When AI Isn’t the Hard Part

What Leading Organizations Are Learning About AI in Practice

AI capability is no longer the constraint. The constraint is the enterprise itself: how it thinks, how it operates, and how fast it can reimagine both. What follows are eight hard truths emerging from the front lines of that shift. They build toward a single conclusion: the organizations pulling ahead are not deploying more AI. They are reorganizing around what AI makes possible.

1. The Constraint Has Shifted from Technology to Imagination

• "The assumption that we can take existing processes and use AI to make them better is a path to sub-optimization."
• "There's an unwritten constraint, which is the imagination of people - and that's going to hold us back."
• "Good leaders used to define tight constraints and execute. Now that's not ambitious enough."

Enterprises are still approaching AI as an optimization layer on top of existing workflows. That approach is already showing diminishing returns. The real opportunity lies in reimagining work from first principles - removing steps, collapsing roles, and redesigning outcomes entirely.

The limiting factor is no longer access to models or tools. It is the inability to break free from inherited ways of working. Deep domain expertise, while valuable, often anchors thinking in legacy constructs rather than enabling reinvention.

Questions:

• Where are we optimizing processes that should no longer exist?
• What would we build if we started from the outcome and worked backward, with no inherited constraints?

2. Value Is Real - but Narrow, Uneven, and Hard to Scale

• "It's easy to build AI applications. It's very hard to make them enterprise-grade."
• "Technology has advanced faster than enterprise adoption and playbooks."
• "The ROI disappears quickly if the business isn't aligned from day one."

Clear value is emerging - but only in domains where work is digital, measurable, and tightly scoped. Software development, customer service, underwriting, and sales are showing real productivity and quality gains.

But scaling remains elusive. Evaluation frameworks, guardrails, integration complexity, and organizational alignment are the true bottlenecks. Most enterprises are still operating in trial-and-error mode without repeatable playbooks.

Questions:

• Where are we seeing real, repeatable ROI vs. isolated wins?
• What specifically is preventing our highest-performing use cases from scaling - and is the blocker technical, organizational, or both?

3. The Operating Model Tension Is Unresolved - and Structural

• "Letting 1,000 flowers bloom doesn't move the needle - but shutting it down kills the culture."
• "People closest to customer problems come up with better solutions than centralized teams."
• "We basically said no to everything else - and that took real commitment."

Enterprises are caught between two competing models. Federated innovation drives high engagement and fast learning, but tends to produce incremental impact. Centralized focus is scalable and efficient, but risks slowing momentum and suppressing initiative.

A hybrid model is emerging, where central teams provide platforms, standards, and reusable components, while distributed teams drive domain-specific innovation. But this balance is difficult to sustain and highly sensitive to culture. And the tension is not a bug to be fixed. It is a permanent condition to be managed.

Questions:

• Are we over-indexed on control or fragmentation?
• What belongs centrally vs. locally - and how explicitly have we defined that boundary?

4. Infrastructure, Cost, and Architecture Are Now Strategic Constraints

• "We can't just keep buying GPUs and throwing them at the problem."
• "Day-two costs are very high - and they're eating into returns."
• "Versioning is becoming a real problem - you have to re-engineer everything every time the model changes."

The next phase of AI is being shaped by economics and operational friction, not capability. Compute and token costs are rising faster than realized value. Model versioning is introducing ongoing rework. And infrastructure decisions are becoming long-term strategic constraints that are difficult to reverse.

This is driving a shift toward smaller models, architectural optionality, and continuous optimization across accuracy, latency, and cost. At the same time, cloud, data readiness, and governance are no longer optional - they are prerequisites for participation.

Questions:

• What is our true "day-two cost" of AI - and how is it trending?
• Are we architected for flexibility, or are we accumulating hidden lock-in?

5. The Interface Layer Is Being Rewritten

• "I'd rather have two systems than zero, don't let perfect be the enemy of good."
• "It's seductive to build one platform centrally - but it's wrought with failure and disappointment."
• "The front end becomes a conversation - you don't go into systems anymore, you just ask."

