Image source: Tasklet official page. The interface shows an agent input surface connected to work systems.
Andrew Lee’s team spent years building Shortwave into an AI email client. By 2025, users kept asking for the same thing: could the AI keep working after they closed the inbox?
The request was concrete. When a sales email arrived, AI should update HubSpot, change the customer record in Notion, and notify a teammate in Slack. The person could go offline, but the workflow could not stop.
It would have been natural to keep adding this automation inside Shortwave. Instead, in June 2025 the team started again and built Tasklet as a separate product: a cloud agent that waits for email, calendar events, webhooks, or schedules, then executes work across systems.
In July, Tasklet got its first paid customers. In October, it launched publicly. Anthropic’s customer story says Tasklet reached $2.5 million ARR in five months, with monthly revenue growth of 160%. Those figures are company-disclosed and not independently audited.
The lesson is clean: when a feature starts working the night shift for customers, it may already be asking for its own budget.
Users closed the inbox, but the work kept going
Shortwave originally helped people process information faster while they were inside email. TechCrunch reported in 2023 that it had more than 20,000 active users and an AI assistant that could search email history, summarize conversations, draft replies, and schedule meetings. The assistant worked when the user opened the mailbox.
Autonomous background work changed the product boundary. An agent may need to receive a lead at midnight, update CRM, access documents, run a browser, and restart when another event occurs. The inbox is only the trigger. The value happens in the systems behind it.
This forced an architecture choice. Andrew Lee explained in Anthropic’s story that Tasklet does not require users to predefine every branch of a workflow. Users describe the desired result; the agent decides the next step, calls tools, and checks intermediate state. A single task can involve dozens or hundreds of tool calls.
That freedom creates a reliability problem. A weak email summary can be corrected by the user. A background agent that has been running for two hours can waste dozens of steps if it goes wrong early. Tasklet needed its own triggers, permissions, execution environment, and reliability system. It could no longer pretend to be just an inbox sidebar.
The team therefore gave up the existing product shell. Tasklet received its own interface, engineering priority, and pricing model. It could also enter companies that did not care about Shortwave.
Once separated, revenue steepened
Five months later, Tasklet customers were creating agents that performed 450,000 actions per day, and users were creating about 2,000 agents daily. Work expanded from email classification into revenue reconciliation, deal desk, competitive monitoring, and research.
In April 2026, Tasklet announced $20 million in funding and said annualized revenue had passed $5 million with 1,200% growth for the year. A month later, Sacra estimated ARR at about $10 million. Company figures and analyst estimates are not audit results, and the revenue numbers use different methods. Together, they point to the same pattern: after the split, Tasklet found a revenue curve Shortwave had not unlocked.
Tasklet’s pricing boundary is wider than email. Each additional system, workflow, model call, browser action, or code execution consumes credits.
Its pricing starts with a free plan, then Starter at $25 per month, Pro at $100 per month, and Custom at $250 per month. Model inference, tool calls, browser use, and code execution all consume credits. That unit matters. Customers are not only paying for a seat. They are paying for completed work.
An old factory pushed the product beyond email
Tasklet’s site includes a striking deployment story. Piping Technology & Products is a 51-year-old U.S. manufacturer with more than 700 employees. CEO Rakesh Agrawal connected Tasklet to his email and calendar over a weekend. Forty-eight hours later, he gathered nine employees from different roles and each built a first agent in half an hour.
Forty days later, more than 60 people were using it. Contract managers used agents to prepare early-morning reports. Operations teams used it to track backlogs. Engineers even built reviewable structural calculation tools. Tasklet says one quoting workflow that previously needed three people and up to four hours fell to one person in 15 minutes.
These results come from Tasklet’s official customer story and are not independently audited. But the deployment illustrates the shift. No one at the factory cared that Tasklet came from an email product. They cared whether the morning report was ready and whether a quote could be sent before the customer moved on.
The rollout also avoided the usual enterprise automation trap. Piping did not begin with a full company-wide process map and a long IT implementation. Each role started with work it understood and could verify. Value appeared in a 15-minute quote and a daily report, then new agents grew outward.
A night-shift feature can become a business
Not every popular feature deserves a separate company. If users merely click one extra button, that usually supports a product update, not a new budget. Tasklet showed heavier signals: independent triggers, continued operation outside the original interface, connections across systems, and willingness to pay as work volume increased.
The risks are also real. General agent markets are crowded with model companies, automation platforms, and startups. The more permissions an agent receives, the higher the cost of mistakes. Tasklet’s usage and revenue figures still come from the company, partners, or estimates; retention and enterprise reliability need time.
But Shortwave did not wait for perfect certainty. The team noticed that customers wanted work to continue after email closed, and they separated the product.
The reported $2.5 million ARR in five months is the market’s first payment for that judgment.
For builders, Tasklet’s lesson is not simply “spin out features.” It is to watch for features that have crossed a boundary. If the feature has its own trigger, runs without the user present, connects multiple systems, creates measurable work volume, and exposes new pricing units, it may no longer be a feature. It may be a business wearing the old product’s clothes.