Cognition, the startup behind the Devin autonomous coding agent, has acquired The Interaction Company of California, maker of the conversational AI assistant Poke. The deal was valued in the "low nine figures," somewhere between roughly $100 million and $300 million, according to Poke co-founder Marvin von Hagen, who confirmed the price in an interview with TechCrunch. Cognition co-founder Scott Wu announced the acquisition in a post on the company's blog on July 23, 2026, titled "Welcoming The Interaction Company" (TechCrunch).

The acquisition brings two product lines under one roof. Poke, which launched in March 2026, lets users chat with an AI agent through iMessage, SMS, Telegram, and, in some markets, WhatsApp. In the three months before the deal's announcement, Poke users exchanged more than 100 million messages on the platform. In June 2026, Poke became the first AI agent approved to operate on Apple's Messages for Business platform. No immediate changes are planned for Poke through the end of 2026, and it intends to remain on Apple's Messages for Business. Starting in 2027, the combined team will experiment with using Cognition's software engineering model SWE-1.7 for some Poke tasks, and fully merging the two products has not been ruled out (TechCrunch).

The strategic logic centers on the interaction layer — the part of an AI product that handles how the agent actually communicates with users, manages context across a conversation, and maintains a consistent personality. Poke's interaction model and personality will be brought to Devin, Cognition's coding assistant, as a result of the acquisition. Conversely, Poke will gain access to Cognition's models and infrastructure to become faster and more reliable. The deal reflects a belief, articulated in TechCrunch's reporting, that AI personality and interaction style are becoming as competitively important as the underlying AI models themselves (TechCrunch).

The broader context here is worth pausing on. For the better part of two years, the competitive frontier in AI products has been defined by benchmarks, parameter counts, context windows, and inference cost curves. Those metrics still matter, but they are increasingly table stakes as leading models converge on similar raw capabilities. The differentiator is shifting toward how an agent communicates, how it manages conversational state across channels, and how it maintains a consistent persona that users actually want to engage with over time. Poke's 100-million-message volume in a single quarter suggests real user pull for a conversational agent that has a recognizable personality rather than a generic assistant tone.

The financial dynamics of the deal are also telling. Poke had been expensive to run and was struggling to turn a profit, according to von Hagen. An agent handling tens of millions of messages across multiple messaging platforms generates substantial inference costs, and monetization in the conversational AI space remains unsettled. Joining Cognition gives Poke access to infrastructure and model resources that could materially lower per-interaction costs, while Cognition gains a proven interaction layer it can fold into Devin. Whether the economics actually improve depends on how efficiently Cognition's infrastructure can serve Poke's traffic load, and how quickly the team can integrate SWE-1.7 into Poke's task pipeline next year.

The pre-existing relationship between the two companies is relevant context. Cognition co-founders Scott Wu and Walden Yan had previously been angel investors in The Interaction Company, meaning the acquisition is an extension of an existing financial relationship rather than a cold-market deal. That prior familiarity likely smoothed diligence and valuation discussions, and it means the two founding teams already had a working understanding of each other's technical approaches and product philosophies (TechCrunch; Cognition blog).

For Cognition, the bet is that a coding agent like Devin benefits from more than just raw code-generation capability. If developers are going to collaborate with an autonomous agent over extended sessions, the quality of that interaction becomes part of the product's competitive moat. Poke's expertise in multi-platform conversational design, honed across iMessage, Telegram, SMS, and WhatsApp, is directly applicable to that problem. The fact that Cognition is willing to spend nine figures to acquire that capability rather than build it internally signals either that the interaction layer is harder to engineer than it looks, that time to market matters more than build-versus-buy optimization, or both.

The decision to keep Poke operating independently through end of 2026, with integration experiments beginning in 2027, is a measured approach. It preserves Poke's Apple Messages for Business presence, which required its own approval process, and avoids disrupting a service with active user engagement while the teams figure out how the two product surfaces should interact. The option to fully merge remains on the table, which leaves the door open to a consolidated product should the integration experiments validate the combined approach.

What this acquisition makes possible, if the integration succeeds, is an AI coding agent that is not only capable of autonomous software engineering tasks but also communicates with developers in a way that feels natural and persistent across the channels they already use. That is a different value proposition from "better code generation," and it positions Cognition to compete on the full developer experience rather than on model benchmarks alone.