Saving lives and creating experiences with Customer support agents at Ascendo.AI; Investor; Speaker on Enterprise AI and Entrepreneurship.

getty

The biggest mistake in enterprise AI has never been a lack of data. It has been the belief that more data automatically creates more intelligence.

That was true in 2018, when I wrote that the fear of not having enough data can stall an enterprise’s digital strategy. It is even more true now, in 2026, as the industry shifts from talking about data volume to talking about tokens. The language has changed, but the underlying problem has not.

Enterprises today sit on trillions of tokens across CRM systems, databases, documents, emails, tickets, Slack threads and wikis. But AI agents do not need all of it. They need the right tokens—the small, precise slice of context that allows them to act quickly, accurately and safely.

That is the real evolution of enterprise AI; not bigger input, but better selection.

The Problem Was Never Scarcity

In 2018, the industry was obsessed with accumulation—collect more data, add more sensors, buy more tools and store more history. The assumption was that intelligence would emerge from scale.

My argument was different. Enterprises should first learn how to extract value from the data they already have. Too often, companies treat data collection as the first step when it should be the last resort.

That insight matters even more now. In the age of LLMs, feeding an agent everything is not a sign of sophistication. It is usually a sign of poor design. The raw fire hose creates noise, raises cost and increases the chance of error. More information is not always better. Sometimes it is just more confusing.

Why Tokens Changed The Conversation

Tokens have made something visible that enterprise software has struggled with for years: Context is finite. A model does not need every document in the company to answer one question. It needs the exact data point, the most relevant passage or the specific ticket that matters right now.

That is why retrieval is becoming central to enterprise AI. It is not enough to store information. You have to know how to find the right piece, at the right time and for the right task. That is where structured search, semantic search and exact lookup are each playing an increasing role.

Text-to-SQL pulls precise facts from structured systems. Vector search finds the most relevant meaning in unstructured content. Reverse indexing locates exact identifiers, error codes and ticket numbers instantly. Together, these retrieval layers can help do something far more valuable than simply reducing token usage. They turn enterprise knowledge into usable context.

The 2018 Insight Was The Blueprint

The most important point in my original Forbes article was this: Enterprises often assume they need more data when what they really need is a better way to use what they already have. That principle now sits at the heart of agentic AI.

If a technician can only handle a few incidents in a day, the system should not optimize for total coverage across every possible issue. It should prioritize the cases that matter most. If a support agent needs to resolve a customer problem, the system should not dump the entire knowledge base into the prompt. It should surface the one answer that is relevant, trustworthy and actionable. That is not just a technical improvement. It is an operating model.

My advice to others is what our direction of thinking at Ascendo AI has always been. The goal should never be to chase data for its own sake. Instead, it should be to build systems that can identify what matters, filter what does not and drive action from the best available context.

The New Enterprise Advantage

The next wave of AI will not reward the companies that collect the most information. It will reward the companies that can narrow it best. That means knowing how to retrieve the right context, enforce permissions, reduce noise and take action with confidence. It means understanding that trillions of tokens are not the asset. The asset is the small, relevant subset that drives the next decision.

In 2018, I wrote that starting with existing data could reveal blind spots, inefficiencies and untapped value. That was true then, and it is unmistakably true now.

The future of enterprise AI will not belong to those who feed models everything. It will belong to those who know exactly what to feed them. And in a world obsessed with tokens, that may be the most important advantage of all.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?