11 - 08 - 2026

AI is ready. but corporate world is still fumbling with the keys

The machines, it seems, are ready. The humans are not.

Artificial intelligence has reached that awkward stage where the technology is racing ahead while the companies expected to use it are still looking for the right files, the right rules and, occasionally, the right people.

That, says Mayank Verma, Global Head–Data and AI at Xebia, is the real problem with AI today. It is not a shortage of clever machines or smarter models. It is the unglamorous business of getting enterprises ready. “Technology is ready for the next revolution of AI. The question is whether we as humans and organisations are ready for that,” Verma said.

Going beyond test phase

There is no shortage of AI pilots. Every large company seems to have one. A chatbot here, a copilot there, an experiment with an AI agent somewhere else. The trouble begins when the experiment is asked to leave the laboratory and enter the real world.

“Most pilots are executed in a controlled environment,” Verma said. “But when they move into actual deployment, they have to deal with the varied nature of the real world, and that is where many of them fail.”

In the laboratory, everyone behaves. The data behaves. The systems behave. The users behave. Real life, unfortunately, has never been known for behaving.

Verma says the AI pilots that survive the journey into production have four things going for them: a clear outcome, someone clearly responsible for them, production-grade technology and governance, and an organisation willing to change the way it works.

The last bit may prove the hardest. For years companies have accumulated data in neat rows and columns, safely tucked away in databases and warehouses. But an AI agent needs more than numbers.

It needs the company’s policies, documents, meeting notes, audio, video and all the other bits of institutional memory that explain how things actually get done. “AI requires not only traditional data but also enterprise context,” Verma said.

He has a useful story to illustrate the point. Xebia once built an HR chatbot for a client. The machine was fed the company’s HR policies and performed beautifully. Then employees began asking questions.

The chatbot gave answers that were perfectly correct according to the policy documents. The employees nevertheless said they were wrong. The mystery was eventually solved. The policy documents did not contain all the information employees needed in practice.

“That was not an AI problem. It was a data problem,” Verma said. This is likely to become more important as companies move from copilots that merely assist humans to agents that can actually perform tasks.

Verma expects a mixture of both. High-risk and customer-facing decisions will continue to have humans looking over the machine’s shoulder. Routine and less risky jobs, on the other hand, are likely to be handed increasingly to autonomous agents.

Access to the right info

Xebia’s answer to this problem is Axis, its agentic data foundation. The idea is fairly simple: before asking an AI agent to run the business, give it access to the right information and a system in which it can operate safely.

Axis helps companies bring fragmented data together, move information from older systems to modern platforms and prepare enterprise data for AI agents. It can also help when things go wrong, which, as anyone who has worked with computers knows, is not an entirely hypothetical possibility.

An AI agent can investigate a failed data pipeline, identify the likely cause and recommend what should be done next. Verma says employees can spend about 80 per cent of their time finding out what went wrong and only 20 per cent fixing it.

“With Axis, the agent can take care of that 80 per cent, leaving the human to review the outcome and take the necessary action,” he said. The numbers become more interesting when one looks at the time saved.

In one Xebia project, a large volume of ETL code was migrated from a legacy database to a modern platform in three months. Had humans done the entire job, the estimate was 11 months.

There is another small matter that companies will have to learn to live with: the AI bill. Verma says enterprises should stop boasting about how many tokens they have consumed and start asking what they got in return.

Not every job requires the most expensive frontier model. Smaller and cheaper models can do many tasks, while the heavyweight models can be reserved for jobs that genuinely require them.

So the next AI revolution may not belong to the company with the cleverest model. It may belong to the company that has cleaned up its data, rewritten its policies, fixed its processes and persuaded its employees to let a machine do some of the work.

The machines are waiting. The question is whether the rest of us are ready.