Big launch today: Shortwave now integrates with @taskletai, our AI automation engine. Your inbox now works for you 24x7 — automatically drafting replies, managing todos, and more. Try it: ask "Connect to my email via Shortwave" at tasklet.ai

This non-technical role is our most urgent need right now -- and no experience required. Are you (1) very smart (2) very hard working (3) great with people (4) ready to drop everything and move to San Francisco? This is a life-changing opportunity for the right person.

We just raised $20M to build the cloud agent operating system for work. The investment is led by @usv & @lightspeedvp and includes @ycombinator, @JeffDean, @patrickc & others. It values the company at $175M. 🧵

Tasklet (@taskletai) is the cloud agent OS for knowledge work. It connects to all your tools, uses computers in the cloud, and runs 24/7 to get real work done. Started by Firebase founder @startupandrew and @jonnydimond, it’s grown >1,200% this year to $5M ARR and just raised $20M. Congrats to the team! ycombinator.com/launches/PsX…

We partnered with @AnthropicAI to build Tasklet — and today they're sharing our story. Tasklet is powering 450K agent actions/day across thousands of businesses, and growing 160% MoM. Every LLM call is Claude. claude.com/customers/tasklet

Sonnet 4.6 & Opus 4.6 are now available in Shortwave. And we've now made Standard intelligence use Sonnet -- a big upgrade for our Pro & Business tier users.

Fundamentally, we think traditional application software is going away. Most of the SaaS you use today will be accessed through AI agents via API rather than directly. Any UI you need will get generated on the fly, purpose-built for the task at hand. You won't adapt to someone else's interface — you'll describe what you need and get it instantly, connected to your real data. 2025 was the year of coding agents. 2026 is the year this extends to all knowledge work. Instant Apps is our first big step toward that future.

Introducing Instant Apps — your Tasklet agent can now build live, interactive apps for you on-demand. Describe what you need & get the perfect custom UI in seconds — connected to your real, live data. 🧵

🚀 Tasklet can now send images and files directly to your iMessage and email, so you can review wherever you are. Need changes? Just reply! Try it out at Tasklet.ai!

You don't need @openclaw to use @moltbook. You can do it with just a prompt in 30 seconds using @TaskletAI (and it works on the free plan)

Just a reminder that we're hiring one amazing SWE in SF. Ideal candidate sits at the intersection of systems engineering & AI. Up to $330k salary lots of equity. $20k bonus if you refer someone to us we hire. Details: shortwave.notion.site/Staff-…

Most of the people talking about clawdbot really want tasklet.

I use @TaskletAI every day to automate tedious tasks. Here I ditch Facebook’s terrible app with a shopping agent to find a laptop for me. This is just shopping, but Tasklet can connect to basically anything and automate ALL KINDS of tasks. It's super fun. Watch it work. 👇

Tasklet agents can now directly email & iMessage anyone. The best part? Communication is two-way—reply to your agent right from your inbox or phone.

At @TaskletAI we’re building agents that do real work - not just chat. A big challenge: how do you architect a long-running agent that doesn’t forget and can reason over massive amounts of data? Here are a few tricks we use 👇 1/ Dynamic, infinite context We want users to continuously tweak an agent in a single “infinite” conversation. But not all context is equal: User intent > tool output. Patterns > individual actions. So we built a context manager that compresses history dynamically, preserving what matters (e.g. “this happened 100 times”) while keeping context cheap and manageable. 2/ JSON-only tool results Every tool returns structured JSON. Yes, it costs more tokens - but we get 3 big wins: • Safer by default (JSON resists prompt injection surprisingly well) • Can be be processed with code - not just vibes • Aggressive truncation, with on-demand field lookup when needed The agent sees structure, not blobs of text. 3/ Everything is a file This is the glue that makes the 2 other tricks work. The agent has a virtual filesystem. Every tool result - even binary data like images - becomes a file the agent can read with code. That means the agent can read, transform, query, and compute over real data - without using many tokens or hallucinating. Example: the agent can average by actually iterating over result JSON and doing math — not guessing. We’ve learned a lot building this — I’ll share more as we go.

An AI agent orders lunch for our team every day. We set it up ONCE, and it runs by itself. We've been doing this for a couple months now, and even discovered amazing restaurants we never would've tried on our own! Prompt to get started here! 👇 tasklet.ai/?prompt=Order l…

Did I mention we're hiring Software Engineers at @TaskletAI?

I built an agent that completely controls its own website. youarethepm.com The agent is taking feature requests on HN and building them live! news.ycombinator.com/item?id…