The future is here.
Within the last month, a handful of new AI tools have pushed the technology past the tipping point — making it more accessible to everyone and more indispensable to those who know how to use it.
“Something big is happening,” Matt Shumer, co-founder and CEO of applied AI company OthersideAI, wrote earlier this week in a post on X that has since gone viral, attracting 75 million views and 34,000 shares.
Shumer described a before-and-after moment in his own work — the point at which AI stopped being a tool he guided and started completing complex, multi-day projects entirely on its own — and warned that the disruption will soon shape every profession.
OpenClaw, an open-source AI assistant that debuted in late January, has already amassed millions of users, and is dominating the conversations and happy hours of everyone in tech.
Meanwhile, OpenAI and Anthropic, arguably the two most prominent AI labs, each released new models on Feb. 5 that were so powerful, some in the tech space believe they can already eliminate white collar jobs like administrative assistants and junior bankers.
With these new tools, people with no coding background are now building and deploying their own AI agents to help them launch startups, book trips, and respond to emails.
“We will be a nation of bots,” predicts John Borthwick, founder of Betaworks, the venture fund that backed Tumblr and Giphy tole me. “In the future everyone will have multiple specialist agents rather than relying on one general-purpose AI.”
This story is part of NYNext, an indispensable insider insight into the innovations, moonshots and political chess moves that matter most to NYC’s power players (and those who aspire to be).
Matthias Luebken, formerly a chief product officer at a cloud management platform, left his corporate job earlier his year to launch a knowledge retention platform for retiring workers called Tavon AI. He’s built the entire company with specialized AI agents — one handles HR, one handles sales, one handles marketing. “It really feels like an assistant,” he told me. “None of this I could’ve done half a year ago.”
He’s already thinking bigger: “I feel comfortable that down the road I can manage my customers with agents.”
Adam Silverman, who runs a custom agent-building agency, used to charge clients $10,000 to $100,000 to create AI agents. Now, tools like Claude Cowork let him compress weeks of work into hours, taking on far more clients as a result.
“You’re not limited by how much you can type,” he told me.
Silverman spends the majority of his workday inside Claude the way most people live in their email — treating the $200-a-month subscription like a team of personal assistants. His bots book his flights, planned his birthday trip, find discount codes that saved him thousands, and ping him proactively with check-ins. “POTUS gets a briefing every day — I get the same thing,” he joked.
But for all the enthusiasm, these people are also cautious about giving bots too much access to their data.
The horror stories are already out there: One autonomous agent, left unsupervised with credit card access, allegedly spent nearly $3,000 on online courses after deciding the classes would help improve its performance.
At present, everyone is sandboxing their AI agents or setting strict limits on what they can access. Sandboxing means isolating AI in a locked-down environment. In other words, it can work inside its designated space, but it can’t touch important files, access the real internet freely, or make purchases on its own.
Part of sandboxing also means giving your bot a secure way to communicate — and that’s where Telegram comes in. Long popular with the crypto crowd, the messaging app has become unlikely infrastructure for the bot revolution. Its built-in tool, BotFather — a nod to the Corleone patriarch — lets anyone spin up a custom AI agent in minutes, and it’s a key reason people are choosing Telegram over other messaging apps.
Of course for all the optimism, there is also an acknowledgement that this will continue to change at an unstoppable pace. And for others its raises the question, what’s the point of learning a new skill when the AI will be able to do it better in six months?
