Headline & intro
AI agents were supposed to arrive as complex developer tools and OS-level copilots. Instead, one of the most interesting experiments now lives in the most boring interface of all: SMS. Poke, a text‑based AI agent that just raised another $10 million, is betting that the future of automation looks less like a dashboard and more like messaging a reliable friend.
In this piece, we’ll look at what Poke actually does, why investors are valuing a 10‑person startup at $300 million, how it fits into the broader agentic AI race, and what its messaging‑first strategy means for Big Tech, regulation and everyday users — particularly in Europe.
The news in brief
According to reporting by TechCrunch, Palo Alto–based startup The Interaction Company has publicly launched Poke, a consumer‑oriented AI agent that runs entirely over messaging apps such as iMessage, SMS and Telegram, with limited availability on WhatsApp in some markets.
Users text with Poke to offload practical tasks: calendar management, email triage, health and fitness tracking, smart home control, reminders, news briefings and more. Instead of coding workflows, people describe automations in natural language or install pre‑built “recipes” that integrate with services like Gmail, Google Calendar, Notion, Strava, Oura, Philips Hue and developer tools including GitHub and Vercel.
Poke dynamically routes requests to whichever foundation model best suits the job, mixing big‑tech and open‑source models. The company has raised a total of $25 million ($15 million seed plus a recent $10 million extension) from Spark Capital, General Catalyst and a long list of prominent angels, and now carries a reported post‑money valuation of around $300 million. Pricing is flexible, with basic, non–real‑time usage often free and more intensive automations charged according to resource use.
Why this matters
Poke is important less as a product and more as a thesis: that the winning AI “agent” may not be the one with the flashiest interface, but the one that quietly embeds itself into habits people already have — checking their messages.
Three things stand out.
First, Poke attacks friction, not just features. Installing OpenClaw‑style systems or enterprise copilots typically means new apps, permissions and mental models. Poke piggybacks on messaging, which everyone from teenagers to grandparents already understands. That makes it a far more realistic candidate to reach the non‑technical majority.
Second, it treats models as interchangeable parts. While OpenAI, Meta and others want you locked into a single model family, Poke is effectively saying: we’ll orchestrate whatever works best. If that stance holds, the real moat shifts from model performance — which is converging — to data access, UX and trust.
Third, the “recipes” ecosystem is potentially disruptive. We’ve seen automation platforms before (IFTTT, Zapier, Make), but they still demanded some technical thinking. Letting users describe automations in plain language, run across email, health, finance and home, then share and even monetize them, pushes AI agents much closer to an app platform in their own right.
The winners in this scenario are consumers who gain low‑friction automation, and smaller AI model providers who can be plugged in behind the scenes. The losers, at least in theory, are incumbents whose assistants (Siri, Alexa, Google Assistant, even ChatGPT in its app form) risk being reduced to just another channel instead of the orchestration layer.
The bigger picture
Poke slots into a wider shift from “chatbots that answer” to “agents that act.” OpenAI’s acquisition of OpenClaw’s creator, Nvidia’s push for agent platforms, and Anthropic’s more action‑oriented Claude offerings all signal the same thing: static Q&A is table stakes; the real value is in reliably doing things on your behalf.
Historically, we’ve been here before. In the 2010s, personal assistants like Google Now, Cortana and early Siri tried to automate context‑aware tasks — but they were tightly bounded by the phone OS and a shallow set of integrations. Then the pendulum swung to no‑code automation tools and SaaS‑specific bots.
The difference now is threefold: foundation models can understand messy human instructions; APIs are more ubiquitous; and messaging has become the de facto control surface of daily life. That makes something like Poke — “IFTTT inside your chat app, backed by GPT‑class models” — possible in a way it wasn’t a decade ago.
Competitively, Poke is a mosquito buzzing around giants. Meta is pushing its own Meta AI in WhatsApp and Instagram. Apple is expected to bake much deeper on‑device AI into iOS. OpenAI is moving ChatGPT toward a more agent‑like posture with tool calling and automations. All of them have distribution advantages Poke can only dream of.
But those same giants are constrained: they are locked into their own models, their own app stores, and increasingly, their own regulatory obligations. A nimble, messaging‑native player that can move fast across platforms, swap models, and iterate on UX has room to explore weird, user‑driven behaviors the big platforms won’t prioritize.
The European / regional angle
For Europe, Poke sits right at the intersection of usability and regulation.
On one hand, the value proposition maps neatly onto European digital habits. WhatsApp and Telegram are core infrastructure for social and business communication from Lisbon to Ljubljana. A text‑based agent that manages calendars, travel, invoices or health reminders inside those channels could easily become a shadow productivity OS for freelancers, small businesses and families.
On the other hand, the regulatory minefield is real. GDPR makes Poke’s access to email, health metrics (Strava, Oura, Fitbit, Withings) and smart‑home data extremely sensitive. The AI Act’s rules on general‑purpose AI and high‑risk uses will force the company to be far more transparent about model sourcing, data flows and human oversight than a typical Silicon Valley startup.
There is also the WhatsApp question. TechCrunch notes that Meta has effectively blocked general‑purpose chatbots on the platform and is now facing antitrust scrutiny from the EU, Italy and Brazil. If regulators force interoperability or fairer pricing for third‑party agents, startups like Poke could suddenly gain access to hundreds of millions of European users overnight.
Local competition shouldn’t be underestimated either. From German enterprise automation firms to Estonian and Polish RPA and chatbot startups, Europe has a deep bench of workflow players. If Poke wants to be more than a novelty, it will need strong localization (languages, local services, EU tax and admin flows) and iron‑clad privacy guarantees.
Looking ahead
Several fault lines will determine whether Poke becomes infrastructure or a feature copied by others.
Distribution is first. Poke’s creator‑driven strategy — paying people per user they bring via shared recipes — is smart, but will need careful tuning to avoid spammy, low‑quality automations. If creators genuinely showcase high‑leverage workflows (for example, streamlining invoices for Spanish freelancers, or automating student schedules in France), organic growth could be substantial.
Second is sustainability of the pricing model. Right now the company is candidly prioritizing growth over profit. But real‑time automations against email, flights or developer tools are expensive. At some point, either prices rise, usage is throttled, or cross‑subsidy appears — potentially via enterprise plans or partnerships.
Third is trust and governance. Even with a multi‑layer security model, an agent wired into your inbox, health data and home devices is a juicy target. A single high‑profile breach or automation gone wrong (sending the wrong email, cancelling the wrong booking) could severely damage confidence. Expect to see much more emphasis on granular permissions, audit logs and “are you sure?” confirmations for sensitive actions.
Finally, the platform question: if Poke’s recipes ecosystem keeps growing, it starts to look like an app store for automations. That opens opportunities for third‑party developers, but also triggers a new set of responsibilities around ranking, moderation and safety — especially in the EU under the Digital Services Act.
The bottom line
Poke is an early, compelling glimpse of what truly mainstream AI agents could look like: invisible apps that live inside our chats and quietly wire together the services we already use. The idea is powerful enough that Big Tech will almost certainly imitate it; the open question is whether a small, model‑agnostic startup can stay ahead long enough to matter.
For users, the key decision isn’t “Should I try Poke?” but “Am I ready to let an AI sit between me and my digital life — and if so, under whose rules?”
