Remember when Devin’s big hook was that it could act like a software engineer, not just answer coding questions in a chat box? That positioning made Cognition one of the more watched companies in AI tooling. Now the company is making a very different kind of bet: Cognition bought Interaction, the California company behind Poke, a personal AI agent that lives in your texts.
The acquisition happened on July 23, 2026, and was valued in the low nine figures. Cognition’s stated direction is clear enough: bring Poke’s interaction style into Devin and focus harder on conversational AI personality. For a company known for a coding agent, buying a texting agent says a lot about where AI products are competing next.
Personality is no longer decoration
As someone who reviews AI toolkits for agntbox.com, I used to treat “personality” as a secondary feature. If the tool did the job, I could forgive a flat tone, awkward phrasing, or a clunky back-and-forth. That was especially true for developer tools. Coders tend to care about outputs, not charm.
But agentic products are different. When a tool is expected to take on multi-step work, ask clarifying questions, handle ambiguity, and keep a user oriented, the interaction layer becomes part of the product itself. A coding agent that cannot communicate well is not just unpleasant. It can become harder to trust, harder to correct, and harder to keep in the workflow.
That is why Cognition buying Poke makes sense. Poke is not described as another code model or developer utility. It is a personal AI agent that lives in your texts. Its relevance is not just what it can do, but how it fits into a familiar communication pattern. Texting is intimate, quick, and conversational. Bringing that style toward Devin suggests Cognition sees the interface as a competitive front.
Devin’s challenge is not only code
Devin already carries the identity of a coding agent. Cognition could have stayed focused on technical capability alone. Instead, this acquisition points to a broader view: an AI agent’s value depends on how well it collaborates with the person using it.
That matters because software work is full of judgment calls. Even without adding any claims beyond the known facts, we can say this much: coding work requires explanation, status updates, tradeoffs, and user intent. If Devin is going to feel more useful, it cannot simply produce outputs. It needs to communicate in a way that helps users stay in control.
Poke’s text-first identity gives Cognition a path to improve that layer. Texting agents are built around short exchanges, tone, timing, and continuity. Those traits map surprisingly well to the way users may want to manage an AI worker: quick check-ins, corrections, approvals, and context passed in plain language.
Low nine figures is a serious signal
The price matters. A low-nine-figure acquisition is not a casual acquihire. It signals that Cognition sees conversational style as strategically important, not merely as a nice interface polish.
For toolkit buyers, that is the part to watch. AI product teams are starting to compete on more than model access, task lists, and integrations. The next sorting mechanism may be how well a product behaves during the messy middle of work. Does it ask the right question? Does it explain itself clearly? Does it sound like a capable assistant or a brittle command line wrapped in a chat window?
Those qualities are hard to measure in a spec sheet, but they shape daily use. In reviews, I have learned that friction often hides in the conversation. A tool can look impressive in a demo and still feel exhausting after repeated use if it cannot keep context clear or respond in a natural rhythm.
What this means for AI toolkit reviews
This acquisition changes how I would evaluate agent tools going forward. I would still care about whether the tool completes the task. That remains the core test. But I would put more weight on the interaction model around the task.
-
Clarity: Does the agent explain what it is doing without burying the user in noise?
-
Control: Can the user redirect it easily when the work starts moving in the wrong direction?
-
Tone: Does the agent feel useful and direct, or does it perform confidence without earning it?
-
Fit: Does the interaction style match the job, whether that is coding, planning, or personal assistance?
Cognition’s Poke deal suggests these questions are moving from “nice to have” to central product strategy. Devin’s future may depend not just on being able to work, but on being able to work with people in a way that feels natural enough to keep using.
My read
I do not see this as a simple feature grab. Cognition bought a company whose product lives in texts because the company appears to believe that AI personality is becoming part of the moat. In plain terms, users may choose the agent that feels easiest to collaborate with, even among tools aimed at serious work.
That does not mean personality can cover weak execution. A charming agent that fails the task is still a bad tool. But a capable agent with a poor interaction style may also lose. The strongest products will likely pair real task performance with communication that reduces user effort.
For Devin, the Poke acquisition is a bet that coding agents need more than technical ambition. They need social usability. They need to know how to talk, when to ask, and how to stay present without becoming annoying. That is a much harder product problem than adding a friendlier greeting.
For the rest of us reviewing, buying, and testing AI tools, the lesson is simple: judge the conversation as part of the work. Cognition just paid low nine figures to do exactly that.
đź•’ Published: