AI Meets Enterprise Reality

IIT2026 brought together the global alumni network in Long Beach for three days of conversations across technology, entrepreneurship, investment, sustainability, health, and global collaboration. The conference celebrated 75 years of IIT excellence and brought together more than 1500 leaders, founders, operators, investors, technologists, and changemakers from across industries. The event had the energy of a room where AI no longer needed a grand introduction. Everyone had seen enough pilots, tools, demos, and boardroom pressure to know the conversation had moved forward.

What stood out was the kind of concern underneath the excitement.

Healthcare leaders were not responding to AI as a shiny interface. They were looking at care workflows where trust, documentation, data movement, and accountability decide whether technology can hold up. Enterprise leaders were weighing custom agents against a crowded market of vertical tools. Technology teams were thinking about older systems where years of product logic sit buried in code, tickets, documents, and memory. Different entry points. Same undertone.

Enterprises are no longer asking whether AI belongs in the business. They are asking what it will take to make AI work without losing control over the systems, knowledge, and decisions that keep the business running.

It all kept circling the same concern: while AI is easy to buy, momentum is hard to build.

Healthcare As the First Stress Test

Healthcare has a way of cutting through loose AI language.

A prior authorization workflow is not the same as a generic productivity demo. It sits inside a chain of documentation, payer rules, patient access, compliance, cost, and clinical context. Value-based care brings another layer of complexity: outcomes, coordination, reimbursement, data quality, and trust.

Here, AI cannot simply sound impressive. It has to hold up.

People wanted to know how the demos were built, how the platform thinking came together, and how collaboration worked across stakeholders. The questions were more about the machinery underneath the experience:

Who owns the data? Where does the intelligence sit? How much of the workflow stays inside the enterprise? What happens when a vendor’s agent becomes the place where your operating knowledge accumulates?

Because in healthcare, the risk is visible.

The market is filling up with vertical AI agents. Some focus on billing. Some on claims. Some on documentation. Some on support. Many will solve narrow problems well. The problem starts when the enterprise slowly loses control of the intelligence layer.

A rented AI tool can make one workflow faster. It can also become the place where data, workflow logic, exceptions, and institutional learning move outside the business. Over time, the company may own the problem while someone else owns the operating memory.

One of the stronger conversations Coditas had at the event:** Enterprise AI is not only a capability question. It is an ownership question.**

The AI Journey Needs a Sequence

Over the last two years, many AI programs have started with spending.

A board asks for AI movement. A CXO approves tools. Teams get access to copilots, cloud AI services, chatbot builders, or coding assistants. Everyone expects productivity to show up because the organization has finally paid for intelligence. That’s rarely how enterprise change works.

McKinsey’s 2025 State of AI survey found that most organizations are still in experimentation or pilot mode. Nearly two-thirds had not started scaling AI across the enterprise, and only 39% reported enterprise-level EBIT impact from AI. At the same time, 62% were already experimenting with AI agents.

Deloitte’s 2026 enterprise AI research shows the same tension from another angle. Workforce access to AI grew by 50% in one year, from fewer than 40% to around 60% of workers with sanctioned AI tools. Still, fewer than 60% of workers with access use those tools in their daily workflow.

Access is rising faster than adoption. Adoption is rising faster than measurable value.

The conversations at IIT2026 reflected the gap. Leaders were no longer satisfied with “we have AI tools” as proof of progress. The better question was: which process deserves AI in the first place? That’s where the sequence becomes important.

First, digitize the process. If work is still happening through calls, spreadsheets, emails, and undocumented manual checks, the business has no reliable data trail. AI cannot learn much from work that leaves no usable trace.

Then connect the systems. Sales, finance, engineering, service, compliance, and operations often hold different pieces of the same truth. If those systems do not talk to each other, AI receives information in fragments. Fragments create weak automation.

Next, add intelligence.

AI performs best when the enterprise has already done some of the less glamorous work. Clean up the workflow. Capture the data. Connect the context. Give teams a reason to trust what the system sees. This may not sound like the shiny part of AI, but it is usually the part that decides whether AI survives beyond the first demo.

