
Sign up to save your podcasts
Or


Everyone at every procurement conference this spring said the same thing: "AI agents are the answer to procurement's problems..."
Almost nobody could tell you the question.
This week Matt and Joël cut through the noise with a working definition of what an AI agent actually is built on five components: trigger, outcome, guides, enablers, and process.
From there, they lay out three heuristic questions to run before deploying an agent anywhere near your procurement process:
Is the path to the outcome actually ambiguous?
Is a directionally-correct output recoverable if it's wrong?
Are your guardrails defined well enough to constrain it?
Plus: what the Anthropic-Salesforce stack battle signals about who owns the workflow layer, and more evidence of the ongoing convergence in procurement tech (Vendr/Vertice, Tropic/Omnea, Coupa/Rossum/Tonkean).
This one's for procurement and IT leaders who keep hearing "AI agent" in every vendor pitch and want a real framework for deciding when that's actually the right tool... Not just the current buzzword.
By Pure ProcurementEveryone at every procurement conference this spring said the same thing: "AI agents are the answer to procurement's problems..."
Almost nobody could tell you the question.
This week Matt and Joël cut through the noise with a working definition of what an AI agent actually is built on five components: trigger, outcome, guides, enablers, and process.
From there, they lay out three heuristic questions to run before deploying an agent anywhere near your procurement process:
Is the path to the outcome actually ambiguous?
Is a directionally-correct output recoverable if it's wrong?
Are your guardrails defined well enough to constrain it?
Plus: what the Anthropic-Salesforce stack battle signals about who owns the workflow layer, and more evidence of the ongoing convergence in procurement tech (Vendr/Vertice, Tropic/Omnea, Coupa/Rossum/Tonkean).
This one's for procurement and IT leaders who keep hearing "AI agent" in every vendor pitch and want a real framework for deciding when that's actually the right tool... Not just the current buzzword.