Shreyas Nair is an AI venture founder and founder of Wordsworth AI.
I recently spoke with a friend at a large company who had just finished another AI pilot review. The pilot went well. The tool summarized documents, drafted follow-ups, extracted action items and made the old workflow seem painfully slow by comparison.
However, the mood in the room shifted when the team spoke about a full-blown production rollout. Data was scattered across too many places; no one was sure which system was the real source of truth; legal wanted an approval step; IT wanted access controls; the business team wanted things to move faster; and the people who knew about the exceptions weren’t even in the room.
To me, this is the real story of enterprise AI right now.
Headlines suggest that AI adoption is already a done deal. Accenture is rolling out Microsoft 365 Copilot to about 743,000 employees after a big internal pilot. General Mills has used AI to analyze thousands of daily shipments and reportedly saved over $20 million in transportation costs. Gartner, Inc. predicted that agentic AI would appear in one-third of enterprise software and services by 2028. On paper, AI is now mainstream in business.
Inside companies, things are much messier. The demos work well and the pilots are real. Employees are actually using these tools, and that matters. However, just using AI isn’t the same as real transformation. A company might have thousands of people summarizing meetings, drafting emails and writing code with AI but still have the same broken workflows underneath.
This is where I think the conversation should go. Enterprises don’t have a model problem anymore; they have a readiness problem. Most leadership teams are still asking which AI tool they should buy. That’s the wrong question. The better question is whether their company is actually ready to use the AI tool.
Once AI goes beyond just writing and summarizing and starts taking action, a company’s weaknesses become clear quickly. The policy exception is buried in an email. The customer-specific rule is in someone’s head. The customer relationship manager (CRM) says one thing, but the contract repository says another. One person in finance manages the spreadsheet that really runs the business. AI fails because the company is hard to understand.
I’ve seen this most clearly in proposal and sales workflows. An AI system can read a 200-page RFP and pull out requirements in minutes. That’s helpful, but the real work comes after that. Which past projects are approved for reuse? Which executive relationships matter? Which claims has legal rejected before? Which resumes are up to date? Which win themes worked in a similar bid, and which ones sounded good but never convinced the buyer?
That knowledge is usually scattered across old slide decks, folders, Slack threads, emails and in the memories of people who have been around long enough to know all of the details.
This is why so many AI projects quietly shrink after the pilot phase. They start as big transformation efforts but end up as assistant tools. They begin as autonomous agents but turn into “human-in-the-loop” dashboards. The CEO might say that the company needs to become AI-first, but six months later, it has 20 disconnected pilots and no clear operating model.
Now, the focus is shifting to systems that can actually take action across workflows with guardrails, escalation paths and audit trails. That difference is important. Getting answers is easy now, but getting things done is still hard, and execution is a company design issue.
The companies that do this well are changing the systems around the AI tool. They clean up data, assign ownership to each workflow, grant agents permissions gradually and measure whether AI actually improves metrics such as cycle time, error rates, costs, resolution time or revenue per employee.
That last point is important because AI usage can quickly turn into a vanity metric. A company might proudly say that 60% of employees use AI every week, but what’s actually changed? Did customers get answers faster? Did engineering throughput improve? Did finance close the books sooner? Did legal review contracts with fewer mistakes? Did sales forecasts get more accurate? If the answer isn’t clear, the company might not have an AI strategy. It just has AI activity.
This is the risk I worry about most—not that enterprise AI fails but that it only half-succeeds. It saves time at the edges, creates enough excitement to get more budget and gives leaders something to talk about in all-hands meetings, but it never reaches the core of how the business actually runs. Three years later, the company has more tools, copilots and dashboards but not much more real leverage.
Now, the boring work becomes the important work. Where does the relevant data live? Who owns it? Which systems are trusted? What can an AI agent do without approval? When should it escalate? What does the audit trail look like? Who’s responsible when something goes wrong? These aren’t glamorous questions, but they separate AI theater from real AI transformation.
The next big divide in business will be between companies that treat AI as just a productivity tool and those that can transform their businesses in ways the world will be surprised to see. This is the first in a series of essays I’m planning on writing about enterprise AI adoption—what’s working, what’s breaking and what leaders need to know as AI moves from experiments to the core of the business.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