The Stack Beneath the Agent

A useful agent is rarely the beginning of the journey. It is usually the visible end of several decisions made earlier.

Coditas’ conversations at IIT2026 often came back to a five-layer view of enterprise AI. The words can sound simple: strategy, people, data, intelligence, platform. The work underneath each layer is anything but simple.

The strategy layer is where the business has to slow down before it speeds up. Which workflows should change? Which processes need digitization? Where is data disconnected? Which use cases have a real path to ROI?

The people layer is where adoption becomes honest. Teams do not change because a new tool appears. They change when the work around them changes, when incentives make sense, when they understand what AI will and will not do, and when leaders explain the new operating rhythm clearly.

The data layer is where the system earns trust. Poor data does more than produce poor answers. It makes users cautious. It makes governance harder. It makes automation risky.

The intelligence layer gives business meaning to the data. This is where semantics, context, rules, relationships, and domain logic start to matter. Raw information becomes more useful when the system understands how things relate.

The platform layer is where repeatability enters. One agent can be built as a project. Ten agents need a platform that can support observability, maintenance, reuse, controls, and future change.

Enterprises should not have to reinvent the wheel every time a new AI use case appears. Once the lower layers are healthier, the next agent should become easier to build, govern, and improve.

THAT is what momentum looks like in practice.

Modernization is Essentially a Knowledge Reconstruction

Modernization came up in a different way at IIT2026.

Some people were less interested in AI agents as a standalone story. They were looking at older systems and asking what happens when the software beneath the business is too old, too undocumented, or too difficult to change.

This is where modernization becomes more interesting than technical debt reduction.

A legacy system is rarely just old code. It holds product memory. It holds workflows. It holds exceptions added for one customer years ago. It holds business rules that nobody fully remembers. It holds decisions made by developers who may have left the company a decade ago.

A traditional migration often moves a function from one technology to another. COBOL to Java. Old Java to new Java. .NET to a newer stack. May be necessary, but not always enough. The deeper opportunity is to recover what the system actually does before rebuilding it.

It means reconstructing the real specification of the product, asking what should remain, what should be simplified, and what should quietly disappear because its reason for existence no longer applies.

If a system has accumulated 100 features over 10 years, the question is not how to move all 100 into a modern stack, but if we were building this product today, knowing what we know now, what would we build?

AI can help recover trapped knowledge from legacy systems. Engineering teams can validate the recovered intent. Product teams can refine it. Business stakeholders can spot redundant workflows. Human review keeps judgment in the room. Hence, modernization starts becoming AI-ready. A company does not get a cleaner version of the same old system. It gets a chance to rebuild with context.

For enterprises looking at AI, this matters more than it may seem. Agents struggle when the systems around them are opaque. Modernization gives AI better surfaces to work with, better context to read, and cleaner paths to action.

ROI is Becoming the Only Serious AI Language

The investment track at IIT2026 focused on capital, company building, valuation, fundraising, and scale. The theme fits the broader market well. OECD data published in February 2026 found that AI firms captured 61% of global VC investment in 2025, or USD 258.7 billion out of USD 427.1 billion. AI infrastructure and hosting attracted USD 109.3 billion in 2025 alone.

Money is not the scarce part anymore. Judgment, definitely.

Every board wants to see AI movement. Every leadership team wants to avoid falling behind. Every function can find a reason to ask for tools. The harder discipline is deciding which AI bets deserve funding because they have a credible path to return.

BCG’s 2026 research found that only 5% of companies generate measurable AI value at scale, while 60% are not achieving material value. Yet within the technology function, the value of AI is rising. The share of companies scaling or fully deploying AI in one or more top tech use cases tripled from 9% in 2024 to 28% in 2025.

The use cases are telling: Software development. Data management. Compliance monitoring. IT project management. Service desk automation. Legacy modernization.

That’s often where ROI begins. A release cycle shortens. A report stops taking three days. Support teams find the right answer faster. A compliance check becomes easier to trace. A modernization team understands old code before rewriting it. A workflow drops steps that nobody needed anymore.

