After speaking with operators and investors who attended Ai4 in Las Vegas, I came away thinking the enterprise AI conversation is becoming less technical and more operational.
That is a good sign.
Two years ago, a lot of enterprise attention went to model choice, copilots and proving that generative AI could do useful work. Those questions still matter, but they are no longer enough. The harder question is whether the company itself is ready to change how work gets done.
Technical readiness only gets you so far
A company can have cloud infrastructure, a data lake, an approved model provider and a security framework and still be unready to scale AI.
The missing pieces are often organizational.
Who owns the workflow? Who can change it? Which decisions can the system make? Which decisions require a human? What happens when the model is uncertain? How is the output recorded? Who monitors performance? What metric tells the team whether the process actually improved?
Those questions sit across product, operations, IT, legal, security and finance. That is why AI programs get stuck even when the technology works.
The unit of transformation is the workflow
I think companies make AI harder by treating it as a horizontal capability before they understand the vertical job.
An enterprise does not experience AI as a model. It experiences AI inside claims processing, underwriting, customer support, software development, forecasting, recruiting, scheduling, procurement or sales.
The workflow is where context, permissions, data, exceptions and incentives meet.
That is also where domain knowledge starts to matter. A good system has to know what “done” means in that job, what can go wrong, what the regulator cares about, what the customer expects and when a human needs to intervene.
The operating model has to catch up
Deloitte’s 2026 survey found that enterprise AI access was expanding quickly while deep business transformation remained much less common. That gap makes sense to me.
Giving employees access to an AI tool is easy. Redesigning a process around it is not.
The second requires job changes, new controls, new metrics and sometimes a different org structure. It may also expose uncomfortable truths: a process exists because of historical handoffs, a team is measured on the wrong output, or nobody actually owns the customer outcome end to end.
Governance works better when it follows risk
I also think companies need to get away from the idea that every AI use case deserves the same governance process.
Generating a first draft of an internal note is not the same risk as making a credit decision, changing a clinical workflow or taking an autonomous action in a production system.
If the controls are too heavy, low-risk use cases never scale. If the controls are too loose, high-risk use cases create obvious problems.
The practical answer is tiered governance: define what the system can do, what data it can touch, how it is evaluated, when a human must review it and what evidence is retained.
The ROI question should come earlier
One of the reasons companies end up with pilot sprawl is that ROI is treated as something to measure after deployment.
I would reverse that.
Before a pilot starts, establish the baseline. How long does the process take today? How many people touch it? What does an error cost? What is the current conversion, throughput or service level? What would have to improve for the project to matter?
That does not mean every experiment needs a CFO-approved business case. It means the team should know what evidence would justify scaling.
The implication for investors
For investors, AI readiness is becoming part of company quality.
I would not give a company much credit for saying it is “AI-first.” I would want to understand whether it has proprietary context, whether the workflow gets better with use, whether implementation is repeatable, whether customers see measurable value and whether the company can defend the role it owns if model capability continues to commoditize.
The same applies to incumbents. The companies with strong distribution, systems of record and trusted customer relationships have a real advantage if they can move fast enough. Startups need to own something more specific than access to a model.
The AI winners inside the enterprise will probably be the companies – and the operating teams – that understand the job better than everyone else.
Sources and notes
Deloitte, State of AI in the Enterprise 2026: https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html – Enterprise AI access, scaling and business transformation.
Gartner, Why 50% of GenAI Projects Fail – And How to Beat the Odds: https://www.gartner.com/en/articles/genai-project-failure – Data, cost, risk and business-value failure points.
Dun & Bradstreet, AI Momentum Survey: https://www.dnb.com/en-us/newsroom/press-releases/dnb-survey-finds-enterprise-ai-returns-continue-to-advance.html – July 2026 data readiness and ROI findings.
Related reading: Why So Many Enterprise AI Pilots Stall Before Production | The Harness Is Not the Moat