Cognition, the company behind an AI coding agent called Devin, paid over $100 million to acquire Poke, a text-based AI assistant launched just four months ago. On the surface, this looks like a strange match: a serious enterprise software company buying a consumer chatbot. The real story is about what the AI industry is realizing, fast, about why people actually keep using these tools.
Devin is not a small product. It is used by Goldman Sachs, Citi, Mercedes-Benz, and NASA, among others. One bank has reported compressing an eight-month software modernization project into eight days using it. Cognition's revenue grew roughly 13-fold in the twelve months leading up to May 2026. The company is currently valued at around $26 billion.
Poke, by contrast, was four months old and losing money. It worked by letting users text an AI as if texting a friend: no app, no login, just a phone number. It reached hundreds of thousands of users and racked up over 100 million messages without spending meaningfully on marketing, growing almost entirely through word of mouth. In June 2026, it became the first AI agent ever approved to run inside Apple's Messages platform, which is used by roughly one billion people.
So why did Cognition spend over $100 million on it? Because Poke had figured out something that most AI products have not: how to make people actually want to come back. Poke chats casually, uses humor, remembers context, and reaches out to users first rather than waiting to be prompted. It treats the person using it like a colleague treats a colleague. That sounds simple. It is, in practice, very hard to build.
Cognition wants Devin to feel the same way. Right now, Devin can only handle one task at a time and has no persistent memory between sessions. The plan is for Poke to eventually coordinate multiple Devin sessions simultaneously and carry memory across them, so that your AI coding colleague actually remembers what it was working on last week. That is a meaningful upgrade for any team using Devin.
The broader pattern here is worth noting. The raw AI capabilities offered by competing products are converging. OpenAI, Anthropic, Google, and others are all closing the gaps between their models. As that happens, the experience around the model, how it speaks, whether it feels like a tool or a collaborator, becomes the main reason someone chooses one product over another. This acquisition is Cognition placing an early and expensive bet on that logic.
For professionals managing teams, the takeaway is practical. The AI tools people in your organization actually use will be the ones that feel natural to interact with, not the ones with the best specifications. As these products become more embedded in daily work, personality and memory will drive habit formation. A cold, robotic tool will see lower adoption, regardless of how capable it is. A tool that checks in, remembers, and communicates warmly will become part of the routine.
Poke's co-founder noted that in the long run, he sees Poke as the layer that orchestrates multiple Devin sessions and maintains continuity between them. That is a meaningful shift in how we should think about AI at work: not a single tool you open and close, but something closer to a persistent colleague that is already working when you arrive and tells you what it found.
Whether that vision arrives in one year or five is an open question. What is no longer an open question is that the companies building AI products have decided personality is worth paying for.