Procurement Insights · May 17, 2026
One of the most important realizations emerging from the current AI wave is that orchestration and agents do not eliminate operational fragmentation. They coordinate on top of it.
That distinction matters.
Across procurement, ERP modernization, AI governance, supply chain transformation, and enterprise architecture, the market increasingly assumes better orchestration will normalize inconsistent environments. The Coupa Inspire conversations this past week brought that assumption into sharp focus. Xavier Olivera at The Hackett Group surfaced it directly in a post that quickly became one of the most substantive public discussions of AI in procurement technology I have seen in 2026, with comments from Tanya Seda, Fabrice Saporito, Jason Busch, and David McClintock that collectively pointed at the same underlying issue from five different analytical directions.
Xavier’s framing of the core question deserves to be quoted in full because it is closer to a structural diagnosis than most procurement-technology commentary ever reaches:
“An agent can only execute the operational capabilities the underlying platform already supports. If procurement, AP, supplier management, inventory, logistics, and finance processes still operate with fragmented workflows and disconnected operational data, adding agents may improve interaction without necessarily improving execution.”
That observation reframes the entire AI-in-procurement conversation. The question is no longer can agents execute tasks. The question is can agents operate reliably across inconsistent enterprise environments. That is a substrate question. And it is the same diagnostic question every other layer of the 2026 industry conversation has been circling without yet naming.
What Orchestration Does Not Resolve
Orchestration is the coordination layer. It moves work between systems, sequences dependent operations, and manages handoffs that previously required human intervention. When orchestration works well, the speed of cross-system coordination improves dramatically and the visible execution becomes smoother.
What orchestration does not do is normalize the underlying conditions being coordinated. It does not resolve:
The fragmented workflows that organizations inherited from previous technology waves and never refactored when the next wave arrived. The ERP customizations that have accumulated for fifteen or twenty years and now operate as undocumented business logic embedded in fields no one remembers configuring. The shadow processes that emerged when the official systems could not support what the operation actually required, and that now coexist with the official systems in a parallel reality the official systems cannot see. The undocumented operating logic that lives in spreadsheets, in personal email folders, in side conversations between functional leads, and in the heads of long-tenured employees who are increasingly retiring. The incentive asymmetries that reward transactional behavior even when the strategy calls for relational behavior. The accumulated coordination debt — the unresolved cross-system workarounds layered across previous technology waves where each wave was deployed on top of the inheritance from the prior wave without resolving any of it.
Orchestration coordinates across all of this faster. It does not make any of it consistent.
That is the structural reality the procurement-technology conversation is now beginning to recognize. And once you recognize it, the question that follows is uncomfortable: what happens when autonomous agents are deployed at scale on top of substrate conditions that were never validated?
The Pattern That Connects the Convergence
Xavier’s observation is not isolated. It is one node in a now-undeniable pattern across the 2026 industry conversation.
The Kearney 2026 Assessment of Excellence in Procurement Study documents a forty-six-million-dollar performance gap between leader and mid-tier procurement organizations that compounds at ten percent annually. The four-pillar Kearney model measures team capability, demand shaping, category strategy, and supplier programs — all operating above the substrate layer.
The McKinsey 2026 State of AI Trust report finds that organizations investing twenty-five million dollars or more in responsible AI report significantly higher maturity scores. The investment threshold is real, and the maturity differential is real, but the report does not name what makes that investment produce returns in some organizations and not in others.
The EDUCAUSE 2025 AI Landscape Study finds that only nine percent of higher education institutions believe their AI policies are adequate to address actual AI risk. As Tiffany Masson observed recently on LinkedIn: policies exist; systems do not.
Bain on the foundation for agentic AI. KPMG on why returns vary so widely. BCG on managing AI as a coworker rather than a tool. Accenture on rebuilding the platform. IBM on operating models for autonomous systems. And now Hackett on agents executing against operational capabilities the underlying platform already supports.
Different vocabulary. Different analytical methodologies. Different industries. Same underlying observation.
The market is converging on the substrate problem without yet having shared vocabulary for it. Xavier called it operational inconsistency. Tanya called it workflows, dependencies, and data quality. Fabrice called it reengineering process architecture itself. Jason called it AI software architecture enablement. David called it enterprise IT complexity reality. McKinsey called it responsible AI investment. Kearney called it the leader-versus-laggard performance gap. EDUCAUSE called it policy-system inadequacy.
