Agentic AI and Procurement (Part 1): What is Agentic AI?

06/26/2025 By Bertrand Maltaverne

Modules

Adobe Stock

Procurement has long been a testing ground for digital transformation, adopting automation and analytics at scale to streamline operations, enforce compliance and extract value from spend. However, the technologies behind these improvements have largely remained reactive. They support decisions and enhance workflows, but they rarely act themselves.

According to The Hackett Group’s 2025 CPO Agenda, 64% of procurement leaders expect AI — especially GenAI — to transform the way their teams operate in the next five years. GenAI is not just hype. 89% of executives are already advancing initiatives to operationalize it.

Deloitte’s 2025 CPO study confirms that trend, stating that GenAI deployment and flexible automation tools are the top strategies to manage potential workload volatility and enable greater agility. Approximately 40% of CPOs said GenAI deployment to be in the pilot stage.

Agentic AI is the next step in this journey.

The emergence of agents

A new generation of intelligent systems that only assists but operates independently is emerging. Known as Agentic AI, these systems are an evolution from previous rules-based automation, which follow predefined, static sequences. In contrast, agentic systems dynamically determine their own processes and tool usage and adapt to changing goals and conditions. Agents can be defined as systems in which large language models (LLMs) decide not only what to complete but also how to complete tasks based on real-time context and learned behavior. They are not tools to be operated but actors in their own right, capable of pursuing defined goals with minimal interventions. While the goal once was to develop AI assistants, the focus has shifted to agents.

Assistants: Intelligent applications that complete tasks for users using natural language and conversational interfaces. These assistants require a prompt of some sort to trigger their actions.
Agents: Intelligent programs designed to autonomously perform tasks on behalf of users or systems. This distinction reflects the agentic leap from user-triggered help to autonomous, goal-driven execution.

Agentic AI is built on the concept of agents. These agents interact with their environments, plan sequences of action, make decisions and refine their behavior over time. Crucially, they do so with a level of autonomy that allows them to initiate and carry out tasks without being explicitly directed at each step.

This ability to take goal-directed actions distinguishes Agentic AI from traditional automation or machine learning. While conventional models may offer insights, predictions or suggestions, agents operate against defined objectives. While scripts follow pre-coded rules, agents adjust strategies as conditions change. And where digital assistants respond to prompts, agents anticipate needs and act on them.

The potential lies not just in enhancing workflows but in enabling a broader value proposition where systems align with business outcomes rather than process steps.

To clarify how agentic AI differs from other technologies, the table below compares key characteristics and indicates the level of support of each technology.

Capability	RPA	Traditional AI/ML	Generative AI	Digital assistants / copilots	Agentic AI
Initiates action	No	No	No	Moderate	High
Responds to goals	No	No	No	Moderate	High
Learns from feedback	No	Yes	Moderate	Moderate	High
Adapts in real time	No	Moderate	Moderate	Moderate	High
Cross-system orchestration	No	Moderate	No	Moderate	High
Handles complexity	Moderate	Moderate	Moderate	Moderate	High
Works autonomously	No	No	No	No	High
Role	Task automation	Insight generation	Content generation	Decision support	Goal-driven execution

Agents redefine how work is distributed across systems and teams. In an agentic model, software no longer waits for instruction. It executes tasks within guardrails, escalates when necessary, and collaborates across functions. Human roles shift upstream: defining goals, setting controls and interpreting results.

At a functional level, several defining traits set Agentic AI apart from previous generations of technology:

Characteristic	Description
Autonomy	Acts independently toward defined objectives.
Proactivity	Initiates tasks based on inferred needs.
Context-awareness and memory	Maintains memory across workflows.
Tool and API orchestration	Connects across systems to execute tasks.
Multi-agent collaboration	Coordinates with other agents or services.
Goal orientation with feedback	Operates with outcome-based logic, not rules.

