Here’s Why AI Tools Are Slowing Some Companies Down — and How to Fix the Problem

In Kurt Vonnegut’s “Harrison Bergeron,” a dystopian government enforces strict equality by handicapping the gifted. For the highly intelligent, the handicap is a government-mandated earpiece that blasts a jarring buzzer every few seconds, scattering their thoughts and halting their momentum. It’s an artificial limit placed on human potential.

Today, companies are building the inverse: Artificial intelligence is amplifying the already-privileged up from the mean, while simultaneously placing artificial limits on the potential of the less-privileged. Call it a privilege gap.

That wasn’t the intention. In fact, when we handed everyone the same AI tools, we declared the playing field level. But the privilege gap didn’t close; it just became invisible. That’s because the inequality was never just about the tools; it was built into the architecture. The modern handicap — the equivalent of that government transmitter — is organizational permissions. It’s the lack of a cohesive strategy communicated across the business. It’s people stuck in a narrow scope or rigid job description.

As a result, every time an engaged, passionate employee tries to use AI to solve a problem and hits a wall, the buzzer sounds.

The Anatomy of the Buzzer

The privilege gap doesn’t show up in Slack or Gmail. It appears in the momentum-killing friction of everyday tasks. For today’s workforce, the buzzer sounds like:

• “Access denied — file a ticket”: Triggered by a data warehouse query the junior analyst cannot run, or an API key a contractor cannot provision.
• “That’s outside your scope”: Heard when a team lead can’t act on a cross-departmental insight because they lack the political capital or clearance to step outside their lane.
• “Awaiting authorization”: Set off when a frontline support rep sees a systemic flaw, but whose job description is to strictly “close tickets,” not “analyze trends.”

An executive’s AI is able to perform multiple actions: pull customer data, cross-reference contracts, draft the response, schedule the follow-up. But a frontline employee’s AI stops at step one, waits for manual intervention, loses context, and has to restart.

The Data: Amplifying the Privileged

Job postings requiring AI skills pay 28% more than those without, according to Lightcast’s analysis of 1.3 billion job postings. The Dallas Federal Reserve found that AI augments experienced workers while substituting for entry-level ones — meaning the premium compounds with seniority, not despite it.

Consider two employees: a machine learning engineer with five years at the company and a customer success rep hired six months ago. The engineer holds database read access, cloud admin roles, and the mandate to “solve problems.” The rep holds read-only Salesforce access and the mandate to “stay on script.”

That difference in access widens the gap, even when both employees have the exact same AI assistant. The engineer’s AI writes the script, runs the query, posts the alert, and files the ticket. The rep’s AI drafts a message and stops, waiting for someone with permissions to act.

The bottom line? Companies pretend they’ve democratized AI by rolling out a web interface to the whole company. But employees don’t get access to agentic tooling just because they have a login to an AI dashboard. True agents require agency, and agency requires system access.

Those companies are missing a huge opportunity. According to research from the National Bureau of Economic Research (NBER), low-skilled workers see a 34% productivity gain from AI — the largest of any cohort. But uptake is concentrated among higher-educated, higher-occupation workers, found another NBER study, mirroring the unequal diffusion of early personal computers. The paradox is stark: Those who stand to gain the most are the least likely to have access to the tools.

How bad is the damage? A Google/Ipsos poll found that the 5% of workers who are genuinely “AI-fluent” earn 4.5 times more and are promoted four times as often. But fluency isn’t prompt engineering. Fluency is system access and organizational permission to act.

How Can Companies Enable Their Passionate People?

Most companies get it wrong. They conflate access to AI with permission to act. They hand someone a tool and call it enablement. But customer service agents with a chatbot and no ability to issue a refund, pull account history, or escalate without a manager aren’t enabled — they’re merely decorated. The tool surfaces the answer, but the org structure prevents the action.

Passionate people already have the motivation. What they lack is the mandate. The fix is a fundamental redesign of both our technology and our culture, starting with these two steps:

• Redesign Identity and Access Management (IAM) for AI: Expand read access to anonymized data and aggregated metrics. Grant scoped write access with audit trails. Replace approval queues with real-time permissioning, where AI requests access, the system grants short-lived credentials, the action completes, and the session ends.
• Redesign organizational strategy: Communicate a clear AI strategy across the business so employees understand what the company is trying to solve. Relax rigid job descriptions to allow passionate employees to follow the insights their AI generates.

Breaking the Chains at Navan

At Navan, we aren’t waiting for the industry to figure this out. We’re actively dismantling the invisible privilege gap by spreading safe, agentic access to everyone — not just the engineers and executives.

We’re building an environment where every employee is genuinely empowered to do their best work in an AI-augmented manner. This means giving our people the systemic clearance to quickly pull outside data into their workflows and bounce it against internal, proprietary data. The result is richer, more informed decisions, faster execution, and communications that are deeply contextualized rather than generic.

Is it risky? Yes — but so is the status quo. Today, people sometimes make the wrong decisions and jump to conclusions. They may act on incomplete information, tribal knowledge, and gut instinct. What changes with AI isn’t that mistakes become possible. It’s that mistakes become traceable, auditable, and correctable in ways they never were before.

We feel that clinging to centralized control is a far greater risk. Accepting the friction of human error is the only way to break the chains of legacy permissions. You can’t build a truly AI-native company if you’re terrified of your own employees having the power to act.

Companies that defer this work will deploy tools widely, train everyone on AI, and watch the outcome gap widen. The buzzer will keep sounding for the lower-level employees. But unlike Vonnegut’s world, where the noise interrupted the gifted, our modern buzzer only interrupts the underprivileged. The executives with full power and access will never hear a thing and wonder why their organizations continue to underperform.