Six months ago, Andrew Lee and Jonny Dimond launched a platform for AI agents that never clock out. Now they've raised $20 million—and their eight-person startup is closing in on $5 million in annual recurring revenue as of mid-April.
The timing may be optimistic, or it may be prescient. Union Square Ventures and Lightspeed co-led the April round, with Y Combinator, Google's Jeff Dean, and Stripe's Patrick and John Collison joining. For founders who built Firebase—the Google-acquired developer infrastructure that now powers millions of applications—the bet is familiar: create the operating system, let others build on top.
Tasklet's pitch sidesteps the usual AI agent script. This isn't about chatbots that simulate helpfulness until you close the browser tab. The company's agents run continuously in the cloud, tethered to real tools and data sources, executing work whether anyone's watching or not. An agent monitoring pipeline health at 3 a.m. doesn't need you awake to notice something's wrong.
"Most AI agents are glorified chatbots," Lee said during the October launch. Tasklet, he argues, is different—though whether the market agrees remains to be tested at scale.
Plumbing Over Personality
Lee and Dimond aren't building another conversational interface. Tasklet operates as what they call a "cloud agent operating system," a persistent runtime environment where agents tap into more than 3,000 prebuilt integrations. Agents trigger off schedules, respond to incoming emails, write and execute code in sandboxed cloud environments, and—when APIs don't exist—control headless browsers to navigate websites the old-fashioned way.
The founders describe this as the "universal connections layer," which is startup speak for the unglamorous plumbing that actually makes software work. After Firebase, Dimond led technical development on Google Cloud Firestore. The pair understand infrastructure. They also understand that developers want programmatic control, not just business users clicking through forms.
Their "Instant Apps" feature, introduced in mid-April, lets agents generate live dashboards on the fly, complete with two-way data sync. An agent tracking sales pipeline can build an interface that updates automatically and accepts manual edits—changes that flow back to Salesforce or HubSpot without human routing. It's a bridge, perhaps, between the conversational future everyone's talking about and the traditional software people still rely on to get work done.
Whether that bridge leads somewhere meaningful is the question investors are now funding.
Not Your Father's RPA
Lee draws a sharp distinction between Tasklet and the robotic process automation tools that have disappointed enterprises for years. In a Hacker News comment posted during the October launch, he explained that traditional RPA relies on brittle flowcharts. Change a website's layout, and the workflow shatters.
Tasklet agents decide their steps at runtime using language models—primarily Anthropic's Claude—adapting to changes without manual reconfiguration. The platform is technically multi-model, but Claude handles the heavy orchestration. Tasklet also uses Claude's "computer use" capability, which lets agents control a cursor and keyboard inside a virtual machine, interacting with websites that never bothered building APIs.
According to an Anthropic customer story, Tasklet now executes 450,000 agent actions daily, with roughly 2,000 new agents created each day. The company claims 160% month-over-month revenue growth in early 2026, hitting $2.5 million ARR within five months of going public. By some internal measures, revenue has expanded over 1,200% since January, though such early-stage growth percentages often reflect small starting bases.
The numbers sound impressive. Then again, most startup metrics do when the denominator is still small.
Customers and Conflicts of Interest
Union Square Ventures didn't just lead the round—it rebuilt its own internal operating system on Tasklet's platform. That's either strong validation or a remarkable conflict of interest, depending on how you look at venture capital.
Tasklet also integrated deeply with Shortwave, the email client Lee co-founded before starting Tasklet. The integration, which went live in January, lets agents draft replies, manage tasks, and trigger workflows directly from an inbox. It's a natural pairing, though it also highlights how tightly networked the founding team's ecosystem has become.
Beyond Lee and Dimond, the roster includes engineers from Firebase, Google Cloud, and adjacent infrastructure projects. Company materials list the headcount as either eight or nine—a discrepancy that likely reflects rapid hiring in recent weeks rather than poor recordkeeping.
The Agent OS Land Grab
Tasklet arrives as everyone with a cloud platform and an LLM partnership scrambles to claim the "agent operating system" category. Anthropic launched Claude Cowork for desktop automation in January, with enterprise availability following in April. Microsoft has been experimenting with a Copilot Cowork variant. Perplexity introduced "Computer," a multi-agent coordinator, in February. Cloudflare and Warp have each shipped infrastructure aimed at autonomous agents running at scale.
The term "agent OS" has already lost whatever precision it once held. Some products focus on desktop workflows, others on enterprise governance, still others on developer tooling. Tasklet plants its flag in cloud-native execution: agents that live remotely, connect to your stack, and operate around the clock without local dependencies.
Whether the market actually needs a new operating system layer—or whether agents will simply nest inside the platforms everyone already uses—remains unclear. The history of enterprise software suggests that infrastructure bets either become invisible foundations or expensive distractions.
But the early traction matters. The investor lineup matters, too. Lee and Dimond built Firebase into a developer staple before selling to Google. They understand how to construct platforms that other engineers want to build on. If they've identified a genuine friction point—the gap between conversational AI and software that actually executes work—then Tasklet may represent more than another wrapper on someone else's language model.
Or it may be infrastructure in search of a problem, arriving just as the AI agent hype cycle begins its inevitable descent. We'll know more once the revenue base grows past the point where percentage gains sound more meaningful than they are.
For now, Tasklet's agents keep running. Whether anyone's watching or not.
