Poke, a 10-person startup based in Palo Alto, has raised $25 million in total funding to build an AI agent that users control entirely through text messages. The company launched publicly in March and is now valued at $300 million post-money, according to TechCrunch.
The product works through iMessage, SMS, and Telegram. Users text Poke plain-language instructions (“alert me to emails from my boss,” “remind me to take my medication every morning,” “tell me if I need an umbrella tomorrow”) and the agent handles the automation. There is no app to install. Users visit Poke.com, enter a phone number, and start texting.
How It Works
Poke routes each request to whichever AI model best fits the task, pulling from major providers and open-source models. Co-founder Marvin von Hagen told TechCrunch that model flexibility is a long-term advantage: “Almost all of our competitors are just big tech and labs that are bound to a specific provider. Like Meta AI will only ever be able to use Meta models, and ChatGPT will only ever be able to use OpenAI models.”
Users can write custom automations in plain text and share them with friends through what Poke calls “Recipes.” The product handles daily planning, calendar management, health and fitness tracking, smart home control, photo editing, and sports scores.
Poke uses Linq, a messaging infrastructure layer, to operate inside iMessage. WhatsApp support is limited after Meta barred general-purpose chatbots from the platform last fall.
Funding and Backing
Spark Capital and General Catalyst led the funding. The latest round added $10 million on top of a $15 million seed from last year. At a $300 million valuation for a 10-person team with a text-message product, Poke’s backers are betting that the next wave of agent users will come through the simplest possible interface rather than the most configurable one.
The launch comes as TechCrunch notes demand for agentic AI is “spiking,” citing OpenAI’s acquisition of OpenClaw’s creator and Nvidia CEO Jensen Huang’s statement that every company needs an OpenClaw strategy. Poke is positioning at the opposite end of the complexity spectrum: if OpenClaw requires terminal access and dependency management, Poke requires a phone number.
The Interface Bet
Von Hagen said Poke pivoted from an email-only AI assistant after watching beta testers try to use it for everything. “People started asking Poke to remind them to take their medication. They asked Poke about sports results,” he told TechCrunch. “We noticed how we needed to become general-purpose much more quickly, because people just like the personality and the humanness of it so much.”
The question Poke’s model raises for the agent ecosystem is whether the next 100 million agent users will look like today’s OpenClaw power users, configuring YAML files and managing skills, or like Poke’s target user, who sends a text and expects something to happen.