These are all the same thing observed from different vantage points. The thing they are observing is the load-bearing layer beneath the orchestration. The substrate.
In the dual-pyramid model I introduced last week following the Nico Bac and Jason Busch orchestration debate, orchestration sits at the visible apex of the technology stack; the substrate is the narrowing point beneath it that everyone above the waterline assumes is stable. What all of these studies are circling without yet sharing vocabulary for is the load-bearing reality of that substrate.
Why Orchestration Improves Interaction Without Improving Execution
The technical reason orchestration improves interaction is that orchestration is a coordination protocol operating on standardized interfaces. An AI agent invokes an API. The API responds. The agent sequences the next call based on the response. The interaction layer becomes faster, more consistent, and more accessible to non-technical users. That is real improvement, and it is the improvement Coupa Inspire and similar industry events showcase.
The technical reason orchestration does not necessarily improve execution is that execution depends on the operating conditions the orchestrated systems sit on top of. If the underlying procurement workflow contains a step that was customized in 2014 by an analyst who has since left the organization, and the customization assumes a supplier classification scheme that no longer matches the actual supplier base, the agent invoking that workflow will execute against the customization perfectly. The agent will not detect that the customization is producing systematically incorrect supplier categorization because the agent operates above the substrate, not on it.
This is the structural failure mode AI deployments are now starting to encounter at scale. The agent executes faster than the human did. The agent makes the same systematic error the human was making, except now at machine speed across thousands of transactions per hour. The error compounds. The downstream effects accumulate. And the post-mortem six months later concludes that the AI was the problem, when the actual problem was the substrate the AI was deployed on top of.
AI does not break broken processes. It perfects them, faster, at scale, with less human friction to slow down the compounding error.
The Long-Term Differentiator
For organizations evaluating AI investment in 2026, the long-term differentiator may not be who deploys the best agents or selects the best orchestration platform. It may be which organizations can sustain reliable coordination under real-world operational load — and that capacity is a property of the substrate, not of the orchestration architecture deployed on top of it.
That reframe has consequences for how procurement leaders, CIOs, and boards should be thinking about AI investment decisions over the next twelve to twenty-four months. The questions that matter shift from feature comparison and vendor selection toward substrate diagnosis.
What inherited fragmentation does our current operating environment contain? Which prior-wave customizations are now operating as undocumented business logic that AI agents will inherit when deployed? Where do shadow workflows coexist with official systems, and what happens when an autonomous agent encounters the inconsistency between them? Which incentive structures will continue to reward transactional behavior even after the AI is deployed to support relational outcomes? How much accumulated coordination debt is the AI being asked to coordinate across, and is that debt visible enough that the deployment team can surface and address it before deployment rather than inheriting it?
These questions do not appear in vendor-led AI procurement sales cycles. They do not appear in most internal AI readiness assessments either, because most internal assessments focus on the visible orchestration layer rather than on the substrate beneath it. The questions appear when the deployment is six months in and the compounding effects of substrate inconsistency become operationally undeniable, by which point the investment has been committed and the remediation cost is materially higher than the diagnostic cost would have been.
That timing asymmetry is the structural reason pre-commitment (Phase 0™) diagnostic work matters. The substrate questions need to be asked before the commitment, not after.
What the Procurement-Technology Conversation Is Now Ready to Name
The most significant thing about Xavier’s post and the comment thread it generated is not the observations themselves. It is the readiness of the procurement-technology analyst community to engage with substrate-level questions publicly. A year ago, the public conversation would have been about agent capability, orchestration architecture, and vendor positioning. The structural questions would have been treated as too abstract for the operational discussion.
That has shifted. The community is now articulating the substrate problem in vendor-neutral language and discussing it as a first-order operational reality rather than as a theoretical concern. Xavier’s question — can agents operate reliably across inconsistent enterprise environments — would not have surfaced publicly twelve months ago. It is surfacing now because the experience of deploying AI into actual enterprise environments has made the substrate problem operationally visible to practitioners who previously did not need to engage with it.
That readiness creates an opportunity for the substrate conversation to mature from observation into diagnostic. Observation tells us the problem exists. Diagnostic tells us what to do about it. The frameworks I have been documenting in this archive since 1998 — Implementation Physics™, the Compounding Technology Shadow Wave™, Hansen Strand Commonality™, and the Phase 0™ Diagnostic — exist to perform that diagnostic work. The eighteen-year contemporaneous archive provides the longitudinal evidence base. The dual-pyramid graphic provides the visual language. The substrate problem the procurement-technology community is now naming in different vocabularies is the problem these frameworks were built to diagnose.
