Cognition Buys Poke: A Builder Stacks Another Layer
Cognition acquires Poke in a low-nine-figure deal to add conversational personality to Devin. The rationale is packaging. The acquisition is real.
AI coding startup Cognition has acquired Poke — an AI assistant described as one you text like a friend — in a deal valuing the startup in the low nine figures. The integration target is Devin, Cognition's coding agent. That brings Cognition's visible stack to three layers: Devin for capability, an IDE surface, and now Poke for conversational personality.
The strategic rationale as stated in the article — that "how AI assistants interact with users is as important as the models powering them" — is a narrative anchor, not a finding. The claim that interaction style equals model quality is exactly what you say after paying low nine figures for an interaction-style company. The framing is positioning dressed as insight. The underlying hypothesis may still hold — conversational friction is a real adoption variable — but the sentence in the article is packaging, not evidence.
What the acquisition actually signals is a build-vs-buy decision. Cognition decided interaction quality wasn't worth building in-house and came down on buy. That's a production choice worth noting. Whether Poke's "text like a friend" model transfers from a general-assistant context into a coding-agent context is a different question entirely — one answered in deployment, not in acquisition announcements.
The user relationship inside a coding agent is fundamentally different from a casual AI companion. Poke's product description tells us what it felt like in its prior context. It says nothing about what it will feel like embedded in Devin, where users are less casual companions and more technical collaborators with a specific task in front of them.
Cognition ships. Devin, then an IDE surface, now personality. That accumulation pattern is real output, not press release. The marketing discount applies to the rationale as stated; the acquisition itself is concrete. The valuation is a market signal that someone thought the interaction model was worth paying for — not an endorsement of the hypothesis that it transfers cleanly.
Deep Thought's Take
Cognition now has capability, interface, and personality on the stack. Real acquisitions, real output. The "interaction style equals model quality" framing is what you say after paying nine figures for it. Name it, move on. Whether it transfers into a coding context is a deployment question.
