This AI agent lives in your texts no app, no install, no setup
The AI agent race has a new front-runner in the accessibility war. Poke a 10-person startup out of Palo Alto wants to be the AI agent for the billion people who will never touch a terminal. You just text it.
Most conversations about AI agents in 2026 involve a lot of setup friction API keys, installs, permission grants, and the ever-present risk of something going sideways on your machine. That’s fine for developers. For everyone else, it’s a dealbreaker. Poke, built by The Interaction Company of California, is betting there’s a massive market in removing all of that friction entirely. Their solution? Skip the app. Just text the agent.
As TechCrunch first reported, Poke launched publicly in March 2026 and is accessible over iMessage, SMS, Telegram, and in select markets WhatsApp. No download, no account setup beyond entering your phone number, no configuration. You just start texting and the agent starts doing things on your behalf.
What Poke Actually Does and Who It’s For
Think of Poke less like a chatbot and more like a proactive personal assistant that lives inside your messaging app. You can ask it to alert you when a specific email lands in your inbox, remind you if rain is in the forecast, update you on sports scores, track your fitness goals, control your smart home, and even automate tasks you define yourself in plain English then share those automations with friends.
That last part is important. Poke calls its pre-built automations “recipes,” and they work with apps and services people already use daily: Gmail, Google Calendar, Outlook, Notion, Linear, Granola for scheduling and productivity; Strava, Withings, Oura, and Fitbit for health and fitness; Philips Hue and Sonos for smart home. Developers can also wire Poke into their workflow through integrations with PostHog, Webflow, Supabase, Vercel, Devin, Sentry, GitHub, and Cursor Cloud Agents.
The recipes marketplace is also becoming a creator economy play. Poke pays between 10 cents and a dollar for every person who signs up via someone else’s shared recipe a smart early move to drive organic distribution through social channels without traditional ad spend.
The Funding Round and Who Believes in It
Poke has just added another $10 million to its balance sheet, on top of a $15 million seed round closed earlier. That brings the total to $25 million raised and pushes the post-money valuation to $300 million for a 10-person team. That’s a high number, but the investor list helps explain the confidence.
Spark Capital and General Catalyst are the lead institutional investors. But the angel list reads like a who’s-who of people who’ve built category-defining consumer internet products: John and Patrick Collison (Stripe), Guillermo Rauch (Vercel co-founder), Ken Howery (PayPal co-founder), Arash Ferdowsi (Dropbox co-founder), and Thomas Wolf (Hugging Face co-founder). Notably, Joanne Jang of OpenAI and Logan Kirkpatrick from DeepMind also invested which signals that even people inside major AI labs see Poke as something different, not directly competitive with what they’re building.
The Logan and Jake Paul involvement is a tell about the distribution strategy. Poke’s near-term growth plan centers on influencers and creators showcasing real use cases a consumer-first playbook that looks a lot more like Snapchat’s early days than the typical enterprise SaaS rollout.
Provider Agnosticism Is Poke’s Real Moat
Here’s something the funding headlines will bury: Poke doesn’t use a single AI model. Under the hood, the platform routes each task to whatever model is best suited for it whether that’s a model from one of the major AI labs or an open-source alternative. Co-founder Marvin von Hagen made this point explicitly when speaking to TechCrunch:
This is a bigger structural advantage than it might seem at first. Every major AI platform right now is also a distribution channel for their parent company’s models. OpenAI’s agents push GPT. Google’s agents push Gemini. OpenAI is actively consolidating everything into one superapp which is great for OpenAI, but it creates a lock-in layer that Poke is explicitly designed to avoid. If a better open-source model emerges next month, Poke can route to it. Its competitors can’t.
The OpenClaw Comparison and Why Poke Is Something Different
TechCrunch framed Poke as potentially “an OpenClaw for the rest of us,” and it’s a fair starting point but the analogy only goes so far. OpenClaw, which grabbed headlines after its creator joined OpenAI and then became a flashpoint in the Anthropic subscription policy debate, is fundamentally a developer tool. It requires deep system access, command-line comfort, and a tolerance for setup friction that most people simply don’t have.
Poke’s value proposition starts at the exact opposite end. The friction is zero. You go to Poke.com, enter a phone number, and you’re live. There’s no app to update, no permission screen to navigate, no mental model to learn. The interface is text messaging which about 5 billion people already use daily.
