Cognition acquires conversational AI Poke in low nine-figure deal
The acquisition of The Interaction Company reflects a strategic shift in the artificial intelligence sector, where user experience and conversational style are increasingly viewed as critical competitive advantages alongside underlying model performance.
AI coding startup Cognition has acquired The Interaction Company, the operator of the conversational AI assistant Poke, in a transaction valuing the startup in the low nine figures. The deal aims to integrate Poke’s distinct, personality-driven interaction model into Cognition’s primary coding agent, Devin, reflecting an industry shift where user experience and conversational style are viewed as critical competitive advantages.
Poke, which launched in March 2026, allows users to interact via iMessage, SMS, Telegram, and WhatsApp, with over 100 million messages exchanged in the past three months. In June, Poke became the first AI agent approved to run on Apple’s Messages for Business platform. The assistant is characterised by a familiar, friendly tone, incorporating slang and humor, contrasting with traditional tool-like AI interactions.
Cognition co-founders Scott Wu and Walden Yan were previously angel investors in The Interaction Company. Wu stated that the acquisition allows the company to make Devin feel less like software and more like a colleague. The goal is to leverage Poke’s proactive and engaging nature to improve user engagement with Devin, which is currently designed as a software engineering tool.
Poke had been expensive to run, making profitability difficult despite hundreds of thousands of users. Under the new arrangement, Poke will leverage Cognition’s models and infrastructure to improve speed and reliability. Cognition plans to potentially utilise its SWE-1.7 software engineering model for some tasks next year.
No immediate changes are expected for Poke through the end of the year, and it will remain on Apple’s platform. Future integration plans include Poke potentially orchestrating multiple Devin sessions and providing persistent memory across tasks, though the timeline and specifics of the technical integration remain experimental.
