Retrofitted AI creates hamster wheels. Native AI drives outcomes. Here’s why

Enterprise software is facing an AI crisis.

While one BCG report found that nearly nine in ten CEOs report seeing some cost or revenue benefits from AI in targeted areas, separate BCG research found that only 5% of companies have managed to generate transformative, bottom-line value at scale.

This gap points to a frustrating phenomenon: Organizations are aggressively adopting AI, yet too often their core operating expenses remain stubborn and their underlying systems are becoming more complex.

One explanation may not lie in the AI itself, but in the infrastructure underneath it. It’s what can happen when you try to attach a new feature to that existing infrastructure, rather than building the feature natively inside the platform.

Here’s why bolted-on AI may not move the needle — and what to look for instead.

The “hamster wheel” of AI-retrofitted software

Most enterprise software providers were built in an era of relational databases, manual form entries, and human-in-the-loop workflows. To keep pace with the AI boom, these legacy providers have taken a shortcut: layering conversational chatbots and predictive widgets onto their existing code bases.

But when you add a frontier AI model onto a legacy system that wasn’t designed for it, you can end up with a structural disconnect that limits what the AI can accomplish.

• The reinvestment gap: While AI tools save sellers an average of 4.8 hours per week, 72% of sales organizations fail to reinvest those time savings into high-value activities, according to Gartner. One possible explanation is that some of the time savings are lost to the double-checking, toggling, and prompt-tweaking often required when systems remain disconnected.
• The verification tax: If retrofitted software relies on rigid, siloed data pools, the AI may lack the full context needed to produce reliable outputs. Employees can generate content in seconds, but then have to spend minutes validating facts, correcting errors, and cleaning data.
• The integration friction: Retrofitted tools may rely on custom, point-to-point APIs that struggle to provide real-time enterprise context. While native architectures leverage open standards like Model Context Protocol (MCP) to give agents direct, standardized connection to data silos and workflows, legacy wrappers can remain trapped behind brittle integrations — forcing AI to operate on incomplete data while humans bridge the gap.
• The token bill shock: Grafting an agentic AI loop onto an inefficient, fragmented architecture can force the system to repeatedly resubmit massive contexts. This has the potential to compress gross margins and push significant API token costs onto the enterprise buyer.

Ultimately, retrofitting software with AI may fail to eliminate underlying inefficiencies and may simply accelerate them. Employees move at a higher velocity, but the organization stays in the exact same place. It’s what one AI expert sagely referred to as the “hamster wheel” effect.

From features to responsibility

Getting off the hamster wheel means redesigning platforms, redefining roles, and realigning thinking.

It’s a big shift, but one that’s absolutely necessary. In the legacy computing era, enterprise software competed strictly on the volume of its features — longer navigation menus, advanced filtering options, and dense configuration dashboards. The winning platform was simply the one that shipped the most code.

The AI-native era is entirely different. It’s defined not by the tools the user is forced to manage, but by the business outcomes the software autonomously owns. The base prompt of the modern application has changed from “display these options for the user” to “own the optimization of this process, anticipate systemic failures, and deliver verified solutions.”

You probably won’t achieve this level of operational autonomy by simply adding a conversational layer to a system that was structurally engineered to wait for human commands.

Many companies are getting the message that their corporate tech stacks require a complete re-architecture. According to Deloitte, 78% of technology leaders plan to integrate autonomous agents directly into their core architecture workflows over … wait for it … the next five years.

That’s an eternity — especially since some platforms have already made the jump.

The T&E example

One place where this architectural challenge is especially visible is in travel and expense management.

A structural problem confronting legacy T&E software is a massive visibility gap driven by user rejection of the corporate travel solution. According to The State of Corporate Travel and Expense 2026, a report by Skift and Navan, a staggering 80% of business travelers admit to booking off-platform at least some of the time — bypassing their corporate tools in search of better inventory, a faster user experience, or cheaper fares.

Every time that happens, it creates an immediate blind spot for the enterprise. Negotiated corporate rates go unused, real-time duty of care tracking breaks down, and spend data completely vanishes from financial view — until weeks later, when an expense report is manually submitted.

When a company relies on a T&E platform that has added AI on top of a system that wasn’t originally designed for it, the AI layer may lack the real-time context it needs to deliver meaningful results. If the underlying architecture can’t natively capture and sync booking data, card transactions, and HR profiles across channels, the AI is working with an incomplete picture of the company’s actual activity.

An AI-native architecture, by contrast, may be better positioned to reduce structural fragmentation. Uniting each element of the T&E process — booking experience, corporate card ledger, expense management workflows — into a single data source creates entirely new processes that maximize the value of AI, enabling it to act proactively and autonomously.

For example, AI is of limited help when it just reads an invoice after the fact. But when AI is woven into the platform, it can actively capture and reconcile the transaction against scores of data points — all in real time — turning a high-friction administrative chore into a zero-touch, automated business outcome.

The next era won’t be won by the past

Macro-technological shifts rarely favor incumbents who are forced to adapt legacy infrastructure to a new paradigm. The commercialization of the internet gave rise to entirely new, native business models, while the shift from desktop to mobile computing rewarded companies that built natively for the smartphone ecosystem.

The agentic revolution is following the exact same pattern. And one lesson is already becoming clear: Grafting intelligence onto a workflow that wasn’t designed for it is a poor bet for producing transformative bottom-line value. True business transformation requires rethinking fragmented middleware and building on a unified foundation where AI is the core architecture — not an afterthought.

The next era of enterprise value won’t be won by racing to patch architectural deficits with thin AI overlays. It will be defined by the systems that were built natively for it.