The way users interact with enterprise systems is shifting fundamentally. Workflows are giving way to conversational and agent-driven interfaces, with AI acting as an orchestration layer across underlying systems.

ERP and core platforms remain - but increasingly as infrastructure beneath this new experience layer. The traditional boundaries between build, buy, and outsource are blurring as SaaS is selectively unbundled and AI fills the gaps.

The strategic question is no longer which system to choose. It is who controls the layer that sits on top.

Questions:

• Where are we over-committed to monolithic systems vs. modular layers?
• What does our future "control plane" look like - and who owns it?

6. The Economics of the Industry Are Breaking the Old Model

• "The same business can go from 50% margins to 90% margins with AI."
• "You're getting arbitrage twice with third-party models - it just doesn't make sense anymore."
• "If I'm spending more on tokens, something else has to give."

AI is fundamentally reshaping industry economics. Services models are compressing. AI-native firms are operating at structurally different cost bases. And traditional outsourcing and labor arbitrage models are under pressure.

At the same time, rising AI infrastructure costs are forcing trade-offs in IT spend, accelerating shifts across the stack. The uncomfortable implication: some of today's cost structures are simply not viable in a market where competitors operate at AI-native economics.

Questions:

• Where are we exposed to margin compression from AI-native competitors?
• What existing spend is being displaced - intentionally or implicitly?

7. Talent Is the Primary Bottleneck - and the Model Is Changing

• "You can't buy your way into engineering talent for AI transformation."
• "Just because someone was a great engineer before doesn't make them effective in AI."
• "The skill set for human-in-the-loop is completely different - and very hard to find."

The talent challenge is not just scarcity - it is structural mismatch. AI-native engineering requires different skills than traditional software development. Data science, engineering, and operations are converging. And new roles are emerging that combine technical depth, domain understanding, and change leadership.

At the same time, work itself is being redefined. Agents handle execution. Humans provide judgment. Roles become fluid rather than fixed. This is where the organizational challenge (Section 8) and the talent challenge converge. You cannot staff a new operating model with the old job descriptions.

Questions:

• Do we have the right talent mix - or are we hiring for the organization we were, not the one we need to become?
• Which capabilities must be built internally vs. accessed externally?

8. The Real Transformation Is Organizational, Not Technical

• "Technology has advanced faster than enterprise adoption and playbooks."
• "If you wait, you lose the muscle - you won't be ready when the time comes."
• "We're not structured for this level of continuous change."

The primary barrier to AI impact is not technical - it is organizational. This is the thread that runs through every theme above: the imagination gap (Section 1), the scaling bottleneck (Section 2), the operating model tension (Section 3), and the talent mismatch (Section 7) are all, at root, organizational problems.

Enterprises are not designed for continuous reimagination. Institutional knowledge risks being lost as processes are automated. And change management remains the hardest problem. The companies pulling ahead are those that align business ownership early, redesign processes end-to-end, and build systems that adapt continuously.

Questions:

• Are we structured to evolve continuously - or episodically?
• If we solved every technical challenge tomorrow, what organizational barriers would still remain?

What This Adds Up To

AI is not struggling in the enterprise. The enterprise is struggling to adapt to what AI makes possible.

The eight themes above point to a single, uncomfortable conclusion: the gap between what AI can do and what organizations are able to absorb is widening, not closing. Technology will continue to advance. The question is whether enterprise operating models, talent strategies, and leadership mindsets can evolve at anything close to the same pace.

The leaders pulling ahead are not those deploying the most models. They are the ones rethinking how work, systems, and decisions come together at a fundamental level - and doing so continuously, not as a one-time transformation.

Synthesized from discussions across the Executive Technology Board, a technology think tank representing 250 Fortune 1000 enterprises worldwide.

Executive Technology Board (c)