A major technical differentiator of agentic systems is their ability to manage context effectively. Unlike traditional automation or chatbots that process each step in isolation, agents maintain a persistent memory of tasks, constraints and interactions. Agents both maintain the background knowledge and operate autonomously, which allows them to make coherent decisions over time. This is essential when navigating procurement processes in which prior context drives the next best action, e.g., multi-round negotiations or supplier onboarding.

It is also important to recognize that there are various levels of agentic systems. Some may initiate only limited actions based on simple goals, while others can orchestrate complex workflows and coordinate other agents across various tools and services. Rather than aiming for full autonomy from the start, organizations can gradually evolve their agentic capabilities by starting with single-agent use cases and layering complexity as confidence and infrastructure mature.

Also, such a pragmatic approach would help organizations learn about agents and the changes they bring. As we will cover in other installments of this mini-series, agentic AI requires a reinforced AI governance because the orchestration of multiple agents introduces new layers of complexity and risk. Without strong orchestration logic, escalation protocols and shared context layers, autonomous agents can conflict, act redundantly or drift from business priorities. Governance becomes essential not just for what individual agents do but for how they collaborate in ways that remain auditable, compliant and value-aligned.

Agents in procurement

The relevance to procurement is immediate. No other business function operates across as many processes, approval gates, data sources, stakeholders and systems. From supplier discovery and contract negotiation to risk alerts and compliance monitoring, procurement processes are ideally highly structured yet often delayed by manual handoffs. Agentic systems are built to overcome this gap by not only automating steps but by executing end-to-end flows.

As we mentioned, while traditional automation can accelerate existing processes, agentic AI expands what is possible. Agents can manage intake triage, detect supplier risk events, coordinate sourcing events or trigger contract reviews based on external signals. These are not theoretical capabilities. Pilot deployments are already exploring these scenarios and bringing results. For example, Hackett’s 2025 research shows that early adopters have already realized productivity improvements of up to 10%, with leading organizations seeing benefits over 25% in effectiveness, cost efficiency and other areas.

When considering agentic AI in procurement, organizations should also consider that humans will still need to be in the loop because human judgment can be required due to inaccessible data, tacit/tribal knowledge and non-quantifiable factors like politics and urgency. Agents can act independently, but in many organizations, not all procurement knowledge is machine-accessible. Much of it lives in the minds of experienced professionals, in negotiation instincts, supplier dynamics and contextual awareness shaped by years of practice. Agentic systems can surface insights and automate action, but they still require human oversight to make sense of nuance, prioritize trade-offs and own the consequences of complex decisions.

Considerations

Real autonomy demands real readiness to ensure data accuracy, system interoperability, escalation clarity and governance protocols that support autonomous but auditable execution. Therefore, Procurement teams must ensure data access, system interoperability and clear escalation policies to facilitate seamless operations. API layers need to support persistent state and flexible interaction.

Governance frameworks must define both what agents can do and how their actions are reviewed and corrected. Frameworks that support agentic systems must expose control over state transitions, error handling and tool orchestration and ideally with transparent and auditable flows. Procurement leaders evaluating platforms should assess not only the capabilities of the agents but also the robustness of the agent orchestration framework itself. These considerations align with the findings of The Hackett Group, which identifies data quality, system complexity and governance gaps as the top barriers to implementation.

Misconceptions

Before we examine what agentic AI can deliver in practice (part 2 of this series will focus on the ROI of agentic AI and part 3 will cover use cases), it is worth addressing some common misconceptions. Although more solutions are branding themselves as ‘agentic,’ not all meet the criteria. Clarity at this stage is essential:

“Agentic AI is just another chatbot.” While both may use natural language, agents are designed to act autonomously, not just reply.
“It replaces humans.” Agentic systems are meant to augment strategic work, not eliminate teams.
“We already use AI, so this is nothing new.” Traditional AI provides insights. Agentic AI executes.

Understanding these aspects is critical because they influence and shape how procurement must reframe its approach to digital transformation. The shift is no longer about accelerating manual steps. It is about delegating them. That requires not just smarter tools. It needs systems that think and act in context.

Stay tuned for more!

The rest of our Agentic AI mini-series will cover: