A YC-backed startup wants to turn iMessage into a command center for AI automation. The pitch: teach it once, text it forever.
Most AI assistants require you to open something—a dedicated app, a browser window, maybe a sidebar that slides out from the edge of your screen. Tidy, a startup fresh out of Y Combinator's summer cohort, thinks that's asking too much. Why juggle another interface when you're already living inside your messages?
The company's wager is straightforward, if ambitious: that the best place for an AI agent isn't a standalone product but the messaging app already consuming hours of your day. Show Tidy how to complete a task on the web—booking a restaurant, checking transit times, ordering lunch—and it'll handle it from then on. Via text. No app-switching required.
It's an old dream with new packaging. And Tidy isn't alone in chasing it.
Learning by Watching
Tidy calls itself an AI assistant "that can do anything you can do on a website." The premise is demonstration-based automation: walk through a web task once—clicking buttons, filling forms, navigating pages—and Tidy captures the workflow. After that, you summon it with a text message.
The interface lives inside iMessage and Slack, not as a separate app you need to remember exists. Want to book a table at a restaurant through Resy? Text Tidy. Need to know if your Caltrain is delayed? Ask. The company's public gallery includes examples from users who've taught it to pull scientific news from Quanta Magazine, check Polymarket betting odds, scan Costco inventory, even monitor BayWheels bike availability.
Beyond custom automation, Tidy offers what it describes as a "filesystem" for notes and memory, plus reminders, scheduled tasks, and Google Calendar hooks. All of it accessible from the same chat thread where you're already coordinating dinner plans or complaining about meetings.
The product comes from Pond AI Inc., a San Francisco outfit run by Aagam Dalal and Brian Williams. They previously built Pond—marketed as "Superhuman for iMessage"—a desktop client designed to speed up message triage with AI-assisted drafting. That earlier project focused on improving the messaging experience itself. Tidy flips the premise: it treats messages as the interface for everything else.
According to the company's privacy policy, last updated in late October, Tidy routes requests through third-party LLM providers including OpenAI and Anthropic, with hosting on Fly.io. Persistent data gets encrypted at rest. Standard stuff for a venture-backed AI tool in 2025, though the policy offers no details on how demonstration recordings are stored or whether they're used to train models.
A Noisy Field
Tidy arrives into what's becoming a surprisingly crowded corner of the AI landscape. Lindy, another assistant targeting the iMessage crowd, promises to put "your entire work life, one text away" with enterprise-level privacy guarantees. Flux, which made a splash on Product Hunt, lets users "build your own agents in iMessage" and emphasizes conversational, proactive behavior—less chatbot, more helpful colleague who remembers context.
Then there's RPLY, a macOS app that scans unanswered iMessages and drafts AI replies—narrower in ambition than Tidy but proof that people are willing to let algorithms into their messages. For the privacy-conscious, OpenClaw and Moltbot offer self-hosted alternatives that work across iMessage, WhatsApp, Telegram, Signal, and Slack. They sacrifice convenience for control.
On the browser automation front, Tidy's "teach once" model echoes tools like Basepilot, which trains AI workers through demonstration for repetitive web tasks, and BrowsePilot, which uses vision-powered agents to manipulate websites through natural language. Axiom.ai offers no-code browser bots with a drag-and-drop builder. None require programming. All promise to automate the tedious stuff.
Tidy's angle—if it has one—is the fusion: demonstration-based learning delivered through a messaging interface. You're not managing a dashboard of bots or remembering to open yet another app. You're texting.
Whether that's genuinely differentiated or just good positioning remains to be seen.
What Remains Unclear
Tidy's website features a login button and a calendar link for booking meetings, but no public pricing. The terms of service mention payments without listing rates. It's not obvious whether you can sign up on your own or need to sit through a sales demo first—a telltale sign of a product still figuring out its go-to-market motion.
The claim that Tidy can automate "anything you can do on a website" invites skepticism. Web apps, sure—but what about sites that change their layouts regularly? Or multi-step workflows with edge cases? The available materials don't say. Neither do they address how the system handles CAPTCHAs, login walls, or the thousand small ways websites break automation.
There's no formal press release, no splashy Product Hunt launch, no Medium post from the founders explaining their vision. Just a live site, a YC directory listing, and a handful of examples. The product exists. It's taking users. But it's early, perhaps more early than the polished landing page suggests.
The Interface Question
What Tidy is testing—along with Lindy, Flux, and a scattering of similar ventures—is whether messaging apps will become the default interface for AI agents. Not a sidebar. Not a standalone app. The place where you already live.
The reasoning has appeal. People check their messages reflexively, hundreds of times a day. The conversational format feels natural. And if an AI can genuinely learn tasks by observation, maybe the real friction isn't building integrations—it's remembering which app to open. A text message is easier than that.
But ease cuts both ways. Will users trust an AI to book reservations or order food without verification? What happens when it misinterprets a request or a website updates its interface? And can a messaging paradigm handle genuinely complex workflows, or does it only work for quick lookups and simple tasks?
The bigger question is whether automation-by-demonstration is robust enough for daily use. It's one thing to show an AI how to check subway delays. It's another to trust it with multi-step purchasing or data entry, where a single mistake isn't just annoying—it costs time or money or both.
Tidy is live. It's booking meetings. The founders are showing off community examples. What they're really selling, though, isn't a product but a hypothesis about how people will interact with AI in the next few years. Whether they're right depends less on the technology and more on whether users actually want their assistant living in their messages.
That's a bet worth watching. But it's still just a bet.