What’s notable here is that the product’s simplicity isn’t just a marketing angle it emerged from real user behavior. Von Hagen described watching beta users of an earlier email-focused product start asking Poke for medication reminders, sports scores, and jacket recommendations. The team didn’t design those use cases. Users discovered them. That kind of organic expansion is usually what separates products with genuine consumer pull from those being pushed by VC narratives.
For those interested in how AI automation is reshaping everyday workflows, our coverage on Gemini’s task automation features shows how the big labs are approaching this same space with a very different philosophy.
The WhatsApp Situation A Real Constraint Worth Watching
Poke works on iMessage, SMS, and Telegram. WhatsApp is trickier. Meta moved last fall to bar general-purpose chatbots from WhatsApp which is a significant problem for any AI company targeting global markets, given WhatsApp’s dominance in Latin America, South Asia, and Europe.
The good news for Poke is that regulators are pushing back. The EU, Italy, and Brazil have all opened antitrust probes, and WhatsApp has already been compelled to allow rival AI agents in those markets though Meta is charging fees for access that von Hagen believes will eventually be challenged further. This regulatory momentum is significant: if the EU decision holds, it sets a precedent that could eventually force Meta to open the door globally.
For now, Poke’s messaging story in the US runs cleanly on iMessage and SMS which covers most of the American market anyway. Internationally, this is a watch item. To enable its iMessage integration, Poke leverages Linq, a startup that raised $20M to embed AI assistants inside messaging platforms a piece of infrastructure that makes the entire product possible at scale.
Pricing: Surprisingly Thoughtful for a Venture-Backed Startup
Most early-stage AI startups either charge too much, give everything away and run out of money, or build in so many tiers that users spend more time navigating pricing than using the product. Poke’s approach is genuinely different and a little unusual.
The service starts free. If you’re primarily asking questions that don’t require live data lookups or real-time processing, you can likely stay on the free tier indefinitely. What costs Poke real infrastructure money and therefore what it charges for is real-time inference: automations that trigger on every incoming email, live flight check-ins, that sort of thing. During beta, users reportedly negotiated pricing directly with the AI agent itself, landing somewhere between $10 and $30 per month. The personalized pricing model is now set by the product based on usage patterns, not a fixed monthly tier.
Von Hagen was direct about near-term goals: “We really don’t want to make money, but we really want to grow. We want to build a product for a billion people and monetization is really secondary.” That’s a statement you’d expect from anyone with $25 million in the bank and a $300 million valuation to protect but the usage-based, low-friction pricing structure at least suggests the team has thought carefully about removing barriers to adoption rather than maximizing ARPU in year one.
Security: The Question Any Thoughtful User Should Ask
Giving any third-party app access to your calendar, inbox, health data, and smart home devices is a significant trust decision. Poke says it uses a multi-layered security model that includes regular penetration testing, security checks, and strict permission scoping for both the AI agents and human employees. By default, the team reportedly cannot view anything inside user tokens unless a user explicitly opts in by enabling access via a settings toggle.
It’s worth noting that TechCrunch has not independently verified these claims through a security audit and neither have we. For most casual use cases, the risk profile is probably similar to connecting any third-party app to Google Calendar or Gmail. But anyone thinking about wiring Poke into sensitive workflows should treat this as an area still worth scrutinizing as the product matures.
What This Means for the AI Agent Market in 2026
The AI agent space is fragmenting fast, and Poke is a clear signal of where the consumer end is heading. OpenAI, Google, and Anthropic are all fighting for the power-user layer the people who want deep integrations, long context, and system-level access. But there’s a different category entirely forming underneath: AI agents for people who just want something done, with no learning curve.
The fact that Poke’s user base has grown tenfold in two months without traditional advertising, purely through creator distribution and word of mouth suggests there’s real demand here that existing AI products aren’t fully serving. ChatGPT and Claude are excellent for research and reasoning tasks. Poke is positioning itself as the thing you turn to when you want something to actually happen: a reminder sent, a flight checked, an automation triggered. The surface area of that use case is enormous.
The broader trend to watch is what happens when agentic AI meets the distribution scale of SMS. WhatsApp alone has over 2 billion users. iMessage is standard on every iPhone in the US. If Poke can win on simplicity and reliability, the text message may become the most consequential AI interface of the next few years not because it’s the most powerful, but because it’s the one everyone already knows how to use.
The companies that figured out how to use AI before competitors have already started to pull away see how businesses are actually deploying AI right now. Poke’s accessibility-first approach may be how millions of non-technical users finally join that wave.
An AI researcher who spends time testing new tools, models, and emerging trends to see what actually works.