers are striking for a product that is still so young and so exclusive, but they suggest that once people experience an assistant that actually follows through, they don’t want to go back.
The bigger companies have started fielding competitors as well, and they are seeing explosive adoption too. Meta launched its assistant, Muse, in early September with a cuddly mascot named Jolly. It now has more than 5 million downloads and overtook ChatGPT as the number one app on the App Store on September 18. Meta’s stock has climbed 27% since the launch, and Forbes reported a single-day gain of more than $25 billion to Mark Zuckerberg’s net worth. OpenAI, meanwhile, launched Dots late last month—cartoonish assistants designed to be “always on,” working on your behalf and even suggesting other tasks to tackle. So why now? Jean-Denis Greze, co-founder of the work-assistant startup Town, credits stronger AI models that arrived late last year, along with better tools for operating browsers and computers. By summer, those tools had become cheap enough for everyday use. Cheaper voice technology has also made AI phone calls practical, which matters because many real-world tasks still require talking to a human on the other end of the line. The combination of better reasoning, more reliable computer use, and affordable voice interaction has finally pushed these agents over the threshold from demo-worthy to genuinely useful. It’s the difference between an assistant that can theoretically navigate your phone and one that can actually navigate the messy, unpredictable world of websites, verification codes, and customer service bots. The infrastructure isn’t glamorous, but it’s what makes the magic possible.
The proof is in what people are actually asking these agents to do. Travel is the standout use case, according to Shinn. Instinct is on track for $1 billion in annualized transaction volume, and travel accounts for half of that. Beyond travel, the tasks are varied and mostly mundane—booking rides, buying groceries, and handling the small administrative chores that clutter up a life. But some users have gotten more creative. One person uploaded photographs of themselves and their entire wardrobe, turning Instinct into a personal stylist that produces daily outfits, complete with generated images of how they will look. Another user, entrepreneur Matt Van Horn, told Forbes in September that Instinct helped recover about $2,000 from old insurance plans and a dormant bank account. The assistant submitted claims after Van Horn secured permission to use his signature. Brianna Guay enlisted an assistant for her sister’s bachelorette party. Her “Bach Bot,” built with xAI’s personal assistant Grok Bot, helped source limo drivers, coordinate vendors, and track deliveries. It even inspected her Amazon order to see how many vases she had purchased when coordinating with florists for a flower purchase. These stories are small, but they illustrate the range of what the new assistants can do. They’re not just answering questions or generating text; they’re operating in the real world, navigating accounts, making decisions, and following through on multi-step tasks. The best part, for users, is that they don’t have to hover over the process. The whole point is delegation, and for the first time, the delegation doesn’t feel like a gamble.
None of this means the technology has become flawless. In August, investor Jason Yeh said on X that Instinct booked a restaurant with a $200 cancellation fee when he had asked it to find available reservations. Context is still important, and the AI can miss the obvious. Greze acknowledges that reliability remains uneven, and some websites, notably Amazon, resist access by agents. When the AI hits a wall, some companies have turned to an old-fashioned solution: humans. Wajo AI’s fallback is to use trained contractors who intervene when its personal AI agent, Fo, encounters an obstacle like a website verification check or a messy hotel negotiation. The idea is to finish the errand instead of punting it back to the person who delegated it. “Customers want the job done,” says CEO Shivani Poddar. “They don’t want a ball dropped on them.” That hybrid approach acknowledges the reality that AI is not yet good enough to handle every edge case, but it also recognizes that the user experience should be seamless. If the AI can’t do it alone, a human behind the curtain can step in, and the user never has to know or care. This is a pragmatic way to build trust, even if it’s not as pure as a fully autonomous system. The goal, after all, is not to prove that AI can do everything; it’s to make sure the job gets done.
Trust remains the deepest question, and there have been a few incidents that rattled users’ confidence. In September, an Instinct user complained that the agent appeared to reveal another person’s information. Shinn said the agent had invented the information rather than leaked it. In another incident, Muse user Matt Robb said the agent shared his address with a Facebook Marketplace buyer who then showed up unexpectedly. Robb later acknowledged selecting an “Allow Always” setting but said he had expected to approve offers. Meta said there had been no privacy breach. These episodes highlight the anxiety that comes with giving an AI access to your accounts, your location, and your personal data. Observant rivals are pitching privacy alongside convenience. Town, whose CEO has a background in financial services and banking, says it does not train on users’ data and deletes most information after 15 days. Poddar says Fo is designed to prevent private conversations from spilling into group chats. Shinn is already asking users to delegate even more. Instinct’s new network lets users link their agents with those of their friends and family, to handle coordination between people, like booking meetings. That means persuading people to trust software with personal information and tasks more consequential than buying flowers. “People have a little bit of agent fatigue,” Poddar says. “But also, they don’t have something they can truly trust to get shit done.” The next chapter of this story will be defined by whether these assistants can cross that final gap—from being clever and convenient to being worthy of real responsibility. If they can, the way we manage our daily lives may look very different. If they can’t, all the funding and hype will be just another reminder that reliability is the hardest part of AI.
