What the Heck is An "AI Agent"?

Archive | Upgrade to Premium | Sponsor

Hi readers,

I’m starting to get excited…

There are only 8 days left before the start of the 2025 ProcureTech Cup tournament!

What is the ProcureTech Cup?

A free, virtual learning event for procurement professionals to learn about the “art of the possible” with procurement technology.

64 leading ProcureTech brands face off in a bracket style tournament and you, the procurement professionals, vote to determine which on is LinkedIn’s favorite solution.

I’ll share all the details in next week’s post.

BUT, in the meantime, make sure you fill in a prediction bracket!

It’s free to play and the person with the best prediction gets a Pure Procurement prize pack.

Who knows… Maybe we’ll get a “perfect bracket”?? 👀

Onwards!

P.S. In tonight’s post I take on the new “buzz word” making the rounds… AI Agents!

📰 In this week’s edition:

• 🛠️ Webinar: Unlock Hidden Savings in MRO Procurement (sponsored)

• 📋 5 procurement jobs that caught my eye

• 🏆 The Road to the ProcureTech Cup : Episode 27

• 🌙 What the Heck Is an AI Agent?

Note: Some of the content listed above is only available in the email version of this newsletter. Don’t miss out! Sign up for free to get the next edition.

👀 In Case You Missed It…
My Best Linkedin post this week:

Play a game with me. I promise you'll learn something useful.

The Journey Continues…

This Friday Fairmarkit comes on the show to demo their market leading autonomous sourcing solution.

They’ve got a great perspective on AI agents to share!

Come check it out:

Last Episode

Mercanis was on the show last Friday to demo their unique Strategic Procurement Suite.

If you missed it, I’ve made ALL the replays available for you on YouTube:

What is the ProcureTech Cup? Start Here.

Did You Know…

There’s a Pure Procurement Newsletter YouTube Channel?

Lots of content planned for 2025. Subscribe now.

What the Heck Is an “AI Agent”?

In a world increasingly saturated with AI buzzwords, "AI agent" has emerged as the latest term that means everything and nothing…

Given all the content I’ve been seeing on AI Agents, I thought it was high time I give you my take on what these “digital workers” actually are, why they matter for your business, and how they're reshaping the future of work.

I promise I’ll sprinkle this perspective with my usual healthy dose of skepticism… I’m not trying to sell you anything…😅

Defining AI Agents for the Business-Minded

At its core, an AI agent is a software system that combines large language models (LLM) with other specialized tools (AI or otherwise) and persistent memory to independently solve problems and execute business tasks.

Unlike conventional software that follows rigid "if-this-then-that" rules, or basic chatbots that merely respond to prompts, agents can make decisions, use multiple tools, maintain context across interactions, and, most importantly, carry out concrete actions without constant human direction.

Think of them as digital procurement specialists with narrow but deep skill sets.

Unlike generative AI systems that can only generate text responses, agents are purpose-built for particular functions… Whether that's automating RFPs, analyzing supplier data, detecting tail spend opportunities, managing contract renewals, conducting market intelligence research, negotiating pricing with established vendors, validating supplier ESG compliance, running three-way matching for invoices, or proactively identifying and mitigating supply chain risks, your creativity is the limit of what an AI agent can do.

What sets agents apart is their ability to take action, not just provide information.

While a traditional LLM might tell you that a contract needs renewal, a properly designed AI agent will actually execute the entire workflow:

• Drafting the renewal agreement (in context),

• Validating this draft against YOUR contract and clause templates,

• Checking for pricing discrepancies against market benchmarks,

• Flagging potentially unfavorable terms,

• and routing it to the appropriate stakeholders for review,

• Etc…

…all with minimal human oversight and extracting/creating data in YOUR SYSTEMS as it goes. 🤯 

Another key distinction lies in who designs and controls the process.

With general LLMs, you're limited by the model's generic capabilities and whatever prompts you manually provide. With purpose-built procurement agents, your procurement experts design the exact workflow, business rules, and decision criteria the agent follows. They determine which systems it connects to, what thresholds trigger escalations, and how information flows through your organization.

