Cognition's acquisition of Poke tells us something important: the era of competing purely on model capability is ending. The new battleground is how your AI product feels to use.
Cognition didn't buy Poke for a better language model or proprietary algorithm. They bought it for its conversational style and interaction design—specifically, how Poke's AI personality could make Devin, Cognition's AI coding agent, more engaging and useful in practice. This acquisition reflects a strategic shift happening across the AI industry: interaction design and AI personality are becoming differentiators, not nice-to-haves.
The Model Isn't the Product
For the past two years, founders building AI products have obsessed over which foundation model to use. GPT-4? Claude? Llama? The assumption was that a better model automatically meant a better product.
That assumption is breaking down. Most cutting-edge models now perform similarly on standard benchmarks. What separates a product users love from one they abandon isn't the underlying intelligence—it's the experience layer on top.
A capable model with a confusing interface loses to a slightly less capable model that feels intuitive and responsive. Users don't care if you're calling the latest API if your product can't explain what went wrong when it fails, if it talks like a robot, or if the interaction flow makes simple tasks feel complicated.
What "AI Personality" Actually Means
AI personality isn't about making your chatbot sound quirky. It's about deliberate interaction design across several dimensions:
Tone and voice. Does your AI sound helpful or condescending? Clear or vague? Formal or conversational? The wrong tone creates friction even when the output is technically correct.
Error handling. How does your product respond when it doesn't understand a request, when it makes a mistake, or when it needs clarification? Products that gracefully recover from errors feel trustworthy. Products that fail silently or defensively feel broken.
Guidance and scaffolding. Does your AI help users understand what it can do and how to ask for it? Or does it wait passively for perfectly-formed prompts? The best AI products reduce cognitive load by guiding users toward success.
Interaction patterns. Does your product support back-and-forth refinement, or does it treat every query as isolated? Can users interrupt, redirect, or undo? The conversational model you choose shapes whether users feel in control or at the mercy of the machine.
Poke understood this. Cognition recognized that bringing Poke's interaction expertise into Devin would make the product more competitive than building a slightly better code-generation engine.
Why This Matters for Founders Building AI MVPs
If you're building an AI product today, this acquisition should change how you allocate development time.
Start with interaction design, not model selection. Pick a capable model—GPT-4, Claude, whichever fits your use case—then spend your energy designing how users will interact with it. Test different prompting strategies, response formats, and error-recovery flows. The model is infrastructure; the interface is your product.
Test with real users immediately. You can't design good conversational UX in a vacuum. Put a rough version in front of users within days of starting development, watch where they get confused, and iterate on the interaction patterns. Does your AI explain its reasoning when users need context? Does it ask clarifying questions or make assumptions? These details determine whether your product feels helpful or frustrating.
Design for mistakes. Your AI will fail. It will misunderstand requests, generate incorrect outputs, and hit edge cases you didn't anticipate. How it handles those moments defines user trust. Build explicit error states, clear explanations, and easy ways to recover or rephrase.
Personality is strategic, not cosmetic. The way your AI communicates becomes your brand. A customer-support AI that's curt and robotic creates a worse brand impression than a slightly slower human agent. A coding assistant that's verbose and unfocused wastes developer time even if the code it generates is correct. Choose a personality that aligns with your product's job and your users' expectations.
Key Takeaways
- Cognition acquired Poke for conversational UX and AI personality, not model technology—signaling that interaction design is now a competitive advantage
- A capable model with poor UX loses to a slightly weaker model that feels intuitive and trustworthy
- Founders should invest development time in tone, error handling, user guidance, and interaction patterns from day one
- Test AI personality and conversational flow with real users early and iterate quickly
- How your AI communicates defines your brand and determines whether users trust your product in practice
Ship Fast, Design Deliberately
Building a working AI product with good conversational UX doesn't require months of design sprints. It requires disciplined iteration: ship a rough version fast, test it with users, refine the interaction patterns, and repeat.
If you're ready to move from idea to working, sellable MVP—with AI personality and interaction design built in from the start—get your MVP built in 3 days. Speed and quality aren't opposites when senior developers lead the process and real user testing starts immediately.