The Bottom Line
Orchestration coordinates faster across substrate that was never normalized. Agents execute reliably against operational capabilities the underlying platform already supports, and they execute unreliably against operational capabilities the platform does not support no matter how sophisticated the agent itself becomes.
That is the structural reality the procurement-technology conversation is now ready to engage with directly. Xavier Olivera surfaced it. Tanya Seda, Fabrice Saporito, Jason Busch, and David McClintock extended it from different analytical directions. The Coupa Inspire conversations made it operationally concrete. The broader 2026 consulting-firm convergence — Kearney, McKinsey, Bain, KPMG, BCG, Accenture, IBM, Hackett — is documenting it across the analytical landscape.
The orchestration layer is visible. The substrate underneath it is load-bearing. That is where AI investments will succeed or fail over the next eighteen to thirty-six months, and the diagnostic question that determines which trajectory an organization is on is not a question about the AI. It is a question about the conditions the AI is being deployed into.
The reports describe the gap. The pyramids name the cause. Phase 0™ is the diagnostic that runs before the next major commitment is made.
This post was developed through the ARA™ RAM 2025™ multimodel validation framework. Five independent models reviewed the Hackett observation, the Coupa Inspire thread, and the substrate diagnosis prior to publication. All five converged on the same structural diagnosis. The internal convergence mirrors the external convergence now documented across Hackett, Kearney, McKinsey, Bain, KPMG, BCG, Accenture, and IBM in their 2026 publications. The convergence itself is becoming evidence of the structural pattern.
The Compounding Technology Shadow Wave™ trilogy executive summaries are available at procureinsights.com. The Phase 0™ Diagnostic — for organizations preparing to commit further AI or orchestration investment — is at hansenprocurement.com/where-does-your-organization-sit-right-now/.
Hansen Models™ · Implementation Physics™ · Compounding Technology Shadow Wave™ · Phase 0™ · Hansen Fit Score™ · Hansen Strand Commonality™
-30-
Procurement Insights
Orchestration Does Not Solve Substrate Inconsistency
Posted on May 17, 2026
0
Procurement Insights · May 17, 2026
One of the most important realizations emerging from the current AI wave is that orchestration and agents do not eliminate operational fragmentation. They coordinate on top of it.
That distinction matters.
Across procurement, ERP modernization, AI governance, supply chain transformation, and enterprise architecture, the market increasingly assumes better orchestration will normalize inconsistent environments. The Coupa Inspire conversations this past week brought that assumption into sharp focus. Xavier Olivera at The Hackett Group surfaced it directly in a post that quickly became one of the most substantive public discussions of AI in procurement technology I have seen in 2026, with comments from Tanya Seda, Fabrice Saporito, Jason Busch, and David McClintock that collectively pointed at the same underlying issue from five different analytical directions.
Xavier’s framing of the core question deserves to be quoted in full because it is closer to a structural diagnosis than most procurement-technology commentary ever reaches:
That observation reframes the entire AI-in-procurement conversation. The question is no longer can agents execute tasks. The question is can agents operate reliably across inconsistent enterprise environments. That is a substrate question. And it is the same diagnostic question every other layer of the 2026 industry conversation has been circling without yet naming.
What Orchestration Does Not Resolve
Orchestration is the coordination layer. It moves work between systems, sequences dependent operations, and manages handoffs that previously required human intervention. When orchestration works well, the speed of cross-system coordination improves dramatically and the visible execution becomes smoother.
What orchestration does not do is normalize the underlying conditions being coordinated. It does not resolve:
The fragmented workflows that organizations inherited from previous technology waves and never refactored when the next wave arrived. The ERP customizations that have accumulated for fifteen or twenty years and now operate as undocumented business logic embedded in fields no one remembers configuring. The shadow processes that emerged when the official systems could not support what the operation actually required, and that now coexist with the official systems in a parallel reality the official systems cannot see. The undocumented operating logic that lives in spreadsheets, in personal email folders, in side conversations between functional leads, and in the heads of long-tenured employees who are increasingly retiring. The incentive asymmetries that reward transactional behavior even when the strategy calls for relational behavior. The accumulated coordination debt — the unresolved cross-system workarounds layered across previous technology waves where each wave was deployed on top of the inheritance from the prior wave without resolving any of it.