The agent becomes an extension of your procurement team's expertise, embedded in a system that operates continuously according to your organization's specific playbook, not generic recommendations from a general-purpose AI.

An AI Agent brings your business process models to life… All feeds them steroids… And feeding those steroids steroids…

Aren't ChatGPT and Other Gen AI Tools Already Agents?

Yes and no.

Large language models (LLMs) like ChatGPT, Claude, Microsoft Co-Pilot and Gemini are indeed forms of artificial intelligence, but they lack several key characteristics of true AI agents.

These general AI systems excel at generating content and answering questions based on their training data, but they're essentially sophisticated prediction engines, generating the most statistically likely response to your prompts.

True AI agents build upon these foundational models but add crucial capabilities:

• Persistence (they remember past interactions)

• Goal-orientation (they work towards much more specific objectives), and most importantly;

• Ability to interact with other systems to accomplish tasks in the real world.

The Probabilistic Problem with Generative AI

The primary limitation of general-purpose AI systems is their probabilistic nature.

They don't "know" facts the way humans do… They predict the most likely sentences based on patterns in their training data. This statistical approach means they can confidently present incorrect information (“hallucinate”) or produce inconsistent results to identical inputs.

And the worst part is… They won’t even tell you they are unsure… Chat GPT is so arrogant it’s crazy 😂

Consider trying to use an LLM to analyze maverick spend in your organization. The model might give you a number that looks reasonable, but it's essentially making an educated guess based on statistical patterns rather than performing actual calculations on your real spend data.

That’s why when ProcureTech simply calls a “big LLM” model as part of their solution, it’s not very impressive…

Why?

It’s not going to produce the business outcomes you’re looking for…

For critical procurement functions, this probabilistic approach simply isn't good enough.

Why AI Agents Are Gaining Traction

This is precisely why AI agents are becoming so valuable.

By constraining powerful LLM models to specific domains and equipping them with the right tools, we can dramatically reduce their probabilistic behavior and increase their deterministic capabilities.

For example, a spend analysis agent might combine the language understanding capabilities of an LLM with direct access to your invoice database, advanced calculation abilities, and visualization tools. Rather than guessing at spending patterns, it operates a process where it performs actual calculations on real transaction data while still communicating results in natural language that business users understand.

This constraint-based approach delivers three primary benefits:

• Higher Accuracy. By narrowing the domain and connecting to authoritative data sources, agents make fewer mistakes.

• Greater Reliability. Deterministic processes produce consistent results every time. You’re not letting loose a Gen AI chatbot on your data… Gen AI is used a specific points in time in a business process supported by a string of different, use case appropriate technologies.

• Actionable Outputs. Agents don't just analyze; they can create transactions through integrated systems (e.g. when you’ve gathered all the specific information for the fields needed to create a requisition in our SAP system as explicitly mapped, create it!)

Gen AI is only but 1 tool in the “AI Agent” toolbox. Everything else you’re currently doing with other methods also fits into that toolbox:

• Data extraction and transformation

• Advanced calculation and modeling

• API integrations and data synchronization

• Automated workflow orchestration

• Document parsing and metadata extraction

• Anomaly detection and pattern recognition

• Database queries and report generation

• Data visualization creation

• Scheduled monitoring and alerting

• Multi-system data reconciliation

• Etc.

The Essential Human in the Loop

Despite their automation capabilities, the most effective AI agents aren't fully autonomous… They operate with humans in strategic oversight roles.

This "human in the loop" approach ensures that AI systems leverage human judgment for critical decisions while handling routine tasks independently. The level of human involvement varies based on risk and complexity:

• High-risk decisions such as strategic supplier selection and major contract negotiations require significant human oversight.

• Medium-risk activities such supplier performance analysis and RFP drafting benefit from human interpretation and review once all the compilation has been done.

• Low-risk, routine tasks such as PO matching, invoice processing and catalog data maintenance can be largely automated with routine human validations that ensure things are done according to your business rules.

This balanced approach allows organizations to capture efficiency gains while maintaining quality and ethical standards.

BUT (and long time readers will see me coming here… 😅)

This all sounds like you need Business Process Management skills to be successful??

