Rocker CEO Alex Vogenthaler was changing his snow tires when the startup idea struck.
Pacing the shop, he caught up by phone with Peter Terrill, a former Stripe colleague who had left the fintech company a few years ago to explore possibilities in AI. Agents for finance? Healthcare?
“F***it,” Terrill told him. As head of product for Stripe’s Connect marketplace and enterprise products, Vogenthaler oversaw 2,000 people and billions in revenue. Terrill, his former engineering leader for Connect, reframed the chat there: “What’s the agent you would want for you?”
“At the end of the conversation, it was just so clear,” Vogenthaler says now. “We’re building this.”
Now, a year and a half later, the duo are sharing that work publicly for the first time. They’re announcing Rocker, a new startup they call “the agent for the busiest people in the world at work.”
An app that connects to your Google Docs, Slack, email and other work tools, Rocker tracks and identifies a user’s highest priority tasks and objectives, then works to expedite them: catching critical decisions or messages that are slipping through the cracks, removing busywork, and reorienting a knowledge worker’s day around their top priorities.
Rocker’s founders know what you’re thinking: Oh God, not another AI assistant.
This one is different, they promise, for a few key reasons. It’s focused entirely on work, and especially the work that happens inside large organizations, not so much small, fellow startup teams. It’s architected in a way that can collect and understand more context than other tools. It’s not a tool for engineers or tinkerers who want to spin up lots of agents of their own. But it can keep track of all the other tools you use, as one more fully-fledged agent supervising up top.
“It’s your Rocker, singular, not your Rockers, plural,” Terrill says. Think of Rocker as less a tool you have to optimize, he argues, and more as an experience like downloading Spotify from the App Store, then streaming a song. “This is a plug-and-play product.”
Rocker launches today with $12.85 million in funding. Greycroft led the round, with participation from Anthropic, Menlo Ventures’ Anthropic-aligned Anthology Fund, Matchstick, and Stripe itself, as well as personal checks from leaders at Stripe, Vercel and Meta.
The startup is opening up its waitlist today, prioritizing larger teams; Rocker is already in discussions and test use with large organizations, including public companies, adds Vogenthaler, declining to identify them by name.
It’s obviously a crowded field of AI assistants, agents and work tools. At Upstarts, we’ve written about a bunch, like work tool Town, which recently confirmed a $108 million funding round, and how people are starting to spend money with personal agents like Instinct and Muse. Two weeks ago, we covered the launch of Scuba, a tool from similarly well-pedigreed founders offering a “context engine” for agents and agile teams to ship products. (Rocker’s Vogenthaler is an investor in Scuba, and says they’re chasing different goals.)
Rocker’s founders offer a couple of explanations for why you should pay attention to their entry into the field. From a product standpoint, they argue that they’ve architected Rocker to have more knowledge, and simply work better, than other work agent tools in the market. And they believe that their approach has the best chance yet to crack the world’s largest corporations, where execs have largely sat out the Silicon Valley agentic free-for-all.
“It’s really easy to make an agent that goes through the motions. It’s really hard to make a good one,” Vogenthaler says. “We think the category to be the Muse or Instinct for work is wide freaking open.”
A MESSAGE FROM OUR SPONSOR
Ship pricing like a product
Different product bets have different business goals. You’re either chasing market adoption, revenue growth, or profits. You need to test fitting pricing models for each of those goals. Fast. Chargebee can help — Price for usage credits, outcomes, complex contracts, or any other modern commercial construct.
More AI assistance
Early Rocker users get the product, its founders say, when they have their first “Oh, shit” moment. That could be a job candidate who never received their offer, or a customer agreement never signed, or a salesperson posting in Slack that they’re about to send the wrong materials to an important prospect.
For executives using Rocker, such moments happen fast enough that it’s become an internal meme: Rocker put the phrase on company T-shirts. But Vogenthaler says the problem is one that’s afflicted business leaders for decades: the question of “What the hell is happening in my company?”
In more than a decade at Google, Vogenthaler’s responsibilities included overseeing Google Drive and much of its Workspace products. At Stripe, he eventually managed products including Connect, Payments, and its consumer wallet, Link, with his teams accounting for more than 90% of the company’s revenue, he says. The mental load, and context, required to oversee all that was taxing. At one point, Vogenthaler counted 250 distinct projects that he was responsible for inside Stripe, he adds.
Terrill, meanwhile, had gone through Y Combinator as co-founder of a startup eventually acquired by Stripe, Payable, then scaled out its Connect engineering team from less than 10 to approaching 100. After leaving at the end of 2021, before the launch of ChatGPT, he spent three years consulting and working on products around large language models, and experimenting with his own bespoke setups and harnesses.
To create a high-context AI assistant that didn’t suck, they decided to rethink how Rocker would learn a business. Instead of starting with a company’s core knowledge and connecting out to its individual employees, Rocker builds a context graph for each individual user, defining their relevant people, projects, topics, and decisions by logic, and then combining those algorithms with math that measures semantic connections between them. Each of those individual graphs is then interconnected within a company’s own cloud environment, meaning that Rocker isn’t accessing sensitive enterprise data, and individual-level data isn’t leaking across a wider company.
