Cognition has acquired The Interaction Company of California, the startup behind the texting-native AI assistant Poke, in a deal valuing the target in the low nine figures. The acquisition folds Poke's conversational interface and personality into Cognition's coding agent Devin, and gives Poke access to Cognition's models and infrastructure. Poke users exchanged more than 100 million messages over the past three months, a signal that a chatty consumer front-end has real engagement to bring to a developer tool.
Poke launched in March 2026 and reached hundreds of thousands of users by letting people talk to an AI agent through iMessage, SMS, Telegram, and, in some markets, WhatsApp. In June 2026 it became the first AI agent approved to run on Apple Messages for Business, Apple's channel for company-to-customer conversations inside the Messages app. Common use cases skew toward productivity — email triage, reminders, to-dos — alongside travel, health, finance, scheduling, and education tasks.
The pitch from Cognition co-founder Scott Wu is that Devin should feel like a colleague, not a command line. Wu and co-founder Walden Yan were angel investors in the Interaction Company before this deal, which suggests the acquisition path was seeded well before terms were signed.
“The Interaction team has built an agent that people love: it's proactive, it knows you, and it's fun to talk to. That's exactly how working with Devin should feel, and now we get to build it together.”— Scott Wu, Cognition co-founder
Key facts
- 01Cognition acquired The Interaction Company of California, maker of Poke, in a deal valuing the startup in the low nine figures.
- 02Poke users exchanged more than 100 million messages in the past three months across iMessage, SMS, Telegram, and WhatsApp.
- 03Poke launched in March 2026 and became the first AI agent approved for Apple Messages for Business in June 2026.
- 04Cognition co-founders Scott Wu and Walden Yan were earlier angel investors in the Interaction Company.
- 05Poke will begin using Cognition's SWE-1.7 software engineering model for some tasks starting next year.
Interaction Company co-founder Marvin von Hagen frames the logic in employment terms: a coding agent with personality is more pleasant to work alongside than a mute tool that returns diffs. That framing matters because Devin's competition — coding agents from Anthropic, OpenAI, and a growing field of startups — largely competes on benchmark scores and pull-request throughput, not vibe.
The bet is that as agent capabilities converge, the interaction layer becomes the differentiator. Two coding agents that both close 70% of GitHub issues will still feel very different if one texts you a joke while the other emits a stack trace.
The financial reality behind the deal is more prosaic. Poke was expensive to run and had struggled to turn a profit despite its user base, according to von Hagen. Cognition brings the infrastructure and model economics — including its newer SWE-1.7 software engineering model, which Poke will begin using for some tasks next year — that a standalone consumer texting agent could not sustain on its own.
For now, nothing changes for Poke users through the end of the year, and Poke will remain on Apple's platform. Next year the two teams will experiment with tighter integration, and a full merger of the products has not been ruled out.
Von Hagen sketched the technical direction: Poke could become an orchestration layer for multiple concurrent Devin sessions, since Devin today can only handle one pull request at a time. He also floated the idea of Poke serving as persistent memory across Devin sessions, so a coding agent can remember what it worked on last week without being re-briefed.
The counterweight is that personality is easy to talk about and hard to defend as a moat. Any frontier lab can tune a model to be warmer, funnier, or more familiar; OpenAI, Anthropic, and Google have all shipped voice and chat modes that lean into affect. Whether Poke's specific tone survives absorption into a coding tool — or gets sanded down to enterprise-safe defaults — is the open question. Consumer engagement metrics inside a texting app also do not automatically translate to willingness-to-pay inside a developer workflow.
For Cognition, this is a bet that the next battleground in AI agents is not raw capability but the interface wrapped around it. Devin competes in a crowded coding-agent market where switching costs are low and benchmark leadership rotates every few months. Owning a distinctive interaction layer — one that hundreds of thousands of consumers already text like a friend — is a plausible way to make Devin sticky in a way that a leaderboard score cannot. Whether personality is worth the low nine figures Cognition paid depends on how quickly rivals can copy the feel without the underlying team.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
