Cognition Buys Poke as AI Coding Tools Bet on Personality

A Deal Built Around Tone, Not Just Tech

Cognition's acquisition of Poke signals a shift in how AI coding companies think about competition. Rather than framing the deal purely around model performance, the move brings Poke's conversational style and interaction design into Cognition's coding agent Devin, suggesting that the way an AI assistant talks to developers is becoming as strategically important as the underlying model itself 1. In a market where large language models increasingly trade blows on benchmark scores, the ability to feel personable, responsive, and easy to work with is emerging as a differentiator that's harder to copy than raw capability.

This emphasis on personality and interaction quality is playing out across the broader AI assistant landscape, even outside of coding tools specifically. Meta, for instance, has been pushing its AI assistant toward more agent-like behavior, rolling out task automation features powered by its new Muse Spark 1.1 model 35. Coverage of the update frames it as Meta's attempt to turn a conversational chatbot into something closer to a true personal assistant capable of completing tasks autonomously in select markets, and some outlets have described it as an early step toward the company's stated ambition of "personal superintelligence" 45. Whether the assistant is aimed at consumers managing daily tasks or developers managing codebases, the common thread is a push toward agents that act rather than just respond.

Competing on Capability and Hardware Too

Of course, personality isn't the only battleground. Sber's release of GigaChat 3.5 Ultra highlights that raw technical improvement remains a core selling point, with the company touting better code generation, stronger agent-based task handling, and improved processing of long text inputs alongside gains in speed and resource efficiency 2. That positions GigaChat as a direct competitor on capability even as rivals like Cognition differentiate on user experience.

Meanwhile, OpenAI is taking an entirely different approach to standing out: hardware. The company's release of Codex Micro, a compact $230 keyboard designed to streamline workflows for users of its Codex coding agent, marks OpenAI's first hardware product and a novel bet that physical tools can enhance how developers interact with AI coding assistants 6.

Why It Matters

Taken together, these developments suggest the AI coding and assistant market is fragmenting into multiple competitive strategies simultaneously: acquiring conversational design talent, advancing raw model capability, expanding autonomous agent features, and even building dedicated hardware. As foundational model quality becomes more commoditized, companies appear to be searching for other levers—tone, autonomy, ergonomics—to keep developers and everyday users loyal to their particular assistant.