“Poke is a personal AI agent that lives in your texts,” Cognition’s acquisition framing says, and that small phrase explains why this deal matters. As Dr. Lena Zhao, I read it less as a product line expansion and more as an architectural admission: for agentic systems, intelligence is no longer judged only by task completion. It is judged by how the system enters the user’s attention, negotiates ambiguity, and sustains a working relationship over time.
Cognition, maker of Devin, acquired Interaction, the company behind Poke, on July 23, 2026. The deal was valued in the low nine figures. The stated aim is to enhance Devin’s interaction style, with a focus on conversational AI personality. Poke brings a mobile, text-centered interface and a model of the agent as something that lives inside messaging behavior rather than behind a formal dashboard.
Why a coding agent wants a texting agent
On the surface, a coding agent and a texting agent may look like different species. Devin is associated with software work. Poke is described as a personal AI agent that lives in your texts. One suggests repositories, tasks, and technical execution; the other suggests short messages, reminders, and personal context.
Yet agent intelligence sits at the intersection of action and interaction. A system that can write code still has to ask questions, clarify intent, report progress, recover from confusion, and preserve user trust when the task changes. That is not decorative polish. It is part of the control loop.
In agent architecture, conversation is not merely the wrapper around reasoning. It is one of the main channels through which the agent receives constraints. Users rarely specify complete goals in a single prompt. They revise, interrupt, approve, reject, and redirect. The agent’s conversational personality shapes how those signals are elicited and interpreted.
Personality as an interface layer
“Personality” can sound soft, especially to engineers trained to value benchmarks and execution traces. But in agent systems, personality often functions as an interface policy. It determines when the agent is terse or explanatory, confident or cautious, proactive or passive. Those choices affect user behavior, and user behavior affects task success.
If Devin is being tuned to work with Poke’s interaction model, the acquisition suggests Cognition sees conversational behavior as part of product performance. A coding agent that feels awkward, abrupt, or opaque can lose the user long before the technical subsystem fails. Conversely, an agent that communicates with the right rhythm can make complex work feel manageable.
This is especially important for agents that operate across multiple steps. Multi-step work creates uncertainty. The user needs to know what the agent believes, what it plans to do next, and where it needs human judgment. A good interaction style reduces cognitive load without hiding the machinery. That is a difficult balance, and it is one reason a texting-native agent is strategically relevant to a coding-agent company.
Texting changes the agent contract
Poke’s key premise is that the agent lives in texts. That matters because messaging is not the same as a command console. Texting is interrupt-driven, informal, and persistent. It is where people issue partial thoughts, not formal specifications. An agent in that medium has to tolerate ambiguity and respond in a way that feels socially legible.
For Devin, that could mean a shift in how users experience software work. Instead of treating an agent as a tool summoned for a task, the user may come to experience it as an ongoing collaborator that can be reached through familiar communication patterns. The architectural challenge is to prevent friendliness from becoming noise and proactivity from becoming intrusion.
This is where personality becomes competitive. Not because users want theatrical bots, but because they want agents that are easier to direct. The agent that asks the better question at the right time can outperform a technically stronger system that asks too late, asks too much, or fails to ask at all.
Low nine figures buys more than a feature
The low-nine-figure valuation signals that Cognition is not treating conversational style as a small add-on. The purchase price indicates a serious bet that interaction design, mobile presence, and agent personality will influence who wins in the agent market.
From a research angle, I would frame the acquisition around three layers:
-
Task competence: the agent’s ability to execute work, such as coding-related tasks associated with Devin.
-
Conversational competence: the agent’s ability to communicate intent, uncertainty, progress, and requests for clarification.
-
Relational persistence: the agent’s ability to remain useful across repeated interactions in a familiar channel such as texts.
Poke appears to strengthen the second and third layers. Devin already represents Cognition’s bet on agentic software work. Poke adds a different axis: how the agent feels during use, and how naturally it fits into a user’s daily communication flow.
Agent intelligence is becoming social infrastructure
The phrase “AI personality wars” may sound overheated, but it points to a real technical issue. As agents become more capable, the bottleneck shifts from isolated output quality to coordination quality. A capable agent that cannot communicate well is like a brilliant teammate who cannot explain what they are doing.
Cognition buying Poke is therefore not just about making Devin friendlier. It is about making Devin easier to trust, steer, and keep in the loop. The future of agent architecture will not be defined only by planners, tool calls, and execution graphs. It will also be defined by interaction style: the subtle protocol through which humans and agents share intent.
For agntai.net readers, the lesson is clear enough. Personality is not a skin on top of intelligence. In agent systems, personality is part of the operating surface. Cognition’s purchase of Poke suggests that the next advantage in AI agents may come from systems that not only do the work, but know how to talk about the work as it unfolds.
đź•’ Published: