Actively AI

The night shift nobody clocked in for

It is 2 a.m. somewhere inside a 1,000-person revenue team, and the work is still moving. Account research is getting written. Buying signals are being flagged. A renewal that was about to slip is quietly being teed up for a human to catch in the morning. Nobody is at a desk. The work belongs to Actively AI, a New York software company that gives every sales account its own dedicated AI agent and then lets those agents run around the clock.

That is the pitch, and it is a strange one in a market exhausted by AI sales tools. Actively does not sell a chatbot that drafts emails faster. It sells the idea that a revenue team can stop being limited by how many hours its people have. The company calls this Intelligence-Led Revenue. Most of its competitors are still calling it, hopefully, "the next quarter."

Spray, pray, repeat

For a few years, the AI sales boom had a default setting: volume. Tools promised armies of automated reps who could send more emails to more strangers than any human could manage. It worked, in the narrow sense that the emails went out. It also flooded inboxes, trained buyers to ignore anything that smelled automated, and quietly proved that doing the wrong thing faster is still doing the wrong thing.

Actively's founders looked at that and reached an uncomfortable conclusion. The earlier generation of AI SDRs, they argued, had mostly failed - not because automation was wrong, but because automation without judgment is just noise with a schedule. The hard part of selling was never sending the message. It was knowing which account deserved the message, when, and why.

Top human reps do this instinctively. They read the call transcript, notice the offhand comment about a budget cycle, cross-reference it against three other accounts, and decide where to spend their finite afternoon. That reasoning is rare, expensive, and impossible to clone by hiring. Actively's wager was that it could be modeled.

Two researchers, one verb

Mihir Garimella and Anshul Gupta met at Stanford, where both studied AI. Garimella's corner of the field was active learning - teaching models to ask for the most useful information rather than drowning in all of it. The company name is a tip of the hat to that idea, which is a rare case of a startup name that actually means something. Before Actively, Garimella's resume ran through Waymo and Facebook, and, earlier, a bird-inspired emergency-response drone he built as a teenager.

The bet they made in 2022 was contrarian for the moment: build custom reasoning models, not faster spam. Combine proprietary models with frontier reasoning from OpenAI and Anthropic. Point the whole thing at a company's own messy data - call transcripts, custom Salesforce fields, third-party signals - and have it think about who to sell to. It is the difference between a tool that acts and a tool that decides.

An agent for every account

The core unit is the Per-Account Agent: one AI assigned to one account, working in the background to research it, spot opportunities, and advance the next step without being asked. Completed work lands in an Agent Inbox, prioritized and ready for a person to approve. An Assistant nudges reps toward the best move. A tool named Watchtower - because of course it is - lets leaders watch the whole fleet at once. An API platform and an MCP server wire it into the data that already lives in the business.

What makes it unusual is the inversion. Most software waits for a human to open it. Actively's agents are supposed to have already done the work by the time the human shows up. The interface is less "here is a blank box, go" and more "here is what I noticed overnight, do you agree?"

Receipts, not adjectives

Skeptics are right to want numbers, and Actively's customers have offered a few. Samsara, the connected-operations giant, deployed agents across its entire 1,000-plus-person go-to-market team - account development, sales, RevOps and customer success - and reported roughly double the conversion on Actively-driven outreach, along with millions saved in compute and token costs. Ramp, the fintech, credits the platform with tens of millions in closed-won revenue and notably higher win rates. Ironclad ties tens of millions in qualified pipeline to Actively recommendations, and a sharply higher qualification rate, while running a leaner team.

The investor side reads the same data and likes it. TCV and First Harmonic co-led the Series B; Bain Capital Ventures, which led the Series A, called Actively the kind of team and traction that could "become the system of intelligence for GTM organizations everywhere." That is a large sentence. The customer logos are what make it more than a sentence.

Revenue at infinite scale, allegedly

Strip away the sci-fi vocabulary - Superintelligence, Watchtower, "Revenue at Infinite Scale" - and the mission is plainer than it sounds. Actively wants reasoning, not effort, to be the scarce resource a revenue team spends. The new Series B money goes toward more products for go-to-market teams, more hiring, more enterprise customers, and a San Francisco outpost to sit alongside the New York headquarters and its roughly 82 employees.

There is an honest tension here worth naming. A platform that promises to remove the ceiling on human-led revenue is also, quietly, a platform that does work humans used to do. Actively's answer is that the agents handle the part people hated anyway - the research, the watching, the never-ending homework - so the humans can do the part that needs a human. Whether buyers agree is the experiment now running at every account it touches.

Back to 2 a.m.

Return to that quiet office. The desks are empty. The pipeline is not. The renewal that would have slipped is flagged, the buying signal is logged, the morning's first move is already drafted and waiting for a human to say yes. Nobody worked late. The work simply did not stop.

That is the world Actively AI is selling, and the honest version of the story is that we do not yet know how far it scales. The numbers so far are customer-reported and promising. The vocabulary is louder than the evidence. But the underlying idea - that the bottleneck in revenue was never effort, it was judgment - is a sharp one, and Actively has convinced serious investors and serious customers to bet on it. The night shift, for now, is staffed by software. Morning will tell us how good it is.