Orchestration coordinates across all of this faster. It does not make any of it consistent.
That is the structural reality the procurement-technology conversation is now beginning to recognize. And once you recognize it, the question that follows is uncomfortable: what happens when autonomous agents are deployed at scale on top of substrate conditions that were never validated?
The Pattern That Connects the Convergence
Xavier’s observation is not isolated. It is one node in a now-undeniable pattern across the 2026 industry conversation.
The Kearney 2026 Assessment of Excellence in Procurement Study documents a forty-six-million-dollar performance gap between leader and mid-tier procurement organizations that compounds at ten percent annually. The four-pillar Kearney model measures team capability, demand shaping, category strategy, and supplier programs — all operating above the substrate layer.
The McKinsey 2026 State of AI Trust report finds that organizations investing twenty-five million dollars or more in responsible AI report significantly higher maturity scores. The investment threshold is real, and the maturity differential is real, but the report does not name what makes that investment produce returns in some organizations and not in others.
The EDUCAUSE 2025 AI Landscape Study finds that only nine percent of higher education institutions believe their AI policies are adequate to address actual AI risk. As Tiffany Masson observed recently on LinkedIn: policies exist; systems do not.
Bain on the foundation for agentic AI. KPMG on why returns vary so widely. BCG on managing AI as a coworker rather than a tool. Accenture on rebuilding the platform. IBM on operating models for autonomous systems. And now Hackett on agents executing against operational capabilities the underlying platform already supports.
Different vocabulary. Different analytical methodologies. Different industries. Same underlying observation.
The market is converging on the substrate problem without yet having shared vocabulary for it. Xavier called it operational inconsistency. Tanya called it workflows, dependencies, and data quality. Fabrice called it reengineering process architecture itself. Jason called it AI software architecture enablement. David called it enterprise IT complexity reality. McKinsey called it responsible AI investment. Kearney called it the leader-versus-laggard performance gap. EDUCAUSE called it policy-system inadequacy.
These are all the same thing observed from different vantage points. The thing they are observing is the load-bearing layer beneath the orchestration. The substrate.
In the dual-pyramid model I introduced last week following the Nico Bac and Jason Busch orchestration debate, orchestration sits at the visible apex of the technology stack; the substrate is the narrowing point beneath it that everyone above the waterline assumes is stable. What all of these studies are circling without yet sharing vocabulary for is the load-bearing reality of that substrate.
Why Orchestration Improves Interaction Without Improving Execution
The technical reason orchestration improves interaction is that orchestration is a coordination protocol operating on standardized interfaces. An AI agent invokes an API. The API responds. The agent sequences the next call based on the response. The interaction layer becomes faster, more consistent, and more accessible to non-technical users. That is real improvement, and it is the improvement Coupa Inspire and similar industry events showcase.
The technical reason orchestration does not necessarily improve execution is that execution depends on the operating conditions the orchestrated systems sit on top of. If the underlying procurement workflow contains a step that was customized in 2014 by an analyst who has since left the organization, and the customization assumes a supplier classification scheme that no longer matches the actual supplier base, the agent invoking that workflow will execute against the customization perfectly. The agent will not detect that the customization is producing systematically incorrect supplier categorization because the agent operates above the substrate, not on it.
This is the structural failure mode AI deployments are now starting to encounter at scale. The agent executes faster than the human did. The agent makes the same systematic error the human was making, except now at machine speed across thousands of transactions per hour. The error compounds. The downstream effects accumulate. And the post-mortem six months later concludes that the AI was the problem, when the actual problem was the substrate the AI was deployed on top of.
AI does not break broken processes. It perfects them, faster, at scale, with less human friction to slow down the compounding error.
The Long-Term Differentiator
For organizations evaluating AI investment in 2026, the long-term differentiator may not be who deploys the best agents or selects the best orchestration platform. It may be which organizations can sustain reliable coordination under real-world operational load — and that capacity is a property of the substrate, not of the orchestration architecture deployed on top of it.
That reframe has consequences for how procurement leaders, CIOs, and boards should be thinking about AI investment decisions over the next twelve to twenty-four months. The questions that matter shift from feature comparison and vendor selection toward substrate diagnosis.
