Vetting AI Vendors in the Age of 'Agent Washing'

'Agent washing' — rebranding an existing chatbot, RPA tool, or basic AI assistant as an 'agent' without adding genuinely agentic capability — has become common enough that Gartner estimates only a small fraction of self-described AI agent vendors offer real agentic functionality. For buyers evaluating AI agent vendors, distinguishing genuine agentic capability from a relabeled existing product is now a core part of procurement, not an edge case to occasionally watch for.

What Actually Makes an AI System 'Agentic'

• Autonomous multi-step planning — breaking a goal into steps and executing them, not just responding to a single prompt with a single output
• Tool use — the ability to actually call external systems, APIs, or data sources to complete a task, not just generate text describing what should happen
• Persistent memory or state across a task — tracking progress through a multi-step process, not treating each interaction as independent
• Some degree of autonomous decision-making within defined boundaries — not every step requiring explicit human direction

Red Flags Worth Testing For During Evaluation

• A demo that only shows a single-turn conversation, never a genuine multi-step task execution with real tool calls
• Vague answers when asked exactly which systems the 'agent' can actually take action in, versus which it can only describe or recommend
• Marketing language emphasizing 'AI-powered' generally, without specificity about autonomous planning or execution capability
• A product that was clearly an existing chatbot or RPA tool rebranded, with 'agent' language added to marketing without a corresponding architecture change

Practical Questions to Ask During a Vendor Demo

• Show me the agent handling a task that requires more than one step and more than one tool call, live, not a pre-recorded demo
• What happens when the agent encounters an unexpected obstacle mid-task — does it adapt, or does it fail silently
• What's the actual mechanism by which the agent decides what to do next at each step — this should be a specific, technical answer, not a marketing description
• Can I see a real, deployed customer reference using genuinely autonomous multi-step execution, not just a proof-of-concept demo

Why This Matters Beyond Avoiding a Bad Purchase

Beyond wasted spend, deploying a relabeled non-agentic tool under agentic governance assumptions — the decision-tier classification and audit-logging practices covered elsewhere in this content series — creates a real gap: your governance framework assumes autonomous decision-making that the tool doesn't actually perform, and your actual risk exposure from genuinely autonomous tools elsewhere goes under-scrutinized by comparison.