Digital Literacy in the Age of AI Agents

Contrary to popular belief, AI agents don't reduce the need for digital literacy… They amplify it.

To effectively deploy and manage these systems, procurement organizations need team members who understand:

• Source-to-pay process design and optimization

• Spend and master data quality management and governance

• System integration principles

• Workflow automation principles

• AI/ML fundamentals and limitations in procurement contexts

• Etc..

However, AI agents dramatically change the need for distribution of this knowledge to be successful!

Instead of requiring broad technical proficiency across an entire workforce, organizations can focus specialized capabilities in a smaller group of "AI champions" who configure and manage these systems for broader use.

Everyone else can just “chat with the process” without needing much training.

The Amplified Productivity Gap

This shift brings us to a critical consideration about workforce productivity…

Typically, across organizations, a small minority of employees typically generate a disproportionate share of the output. This isn't just corporate folklore… It's a pattern observed repeatedly across industries and disciplines, where a relatively small percentage of people produce the majority of results and value.

AI agents are poised to dramatically amplify this effect, creating what I call the "AI productivity multiplier."

Your most productive team members become AI orchestrators. Technical specialists who design, implement, and refine business processes with AI systems. These AI champions gain exponentially more leverage, potentially doing the work of dozens of knowledge workers. The productivity gap between organizations with effective AI champions and those without widens dramatically.

This isn't just a minor shift in efficiency… It's a fundamental restructuring of how value is created and distributed within organizations.

This is exactly what has happened with the internet in the last 30 years… But on a different level…

The Open Question: Valuing the Majority

This brings us to the most challenging question of the AI age:

In a world where a small percentage of technical superstars can accomplish exponentially more with AI agents as their force multipliers, how do we ensure the vast majority of society remains valued and retains their dignity?

This isn't just a philosophical question but a practical business concern.

When your most skilled knowledge workers can do the work of dozens of others through AI augmentation, organizations must reconsider:

• How to measure and reward contributions beyond technical output

• What skills and attributes become more valuable (empathy, creativity, ethical judgment)

• How to create meaningful roles that leverage uniquely human capabilities

• What responsibility businesses have in reskilling and supporting workforce transitions

Our businesses and societies also have to decide how the additional wealth generated by AI-led improvements will be distributed…

• Only to the relative few who learn to wield it successfully and their shareholders?

• To society at large?

(especially since LLM relies on all the knowledge made available on the internet by others… For free…)

Realistically, just as it happened during the internet age, businesses will need smaller workforces to accomplish growing amounts of work… Those who don’t (or fail to) compete, will get left evermore behind.

But again, this is the “next level.”

So… Where do you fall on this issue?

This won’t necessarily guide how you should tackle AI Agents… But it should start informing how you manage the inevitable improvement/fallout (depending on where you’re sitting).

Moving Forward with AI Agents

Regardless, for business leaders looking to explore AI agents, I recommend a measured approach:

• Identify specific procurement processes with clear metrics and high routine components (invoice processing, catalog management, basic sourcing)

• Start small with focused agents that solve discrete problems (e.g., an agent that analyzes payment terms or identifies consolidation opportunities)

• Identify and empower your most productive team members who can drive disproportionate value

• Implement strong governance and quality control mechanisms with clear approval workflows

• Create feedback loops to continuously improve agent performance in procurement contexts

• Continue deploying agents in riskier use cases, while applying your lessons learned.

The future of procurement isn't humans versus AI… It's humans and AI versus outdated ways of working.

By thoughtfully implementing AI agents with appropriate human oversight, procurement organizations can achieve both efficiency and strategic value while creating meaningful work for people at all levels.

The question isn't whether AI agents will transform procurement, but whether you'll be leading that transformation or be left behind by competitors who got there first.

Did I miss anything?

Let me know in the comments 👇

1. Need Help Building Your Digital Procurement Roadmap?
   I’ve been helping global procurement teams digitalize their processes and practices for 12+ years. Reply to this email to get a conversation started.

2. You have something to share with our 10,000+ readers?
   The digitally-minded procurement professionals reading this newsletter are thirsty for knowledge, insights, and solutions. Reply to this email to get in front of them.