What inherited fragmentation does our current operating environment contain? Which prior-wave customizations are now operating as undocumented business logic that AI agents will inherit when deployed? Where do shadow workflows coexist with official systems, and what happens when an autonomous agent encounters the inconsistency between them? Which incentive structures will continue to reward transactional behavior even after the AI is deployed to support relational outcomes? How much accumulated coordination debt is the AI being asked to coordinate across, and is that debt visible enough that the deployment team can surface and address it before deployment rather than inheriting it?
These questions do not appear in vendor-led AI procurement sales cycles. They do not appear in most internal AI readiness assessments either, because most internal assessments focus on the visible orchestration layer rather than on the substrate beneath it. The questions appear when the deployment is six months in and the compounding effects of substrate inconsistency become operationally undeniable, by which point the investment has been committed and the remediation cost is materially higher than the diagnostic cost would have been.
That timing asymmetry is the structural reason pre-commitment (Phase 0™) diagnostic work matters. The substrate questions need to be asked before the commitment, not after.
What the Procurement-Technology Conversation Is Now Ready to Name
The most significant thing about Xavier’s post and the comment thread it generated is not the observations themselves. It is the readiness of the procurement-technology analyst community to engage with substrate-level questions publicly. A year ago, the public conversation would have been about agent capability, orchestration architecture, and vendor positioning. The structural questions would have been treated as too abstract for the operational discussion.
That has shifted. The community is now articulating the substrate problem in vendor-neutral language and discussing it as a first-order operational reality rather than as a theoretical concern. Xavier’s question — can agents operate reliably across inconsistent enterprise environments — would not have surfaced publicly twelve months ago. It is surfacing now because the experience of deploying AI into actual enterprise environments has made the substrate problem operationally visible to practitioners who previously did not need to engage with it.
That readiness creates an opportunity for the substrate conversation to mature from observation into diagnostic. Observation tells us the problem exists. Diagnostic tells us what to do about it. The frameworks I have been documenting in this archive since 1998 — Implementation Physics™, the Compounding Technology Shadow Wave™, Hansen Strand Commonality™, and the Phase 0™ Diagnostic — exist to perform that diagnostic work. The eighteen-year contemporaneous archive provides the longitudinal evidence base. The dual-pyramid graphic provides the visual language. The substrate problem the procurement-technology community is now naming in different vocabularies is the problem these frameworks were built to diagnose.
The Bottom Line
Orchestration coordinates faster across substrate that was never normalized. Agents execute reliably against operational capabilities the underlying platform already supports, and they execute unreliably against operational capabilities the platform does not support no matter how sophisticated the agent itself becomes.
That is the structural reality the procurement-technology conversation is now ready to engage with directly. Xavier Olivera surfaced it. Tanya Seda, Fabrice Saporito, Jason Busch, and David McClintock extended it from different analytical directions. The Coupa Inspire conversations made it operationally concrete. The broader 2026 consulting-firm convergence — Kearney, McKinsey, Bain, KPMG, BCG, Accenture, IBM, Hackett — is documenting it across the analytical landscape.
The orchestration layer is visible. The substrate underneath it is load-bearing. That is where AI investments will succeed or fail over the next eighteen to thirty-six months, and the diagnostic question that determines which trajectory an organization is on is not a question about the AI. It is a question about the conditions the AI is being deployed into.
The reports describe the gap. The pyramids name the cause. Phase 0™ is the diagnostic that runs before the next major commitment is made.
This post was developed through the ARA™ RAM 2025™ multimodel validation framework. Five independent models reviewed the Hackett observation, the Coupa Inspire thread, and the substrate diagnosis prior to publication. All five converged on the same structural diagnosis. The internal convergence mirrors the external convergence now documented across Hackett, Kearney, McKinsey, Bain, KPMG, BCG, Accenture, and IBM in their 2026 publications. The convergence itself is becoming evidence of the structural pattern.
The Compounding Technology Shadow Wave™ trilogy executive summaries are available at procureinsights.com. The Phase 0™ Diagnostic — for organizations preparing to commit further AI or orchestration investment — is at hansenprocurement.com/where-does-your-organization-sit-right-now/.
Hansen Models™ · Implementation Physics™ · Compounding Technology Shadow Wave™ · Phase 0™ · Hansen Fit Score™ · Hansen Strand Commonality™
-30-
Share this:
Related