What Happened
A piece by Lex Freer draws a pointed comparison between AI agents and Android phone cameras. The argument: just as Android devices carry a perception penalty despite hardware that matches or exceeds iPhones, AI agents are fighting a losing battle against years of terrible chatbot experiences.
The core claim is blunt. It does not matter if your AI agent is technically the best one available. If people have used agents before and hated them, they will not try yours. Freer describes the universal frustration: "Every time I connect with a custom support chatbot I want to smash my phone." That sentiment is the real competition, not other AI products.
The piece identifies what it calls an "authenticity gap." Users know the agent is not human, and no amount of technical polish compensates for that fundamental substitution. The solution, Freer argues, is not better models - it is a breakthrough brand moment comparable to what the iPhone did for smartphone cameras. Someone needs to build the agent experience so good that it resets expectations entirely.
Why It Matters
This diagnosis rings true for anyone who has tested AI agents in production. The technology has improved substantially over the past year. Tools like ChatGPT's custom GPTs, Claude's tool use, and dedicated platforms like Ada and Drift have real capabilities. But user sentiment lags behind the technology by a wide margin.
If you are evaluating AI agents for customer support, sales, or internal workflows, the technical benchmarks tell only half the story. The other half is whether your users or customers will actually engage with the agent instead of immediately looking for the "talk to a human" button.
This matters for tool selection. A slightly less capable agent with better UX, clearer branding, and more thoughtful conversation design will outperform a technically superior one that triggers chatbot PTSD from 2023-era experiences.
Our Take
Freer is right about the diagnosis but underestimates the speed of the fix. The iPhone comparison is instructive but misleading in one way: Apple had to ship physical hardware to change camera perception. AI companies can update their agents overnight.
We are already seeing this happen. The gap between a 2024 support chatbot and a 2026 agent built on current models is enormous. The problem is that most companies deploying agents are still building them like chatbots - rigid decision trees wrapped in natural language. The brands that win will be the ones that let the model actually think, rather than constraining it into the same bad patterns that gave chatbots their reputation.
The real takeaway for practitioners: if you are deploying agents, invest as much in conversation design and user onboarding as you do in model selection. The perception problem is real, and it will not solve itself with better benchmarks.