Why AI’s Productivity Promise Falls Apart Without Human Expertise
AI gets you to 60% quickly, but human expertise is what unlocks 100%.
Agor2012 | iStock
Executives are salivating over AI’s promise of productivity gains. A Stanford University study found workers using generative AI completed tasks three times faster, cutting 90-minute assignments to just 30 minutes. Technology vendors have seized on this “tripling of productivity” to convince business leaders that they can’t afford to be left behind in the AI-driven efficiency boom. Simple math says if tasks take a third of the time, you may only need a third of your staff.
It’s a seductive equation, but it’s also deeply flawed.
I recently used ChatGPT to create a comic strip for a LinkedIn post. Within minutes, I had something impressive. Even though I possess some artistic talent, I couldn’t have produced the panels and visual style this quickly.
After my initial elation, I then noticed subtle flaws in the comic. I realized I had 60 percent of what I needed and further prompting wouldn’t close the gap—at least, not in a timely fashion. Instead, I employed my knowledge of Photoshop to remove unwanted elements and fix composition issues. That got me to about 85 percent.
The first draft generated by ChatGPT had several problems, which I was able to address with my Photoshop expertise.
Brent Dykes | AnalyticsHero, LLC
One panel still had a problem where an object didn’t match its shape in the preceding panels (box dimensions). I could have redrawn that part of the panel, but was the effort worth it for a LinkedIn post? Probably not, so I shipped it at 85 percent.
This experience reveals why the bold productivity claims for AI will underdeliver. The AI tools can get everyone to 60 percent in minutes. In this case, someone without Photoshop skills could probably push to 70-to-75 percent through patient, iterative prompting, but it would take significantly longer. My design expertise wasn’t just about reaching a higher quality threshold—it was about reaching it efficiently.
The Stanford study found that workers only applied AI to about one-third of their tasks. The bottleneck isn’t creating the initial draft—AI excels at that. The challenge is refining that draft into something effective and shippable, which requires expertise and human judgment that AI can’t provide. Without sufficient domain expertise, the worker may not even know what good or effective looks like, limiting what they can achieve with AI.
Many leaders make a costly mistake with AI and productivity. They assume the gains are evenly distributed across the work, when in reality they are heavily concentrated in the earliest stages.
The Four Zones of AI Productivity
To understand where and how this plays out, I’ve mapped the process to four distinct zones between zero percent (nothing created) and 100 percent (excellence). Each zone requires different capabilities and shows how expertise still influences speed and quality.
Generative AI accelerates the start of work, but human expertise determines how far it goes.
Brent Dykes | AnalyticsHero, LLC
The Launch (0%→60%)
This first zone drives the productivity hype. You provide a prompt, and within minutes the tool produces an initial draft—website copy, a business presentation, a prototype of software code, or in my case, a comic strip. While an expert might craft a better prompt and reach 60 percent faster, even a novice gets there through simple iteration. AI does most of the work, and everyone reaches this threshold far faster—creating the illusion of equal capability.
The Rework (60%→80%)
To produce something shippable, humans must take control and edit the initial output. The work begins with verifying facts, aligning content to purpose, personalizing the voice and fixing layout or composition issues. Novices must continue refining through prompts—a slow, iterative process that some may abandon prematurely if they’re impatient. Experts can deploy other tools and skills to make adjustments far more quickly. In my comic strip example, I used Photoshop to make corrections rather than trying to coax ChatGPT to adjust specific elements without breaking something else.
The Forge (80%→90%)
Now you reach a point where you must create, not just refine. In my comic, this meant drawing additional objects to complete the visual story. For a marketer, it might mean adding an additional paragraph based on a real-world case study that AI doesn’t know about. For a developer, it’s hand-coding a custom integration with a legacy internal system that AI can’t access. The productivity gains disappear here because the work requires context, knowledge and capabilities that AI doesn’t possess. While AI tools may still be able to assist with the process, the work is time-intensive even for experts and often impossible for novices without the requisite skills.
The Summit (90%→100%)
The last zone is both expertise-gated and optional. To push beyond very good to excellent, you need deep specialized skills. But it isn’t just about possessing the right technical abilities, it’s about discerning when something is good enough and when excellence actually matters. Human judgment determines when the effort is worthwhile and when you’re pursuing pointless perfection. For my comic strip, one panel had an inconsistency—an object didn’t match its appearance in other panels. Redrawing it to match the AI-generated style would have taken hours with uncertain results. I decided it wasn’t necessary and shipped it at 85 percent. The LinkedIn post ended up being my second most popular one in 2025.
Here’s what this means in practice. Consider how long it takes workers at different expertise levels to move through each zone:
AI compresses early work for everyone, but expertise determines speed, reach and final quality.
Brent Dykes | AnalyticsHero, LLC
If we’re not careful, the implications can be stark. An expert completes the entire journey—start to perfection—in about three and a half hours. An intermediate worker needs six hours and can’t reach excellence. A novice? They might spend four hours just to reach 60-to-70 percent, unable to progress further without expertise they don’t possess.
Even if more sophisticated AI can compress the early zones, it won’t eliminate the expertise required in later ones. This is why the “3x productivity equals one third the headcount” equation falls apart. You’re not just losing speed when you cut experts—you’re losing the ability to reach zones 3 and 4 entirely. The work gets stuck at 60-to-70 percent, which means more output but less that’s shippable. The productivity gains are concentrated in The Launch phase, but business value is created in the later zones where expertise still defines outcomes.
While it’s seductive to calculate the headcount savings from AI-driven productivity, it overlooks a fundamental truth about how AI actually works. Recent research from Anthropic’s Economic Index reveals a near perfect correlation (r > 0.92) between the sophistication of a user’s prompt and the sophistication of AI’s response. How humans direct the AI determines how effective it can be and what it can accomplish—at every stage of the process. Expertise shapes both what you can extract from AI and what you can achieve with its output.
Some will argue that future AI improvements may extend its impact into the later zones. As generative AI tools become more capable in these later phases, the expertise advantage won’t disappear. Experts will extract better outputs faster because they know what to ask for, how to evaluate it and how to finalize it. Better AI tools won’t close the capability gap but amplify it further. The smartest organizations will figure out how to deploy AI to amplify their collective expertise, not eliminate it. If you get it wrong? Your “3x productivity gain” becomes a flood of work that can’t or shouldn’t be shipped.
