Agentic AI Won’t Work in Sourcing Until We Redesign the Work Itself 

Geraint John

May 29, 2026 • 5 min read

---

This month, in both our sourcing and talent roundtables with the community, we heard from many leaders about the challenges that come with moving from curiosity about agentic AI into deployment. In particular, CPOs and procurement leaders are mulling issues such as where agents can create value now, how much autonomy is realistic in direct vs indirect spend, and what guardrails need to be in place before agents start acting on the function’s behalf.

These concerns point to a larger shift in what matters when implementing agentic AI: this is an operating model design challenge.

Appetite for Change 

The appetite for implementation is clearly there. In Zero100’s most recent sourcing survey, 60% of respondents said they expect AI agents to perform more than a quarter of their sourcing team’s current work by the end of 2028. Almost one-fifth (19%) said it would be more than half. 

Sourcing leaders are taking a pragmatic view of how this will happen. Initially, they are segmenting the work based on risk. The strongest early use cases are the ones that are repetitive, rules-based, and document-heavy, think tail spend management, RFx support, supplier discovery, bid analysis, and intake triage. 

Several Zero100 community members, including an energy major and a global CPG company, for example, are focusing on previously unmanaged tail spend, where the stakes are lower, the economics are clearer, and the case for autonomous action is easier to make. 

For this type of discrete use case, the majority of sourcing leaders, according to our survey, expect to use third-party developed and managed agents within off-the-shelf solutions, such as those from Pactum, Fairmarkit, and Keelvar.

For more expansive and/or tailored use cases, on the other hand, there is an acknowledgment that sourcing-IT fusion teams may be needed to build an entirely new “virtual workforce” in-house.  

A large quick-service restaurant group and an industrial manufacturer shared with us that they are working to develop their own “super agents” across the source-to-contract process, and to connect planning signals with sourcing execution faster and more efficiently than has previously been possible (an example of what Zero100 calls PowerThreads). 

Governance and Data Challenges 

As procurement leaders consider the build vs buy question (and the new, non-negotiable questions the agentic era now forces), our discussions with both sourcing leaders and the wider supply chain community reveal data readiness and governance as a consistent top concern. 

On governance, the hard part is figuring out how AI agents should move through approvals, handoffs, and controls without creating more friction than they remove. The complexity here is magnified when the focus is highly engineered direct materials, where supplier decisions carry more technical, commercial, and continuity risk. 

One Zero100 community member, for instance, has been working through how multiple sub-agents might support supplier discovery, event execution, and analysis without breaking existing decision rights or governance structures.  

In terms of data readiness, the question is whether the underlying data is good enough to support meaningful action by AI agents. Members consistently raise doubts about whether their supplier, contract, and spend data are clean enough to support greater autonomy – often with good reason. This reinforces our prediction that master data management will be the #1 use case for AI this year. 

Critically, governance and data readiness underpin workflow and operating model design. Most sourcing leaders can already imagine a useful copilot or a task-specific agent. What they are struggling with is how to connect supplier data, spend signals, contracts, intake requests, market inputs, approvals, and escalation rules into a workflow that an agent can navigate. 

Operating Model Design 

In parallel, leaders need to address the role of humans in the autonomous sourcing organization of the future. Our conversations in recent months have surfaced understandable fears, but also a genuine desire on the part of leaders to expand the scope of what human-machine collaboration can achieve. 

Their ambition is not to remove people from sourcing; it is to move them away from low-value manual and tactical work and further into judgment-heavy work. Agents can handle repetitive execution, preparation, and analysis, while people retain control in areas such as category strategy, exception handling, supplier relationships, and final accountability.  

The key takeaway here is that leaders need to get precise about which tasks should move to agents, what should stay human, and how workflows need to change in between. That means reviewing the complexity, value, and risk associated with key steps within sourcing “jobs to done” (for example, selecting the best suppliers, ensuring cost competitiveness, and writing appropriate contracts). 

The Bigger Challenge 

While sourcing has spent years digitizing spend data, purchase orders, and payment processes, but agentic AI raises a much bigger hurdle, which is not solved by just adding more tech, even if it’s the right tech. It’s solved by redesigning work itself.