The first valuable AI use case may not be the one that looks best in a demo, but can remove friction from work the business already knows is slowing it down.

A demo earns attention. ROI earns permission to scale.

Governance Belongs in the Build

Governance tends to enter AI conversations late, usually after enthusiasm has already done its damage. But it won’t work for long.

Once AI starts retrieving information, supporting decisions, calling tools, or moving work forward, governance becomes part of the build. It belongs in access design, logging, monitoring, escalation, model evaluation, data handling, and release practices.

NIST’s AI Risk Management Framework was created to help organizations manage AI risks and incorporate trustworthiness into the design, development, use, and evaluation of AI systems.

If an AI system in an enterprise touches a workflow that carries cost, customer impact, regulatory exposure, or operational risk, the business needs to know what happened, why it happened, and who can intervene. That requires audit trails, access rules, human review, monitoring, clear escalation paths, incident response, and most importantly, ownership.

None of these is glamorous. All of them matter. Good governance does not slow AI down. It keeps AI from breaking trust faster than the business can repair it.

Why the Room Still Matters

There is a reason a conference like IIT2026 matters even when most information is available online.

A room changes the question.

A business leader describes a workflow problem. An engineer hears an integration issue. A healthcare leader sees risk. An investor asks whether the motion can repeat. A modernization expert hears trapped knowledge inside legacy systems.

The same problem becomes clearer when different people turn it around from different sides — the value of IIT2026’s cross-industry setting. It brought together people who were not all solving the same problem, yet many were dealing with the same underlying constraints: data, adoption, cost, ownership, trust, and scale.

AI execution does not move through tools alone. It moves through conversations that sharpen judgment. It moves when the right problem meets someone who has seen a version of it before.

What Leaders Should Carry Forward

The next “AI-driven” conversation should begin before the tool is chosen.

1. Start with the business friction. Where is the organization losing time, money, quality, trust, or revenue? It could be claims, reporting, engineering, customer support, sales operations, compliance, onboarding, approvals, documentation, or modernization.

2. Revisit the workflow. If a process has 10 steps, do all 10 still deserve to exist? AI should not speed up a bad workflow before someone asks why the workflow looks that way.

3. Check whether the process is digitized. Manual work leaves weak evidence behind. Without a reliable data trail, AI has little to learn from.

4. Look at connection. Does the process need context from finance, sales, engineering, operations, service, or compliance? If those systems remain separate, AI will keep seeing pieces rather than the whole.

5. Examine the data and knowledge layer. Which data can be trusted? Which rules are documented? Which decisions live only in someone’s memory? Which product behaviors are trapped in code, tickets, and old spreadsheets?

6. Look at modernization. Which systems can support AI as they are? Which should be wrapped, rebuilt, simplified, or retired? Which parts of the old system carry logic the business cannot afford to lose?

7. Define governance. What needs approval? What needs an audit trail? What needs human review? What should never be automated? Who owns the outcome when AI acts?

8. Plan adoption. How does work change for the people using the system? What will make them trust it? What will make them avoid it? What needs to change in roles, handoffs, reviews, and incentives?

9. Measure ROI in business terms. Speed. Cost. Quality. Risk. Revenue capture. Throughput. Margin. User satisfaction. A serious AI program should know what it expects to prove after 30, 60, and 90 days.

From Attention to Momentum

Answering the right questions will help create better AI decisions. Why? Because they compel the enterprise to look beneath the surface.

The entire outlook is grounded in execution at Coditas. Enterprise AI creates value when strategy, data, knowledge, platform engineering, modernization, governance, and adoption move together. Progress slows when those decisions happen in isolation.

The conversation at IIT2026 was full of energy, but energy alone does not create movement.

Momentum comes from clearer choices. Stronger foundations. Better systems. Measurable value. People who know how to make technology hold up under real operating conditions.

AI already has the spotlight. Momentum will belong to enterprises that can make it work when the lights are off.

Ready to move from AI attention to AI momentum?

Coditas helps enterprises build AI strategies, data foundations, knowledge systems, modern platforms, and adoption models that can stand up to real operating conditions.

Let’s build what works beyond the pilot room. Talk to us.