Akhil Verghese on Why the Next AI Shift Is Leaning Toward AI-Native

The conversation around AI has moved quickly from curiosity to implementation. A few years ago, business leaders were asking whether AI had a place inside their companies. Today, many are asking a more important question: why are so many AI investments still failing to change how the business actually runs?

For Akhil Verghese, CEO of Krazimo, the problem isn’t that companies aren’t using AI enough, but that many haven’t done the foundational work that transforms the technology from being a useful tool into a superpower. 

Employees today may ask chatbots for help drafting emails, summarizing notes, cleaning up spreadsheets, writing code, and speeding through repetitive tasks. In many cases, those applications help, but many people believe that they could still be more efficient. Employees may have to explain context that the agent should already have or fail to provide it at all, which can potentially result in tasks being performed to less-than-satisfactory expectations.

While those are some examples of AI-assisted work, Verghese is focused on something more ambitious: AI-native companies.

What it means to be AI-native

Verghese says that an AI-native company isn’t defined by how many subscriptions it buys or how many pilot projects it launches. It’s defined by whether AI can access the company’s tools, data, workflows, permissions, and processes in a way that lets it answer, act, monitor, and improve safely inside the business itself.

“These are the kind of interactions I have with our AI-native assistant on a daily basis,” Verghese shares, describing examples such as tracing revenue received in the company bank account back to specific marketing channels, summarizing every touchpoint on a project across email, Slack, and meeting notes, or identifying prospects in line with the business’s ICP and drafting outreach for approval. “They wouldn’t have been possible with an AI-assisted dashboard or a RAG system on top of a sales database. They can only happen when every interaction, datasource, tool and process is accessible to your intelligence layer.”

The difference between AI-assisted and AI-native companies

To explain the difference, Verghese separates companies into stages. Some still aren’t using AI in any meaningful way. Others are “tab switchers,” with employees using AI tools on the side while the real work remains untouched. A more advanced group builds individual AI workflows or agents, but those systems often remain a patchwork of point solutions.

In AI-native companies, AI isn’t an accessory. It is integrated into every employee’s daily work.

The unglamorous work behind AI transformation

AI’s larger promise is a shared operational language for companies. Verghese describes going AI-native as the first step that made Krazimo’s 10-person team feel as if it had “15 extra employees that work around the clock,” but he also makes clear that the path to that kind of leverage starts with unglamorous work.

“You can’t automate what you haven’t understood,” Verghese says.

Before a company can build an intelligence layer, it has to map itself. That means understanding which tools each function uses, which data employees rely on, where money flows, and which tasks happen every day.

For Krazimo, that is where many AI projects either succeed or just completely fail. A company may want an AI agent that can update a CRM, draft customer messages, analyze spend, or monitor account health. But before the agent can do any of that reliably, the company has to answer more basic questions:

• What information can it access? 
• What can it change? 
• Which actions require approval? 
• What happens when the data is incomplete? 
• Who owns the result when the system goes live?

The answers determine whether AI becomes a dependable part of the company or just another impressive demo that never survives in real-world conditions.

Why governance comes first

Verghese says that governance is central to his vision. In an AI-native company, access can’t depend on a model “behaving” and has to instead be built into deterministic systems. A system shouldn’t be able to send an outward-facing message, approve a spend, or act on sensitive information unless the business has already defined the particular circumstances under which it can.

“The intelligence layer is only as trustworthy as the boundaries you draw,” Verghese says.

When AI stops waiting to be asked

Once the structure is in place, the first phase is reactive. Employees can ask questions and get answers grounded in company data. An assistant can explain why revenue dipped, summarize an account, identify which marketing channel is worth the spend, or retrieve the status of a project.

But Verghese believes the larger transformation happens when AI becomes proactive.

Instead of waiting for employees to ask how the pipeline is doing, a proactive agent may surface that three deals have gone quiet and prepare follow-up messages for review. Instead of waiting for a marketing manager to check performance, it may notice that one campaign is producing a higher return and suggest reallocating spend.

“That was definitely the biggest surprise to me in going AI-native,” Verghese says. “At some point, you stop interrogating your AI, and your AI starts briefing you.”

The human still owns the judgment

Even though AI is taking on more, it doesn't mean humans should hand over control. In Verghese’s model, agents can operate within limits, but large, outward, expensive, or irreversible actions still require human approval. The agent proposes; the human reviews, edits, approves, or rejects; and the system brings reasoning and evidence with the recommendation.

Verghese isn’t arguing for companies to outsource their judgment to AI. In fact, he warns against it. 

“Human understanding and responsibility remain essential, which is why monitoring becomes the final loop,” he says.

A company has to know whether its AI answers are correct, whether its agents are taking the right actions, and whether the automation is actually producing business value. A lead that was followed up on isn’t the same as one that was closed, nor is a deflected support ticket a resolved problem. If businesses don’t begin by creating a measurable definition of success that they can anchor to their pre-AI starting point, they may never know if their AI is just doing work, using tokens and not impacting the bottom line.

“The metric has to match the outcome you care about,” Verghese adds.

For Krazimo, this is where AI-native work becomes an operating philosophy. The company is building for a world where AI isn’t a side tool employees consult, but a structured layer that helps organizations improve how they work and learn.

In Verghese’s view, that’s what makes the AI-native shift so significant. It’s not a project a company finished, but a loop it keeps running.

And for Krazimo, it’s the future of enterprise AI.