AI Summary View: Reactive to Proactive.

While preparing for the "AI in Procurement" webinar, I stepped back to consider the broader picture of AI’s developmental streams.

Spoiler alert: (Unfortunately,) Microsoft didn't sponsor it. As a business user predestined to utilize the Office 365 suite, I found no better way to use existing Microsoft apps natively, spiced with AI.

Diagram illustrating the evolution from Reactive AI to Proactive AI, featuring intersections of Assistive AI, Agentic AI, and Generative AI. It highlights key elements like Robotic Process Automation (RPA) for simple tasks, Intelligent Process Automation (IPA) for process creation, and Autonomous Agents for contextual adaptation. Microsoft Copilot and Power Automate are shown as tools enhancing AI capabilities.

These categories overlap. Generative AI describes a capability; assistive AI describes its supporting role; agentic AI describes how a system selects and executes actions. Autonomy concerns how much authority it has to act without human intervention. Read the diagram as a map of related approaches, not a fixed maturity sequence.

Robotic Process Automation (RPA)

I realized this recently when I started operating Microsoft Power Automate (formerly Microsoft Flow). RPA has been around for years, nicely packaged and readily available, yet I expected it to come prepackaged with e-procurement or ERP platforms.

RPA excels at executing simple, structured workflows with military precision: invoice data entry, PO validation, or three-way matching.

No learning, no adaptation—just relentless rule-following. Reliability and speed matter for repetitive tasks, and you can implement them without waiting for the "digital transformation."

Intelligent Process Automation (IPA)

Then Microsoft suggested moving up a level to IPA, where task automation gets a pinch of AI. IPA combines RPA with NLP, essential ML, and rule-based decision trees. You’ll see it in chatbots that understand queries or tools that triage supplier emails and route them appropriately. It’s brighter and fancier but still reactive.

Assistive AI

Now, we’re stepping into more complex territory. This stream supports decision-making, often by crunching large volumes of data and presenting recommendations. Contract risk scoring, supplier evaluation models, and spend forecasting tools fit here. It doesn’t do much on its own — but it gives us humans sharper tools.

Agentic AI

But what about AI that acts? Agentic AI can plan and execute multi-step tasks. It can take a business goal like “optimize supplier onboarding” and map out a series of steps: gathering data, reaching out, sending reminders, and escalating when stuck.

It’s still semi-dependent but shows initiative. It’s like hiring an intern who reads the policies and procedures.

Generative AI

Generative AI produces content such as supplier emails, document summaries, and draft market reports. It can support both a human user and an automated workflow.

Providing documents or feedback can improve an answer by changing the information available during the task. This does not necessarily mean the underlying model has learned permanently or been retrained.

For procurement, useful output requires relevant evidence, clear instructions, and checks proportionate to the consequences of an error.

Autonomous Agents

An agent can use tools, inspect results, and choose subsequent steps toward a goal. The degree of autonomy depends on its permissions, approval requirements, and stopping conditions. Managing other agents is an optional design choice.

For example, a supplier-onboarding agent could identify missing documents and prepare follow-up requests. Allowing it to send those requests or change supplier records requires explicit authority and controls.

Choose the Required Level of Autonomy

My recommendation is to start with the task and its consequences. Use a predefined workflow where the steps are stable. Consider an agent where the next action genuinely depends on what the system discovers.

Define what it may access, change, and communicate; when approval is required; and when it must stop or escalate.

Greater autonomy needs to justify its additional cost and risk through measurable improvements in service, quality, or capacity.

The name of the game: Reactive to Proactive

Essentially, the AI journey goes from “tell me what to do” to “I’ll figure it out.”

While Generative gets the limelight, the real workhorses are the agentic and autonomous types, which we need to master soon.

That’s where future value will be created—not by copy-pasting tender templates, but by systems that navigate ambiguity and act autonomously within governance remits.