Firebase co-founder Andrew Lee returns to YC with AI agent platform Tasklet, now at $7M ARR

Andrew Lee built Firebase and sold it to Google in 2014. Now he's back at Y Combinator with Tasklet, an AI agent platform that connects across work tools, runs 24/7, and takes ownership of recurring workflows including email, search, and cross-tool integrations.

The growth numbers are hard to ignore. Tasklet launched in October 2025, entered 2026 at roughly $333K ARR, and is now at $7M ARR — a roughly 20x ramp in under six months. Lee says Tasklet has the highest revenue of the current YC batch. The company closed a $20M round in April with USB Ventures and Lightspeed participating, and is actively raising more, citing direct competition with Anthropic and OpenAI.

“At the beginning of the year, we were at about a third of a million of run rate. We're at a $7,000,000 run rate now... We actually already closed a $20,000,000 round in April with USB and Lightspeed and some other folks. But we're competing directly with Anthropic and OpenAI.”

From email client to agent platform

The path to Tasklet was indirect. Lee started the company in 2020 as a better Gmail replacement, raised money, built a team, and was preparing to shut it down when LLMs became capable enough to change the calculus. The pivot into AI-powered email search and drafting gave birth to a capable agent inside the email client — but users kept asking Lee to strip out the inbox UI and just give them the agent. Tasklet is that product, built on an entirely new codebase.

Dynamic integrations

The core technical differentiation is how Tasklet handles integrations. Rather than hand-coding connectors the way Zapier or legacy automation tools do, Tasklet uses AI to generate integrations dynamically at the moment they're needed. It ships some pre-built connectors, but if a user requests something outside that set — including internal, bespoke company APIs — the model generates the integration on the fly. Lee argues this is the capability that lets Tasklet reach past walled gardens and connect to tools that no off-the-shelf platform supports.

Search without indexing

On search, Lee makes a counterintuitive point. Tasklet doesn't pre-ingest and index your email into a vector database. Instead, it runs large numbers of parallel queries against live APIs, looks at results, and iterates. The output is surprisingly accurate without the infrastructure overhead — though it's slow and can be expensive. Lee's own test run ahead of Demo Day, using Claude with the same approach, spent about $60 over 30 minutes to scan ten months of cold inbound email and produce a ranked spreadsheet of 100-plus contacts. Useful, but not yet a casual consumer feature.

The cost and latency constraints are real limits for now. Scaling that approach to Google's billion-user base, for free, would be prohibitively expensive — so the window for Tasklet and similar tools to build a position likely stays open for at least a few more years.