Quoting Boris Mann

The Spark: A Resonant Critique Developer Simon Willison's blog featured a quote from Boris Mann: “‘11 AI agents’ is meaningless as a phrase.” The reason this snippet resonated and spread quickly through developer circles is that it precisely hit a common pain point in the AI field today: terminological inflation and ambiguity. When everyone talks about “AI Agents” but each person might have a completely different definition, the word loses its communicative precision. Mann used a brilliant analogy—saying “I have 11 AI agents” is like saying “I have 11 spreadsheets” or “I have 11 browser tabs.” It states a quantity but says nothing about what these “things” are, what they can do, or how they work together. Deconstruction: The Semantic Slide from “Tool” to “Agent” To grasp the depth of this comment, we need to unpack the concept of “AI Agent.” In an ideal technical vision, a true AI Agent should possess autonomy, goal-orientation, the ability to use tools, and the capacity for complex interaction with other agents or the environment. However, in current marketing and everyday discussion, the bar for “Agent” has been drastically lowered. It might refer to a simple chatbot with a preset flow, a script that can call a few APIs, or a highly complex autonomous system. When everything is lumped under the umbrella of “Agent,” using “how many agents I have” to boast about capability becomes a form of jargon devoid of real information. It’s like before standardization, everyone called their product “smart,” ultimately devaluing the word itself. Trend Insight: A Harbinger of AI Application Stratification and Conceptual Clarity Boris Mann’s quip reveals a deeper trend: AI applications are rapidly differentiating and stratifying, and our language is already lagging behind the evolution of product forms. The industry is moving from “single model calls” to complex systems involving “multi-step, multi-tool collaboration.” In this process, there’s an urgent need for new, more precise vocabulary to describe different levels of automation and autonomy. For instance, distinguishing between “workflow automation” (fixed processes), “Co-pilot” (human-in-the-loop leading), and “autonomous Agent” (making independent decisions within a defined goal) becomes crucial. This conceptual clarification is often a hallmark of a tech field’s maturation. When the hype fades and people start scrutinizing “what this actually is,” genuine innovation and solid engineering practices will emerge. Practical Value: How to Think About and Evaluate “Agent” Products For developers, product managers, and entrepreneurs in the industry, the practical value of this statement lies in providing a “disenchanted” thinking framework. The next time you see marketing like “we have X AI Agents,” you can instinctively ask a few questions: What are the autonomy boundaries of these agents? What is the most complex task they can handle? How do they communicate and coordinate with each other? What are the failure and rollback mechanisms? Instead of focusing on quantity, it’s better to focus on the depth of a single agent’s capability, the architectural design of multi-agent systems, and the actual problems solved for users. This helps us see through marketing rhetoric and more rationally assess the true value of a technical solution. The Counterintuitive and Overlooked Angle An angle that might be overlooked is that this terminological confusion itself reflects the sheer speed of the shift in AI development paradigms. Developers are eager to embrace new possibilities, but the existing vocabulary (like “script,” “service,” “application”) is insufficient to describe these new entities that combine software characteristics with a degree of cognitive ability. Thus, “Agent” has become a temporary “container” bearing too many expectations. Boris Mann’s statement is less a criticism and more a call for the industry community to quickly establish a clearer, more pragmatic set of “jargon” to support the next phase of innovation and collaboration.