What Rocker looks like in practice: after a user downloads the app to a local environment, Rocker starts to learn their work habits within 30 seconds, ready to flag immediate concerns. For many users, those come via a ping at the end of the workday, telling a user the most important things they still need to execute before logging off and having a restful night’s sleep. (Why not start with a daily digest of tasks, like many other tools? Those feel more like a vitamin than a painkiller, Rocker’s founders argue, suggesting ways you could approach your day with less urgency.)
For businesses that use AI tools like ChatGPT or Claude, Rocker can make them more valuable without creating a new vector for risk through data permissions, Vogenthaler and Terrill say; Rocker can also execute tasks on a user’s behalf with a company’s own internal AI agents, like with Stripe’s company-wide tool, called Kai.
For example, Rocker might flag that a user didn’t complete a deep research report within Kai prior to a meeting, and instruct it to do so; for more rank-and-file users, it could automate status reports needed by bosses, reducing hours of work to a minute of approvals.
At Irrational Labs, a behavioral economics-based product design firm that’s been working with Rocker, co-founder and CEO Kristen Berman says she had one of the “Oh, shit” moments when a salesperson was going to send the wrong document to a client. With 20 projects to manage at any given time, she’d missed the Slack message, but Rocker caught it.
The app catches something else she’s missing seemingly every other day, she says. “That makes me have more internal reflection than I care to admit,” Berman quips. “The context of what it knows about me just somehow nails it. I don’t know how they do it, but it gets my work world.”
Another partner working with Rocker who received early access (it must be noted that, obviously, both of these customers have closer ties to the startup than a neutral party) is Lauren Hawker, a senior project manager at MultiAdaptor, a London-based digital branding agency. Hawker checks Rocker several times a day, including just after lunch, to refocus her day, and triage at day’s end.
Where Rocker has proven most valuable to her over the past two months using it, however, is as a mental strain reliever. Internally, Hawker is known as a knowledge keeper of company lore, best practices and quirks, she says, meaning people come to her continually for wider context. With some of that offloaded to Rocker, she can spend more time thinking strategically and creatively, she adds.
“It’s been a real shift in what my own role could be, what I pride myself on, and what I do with my job,” she says.
A crowded fight
At Upstarts, I work with a cracked team of specialists part-time, but I am the only full-time employee. So I’m not an ideal target customer for Rocker, which is designed to appeal to big company executives, and to get their approval for widescale use. It’s not a one-for-one swap for the conversations I have with Instinct and Muse, the automated workflows I’ve set up through Town, or the times I ask ChatGPT or Claude for spellcheck assistance, or podcast research prep. (My favorite prompt: What are the 5 questions a guest has already answered to death, that I shouldn’t ask?)
Vogenthaler admits that he was concerned that the biggest AI labs might move more in this direction. He’s not now, he says.
“When we started working on this a year and a half ago, I was just worried. Are the models going to be able to pick the needle out of the haystack?” he tells Upstarts. “And yeah, they’re able to pick up the whole haystack, and then they drop the whole haystack on you.”
And Rocker’s founders are betting against the long-term viability of agent-builder tools to become popular at scale, outside of a technical or tech-adjacent startup set. While tech-savvy practitioners will continue to build their own agents and harnesses, such tools are too complicated, and too annoying to maintain, to truly take off, they argue.
“I think agent builders are an evolutionary dead end,” Vogenthaler says. “Magic doesn’t happen. You can imagine it being good, but it’s not actually good.”
At Greycroft, lead investor Ian Sigalow believes that Rocker’s setup will allow for more effective ‘multiplayer’ use of such agents across a company, versus single-player, self-obsessed experiences with personal agents.
Still, he concedes that he expects as many as 10 to 20 startups to receive significant venture funding to go after agentic enterprise use cases.
“This fight right now is basically for the future of the operating system,” he says. “I log into my workspace every day, and I get a blank canvas with a bunch of apps on it. It’s just not going to look like that.”
The end-of-day ping is the most revealing product choice in this piece, and it is also the one I would interrogate hardest. Rocker is pitched as a painkiller rather than a vitamin: instead of a morning digest, it tells a busy executive what is still slipping at the end of the day, from the unsigned agreement to the offer that never went out. That is genuinely useful for catching dropped balls, but a system that excels at surfacing what is slipping will systematically privilege the urgent and the already-committed over work that has no thread, no deadline, and no one chasing it yet.
For someone overseeing 250 projects, the scarce resource is not really another reminder to finish today's list. It is the judgment to decide what should be dropped, delayed, or never started. So the test I would apply to a singular supervisor agent is not only whether it catches the right near-misses, but whether it ever tells a leader that a priority is wrong. Does Rocker push back when the context graph is faithfully executing a bad ranking, or does it mainly make a questionable priority list run more efficiently? A supervisor that only expedites, and never challenges, could quietly turn urgency into strategy.