The transplantable skeleton: Why agentic AI infrastructure must survive corporate surgery

Gartner predicts that more than 40% of agentic AI projects will be canceled by 2027. The debate has centered on escalating costs, unclear business value and inadequate risk controls. But having led IT infrastructure through major divestitures and cross-border integrations over the past two decades, I see a more fundamental problem: most agentic infrastructure is built as if enterprise boundaries are permanent.

They are not. The average Fortune 500 company undergoes a significant merger, acquisition or divestiture every 3.5 years. Business units get carved out. Acquisitions get absorbed. Regulatory shifts force operational separation. Yet most agentic AI implementations would fail catastrophically the moment someone says, “We’re spinning off that division.”

This is the portability crisis hiding inside the agentic revolution.

If 2025 was the year of the brain — the LLM — then 2026 must be the year of the skeleton: the structural framework that determines what the organism can become. But we need a skeleton that can be transplanted. The infrastructure we build today must withstand the corporate surgery every enterprise inevitably undergoes.

The problem: Tight coupling creates transformation debt

When agentic systems are tightly coupled to a single enterprise’s identity infrastructure, data lakes and application programming interfaces (APIs), a divestiture becomes extraordinarily complex. AI agents cannot simply be copied to the new entity. Their entire operational context — the knowledge graphs they rely on, the observability systems that enable traceability, the metadata layers that provide business understanding — must be surgically separated.

I have witnessed this cost firsthand. Across major divestitures I have participated in throughout my career, disentangling infrastructure dependencies has consistently consumed 12 to 24 months and $40 million or more in unplanned spend. This budget could have funded multiple new digital initiatives. The technical debt was never in the AI models themselves. It was baked into the infrastructure assumptions behind every integration decision.

The problem compounds for global enterprises operating across regulatory jurisdictions. An agent designed for GDPR-compliant European operations may be architecturally incompatible with a newly acquired U.S. subsidiary operating under different data residency requirements. The system that was supposed to unify intelligence across the enterprise becomes a liability that must be replicated, forked or abandoned.

This is the problem CIOs must solve: How do you build agentic infrastructure that delivers value today while remaining portable enough to survive the organizational changes every enterprise inevitably faces?

Designing for divisibility: The mindset shift

Building agentic infrastructure that survives enterprise transformation requires a fundamental shift in how architects think about system boundaries. Rather than optimizing for a single organizational context, every design decision must account for the possibility that components will need to operate independently or within entirely new structures. This is not about adding features — it is about embedding divisibility into the foundational assumptions of every integration.

Semantic clarity matters more than semantic unity

Traditional approaches to enterprise data architecture treat the organization as a single context with unified definitions. But enterprise vocabularies inevitably fragment during transformation. What “customer” means in one division may carry fundamentally different weight in another — and those differences must be preserved, not papered over.

The practical implication: build data architectures that maintain semantic coherence internally while supporting interoperability externally. When regulatory requirements demand carving out a business unit in 90 days, well-bounded semantics allows agents to continue operating in both the parent and the carved-out entity without the massive translation exercises that consume transformation budgets.

Ownership must be explicit from day one

Event-driven architecture for agentic systems is well established. What receives less attention is who owns the operational data that flows through these systems. When a divestiture occurs, which entity retains historical event streams? Which gets the derived insights?

Too many architectures defer these questions until crisis forces them. The organizations that navigate transformation smoothly are those that have already documented ownership at the system design phase. They build data flows where business attribution is intrinsic rather than inferred. When a business unit moves to a new organization, the question of what data belongs to whom has already been answered.

Metadata should coordinate, not centralize

Enterprise metadata implementations typically take one of two forms: distributed chaos, where every system maintains its own metadata with no coordination; or monolithic data lakes, where all metadata flows into a single repository. Neither survives transformation well. The chaos approach creates archaeological expeditions during carve-outs. The monolithic approach creates dependencies that cannot be cleanly severed.

The alternative is coordination without forced centralization. Each business domain maintains its own metadata stores following shared standards for discovery and exchange. During transformation, domains can be detached from enterprise-wide catalogs while maintaining internal coherence, then reattached to a new enterprise context with minimal disruption. The metadata architecture serves the business structure rather than constraining it.

Operational automation requires boundary awareness

Artificial intelligence for IT operations — what Gartner defines as platforms combining big data and machine learning to automate event correlation, anomaly detection and causality determination — represents one of the most promising applications of agentic AI. Autonomous systems that diagnose issues, initiate remediation and reduce mean time to repair (MTTR) offer significant operational value.

But these systems must understand their operational boundaries. An agent authorized to restart services in one domain should not automatically gain that authority across newly merged entities. Effective AI-driven operations depend not just on intelligent agents but on careful boundary definition that prevents unauthorized actions across organizational lines.

During transformation, boundaries must be reconfigurable without rebuilding the agents themselves. The intelligence stays constant; the operational scope adjusts. This requires externalizing authorization decisions from agent logic — treating permissions as configurable infrastructure rather than hardcoded behavior.

Protecting proprietary intelligence during transformation

Beyond operational automation lies a category that demands even more careful architectural consideration: the organization-specific datasets, proprietary business logic and custom-trained AI models that constitute a genuine competitive advantage. These strategic intelligence assets must never leak during transformation.

Portable infrastructure requires knowing exactly which data sources, training runs and business rules contributed to each agent’s behavior. The organizations that document these relationships as they build — rather than attempting to reconstruct them during crisis — find that transformation negotiations proceed smoothly because the question of what belongs to whom has already been answered.

Without this documentation discipline, divestiture becomes an archaeological expedition through undocumented decisions, consuming months of effort that should have been spent on integration. With it, there is a clear manifest of what belongs to whom, and deal timelines that actually hold.

Regulatory complexity demands architectural awareness

Coalfire’s 2026 outlook suggests regulatory fragmentation will accelerate as data sovereignty requirements evolve. GDPR, PIPL, HIPAA and the EU AI Act each carry different implications for how agentic systems can process data, make decisions and transfer information across borders.

Rather than treating regulatory compliance as a layer added after the fact, transformation-ready infrastructure builds jurisdictional awareness into its foundational design. When a carve-out requires separating data subject to different regulatory frameworks, the architecture should already understand those distinctions rather than requiring manual classification under deal pressure.

This is not about compliance checkboxes. It is about building infrastructure where regulatory requirements become visible properties of the system rather than assumptions buried in implementation details that surface only when a deal team asks uncomfortable questions.

The transformation readiness test

Before approving any significant agentic AI investment, CIOs should subject proposals to a simple but revealing test: Can this system operate under a different corporate identity? If a deal announcement were to come tomorrow, could this agent continue to function within an entity that does not inherit the parent company’s identity systems, security infrastructure or enterprise agreements? The answer exposes how much transformational debt the investment would create.

The related question: Is the boundary of the intellectual property clear? When business units separate, can the organization demonstrate which training data, business rules and model weights belong to each entity? Organizations that cannot answer this question cleanly will find that transformation negotiations become mired in disputes over AI asset ownership — disputes that can delay deals by months and cost millions in legal fees.

If the honest answer to either question is unfavorable, transformation debt is accumulating. That debt will come due at the worst possible time — when a deal timeline leaves no room for architectural rework and every week of delay threatens the transaction itself.

Building for the inevitable

Enterprise transformation is not a risk to be mitigated; it is a certainty to be designed for. The organizations that get this right will possess an agentic infrastructure that remains a competitive advantage regardless of how corporate boundaries evolve.

In a world where the only constant is transformation itself, the CIOs who build divisible infrastructure today will be the ones leading both sides of tomorrow’s deal table.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?