The Competence Debt: What We’re Losing in the Rush to Become AI-First

I’ve spent a lot of my career talking about productivity.

For much of that time, that meant talking about Microsoft Office, and later Microsoft 365. Word, Excel, PowerPoint, Outlook, OneNote, SharePoint, Teams, OneDrive, and the vast ecosystem of tools and services that have accumulated around them. I’ve written about them, trained people on them, spoken about them at conferences, and spent countless hours helping organizations figure out how to weave them into their everyday work.

Over those years, I’ve watched the definition of productivity shift repeatedly.

We moved from isolated desktop software to connected cloud suites. From local file shares to SharePoint and OneDrive. From emailed attachments to live co-authoring. From individual output to collaborative workflows. Every transition expanded what people needed to learn, but crucially, it never eliminated what came before.

Now, we’re navigating another massive platform shift with artificial intelligence. And this one worries me.

Not because we’re talking too much about AI—I spend a substantial amount of my own time writing, speaking, building, and training around Microsoft Copilot, autonomous agents, and AI-enabled work.

My concern is what we’ve completely stopped talking about in the process.

Where Did All the Productivity Training Go?

Think back just a few years.

Community user groups and industry conferences were packed with sessions on Excel formulas, PowerPoint storytelling, Word template hygiene, OneNote information architecture, and Teams governance. Organizations ran regular lunch-and-learns. Power users shared clever tips over coffee. Someone would show you a keyboard shortcut or a cleaner way to structure a spreadsheet, and you’d quietly save five minutes every workday for the next decade.

There was a vibrant, informal learning ecosystem built entirely around foundational digital craft.

Somewhere along the way, that conversation was swallowed whole by AI.

Scan today’s conference agendas, corporate training calendars, webinars, social feeds, and tech blogs. An overwhelming percentage of the airtime is consumed by Copilot, generative models, prompt engineering, agent orchestration, and automated workflows.

The excitement is understandable. AI represents an enormous technological leap. Organizations are investing heavily to capture early returns, software vendors have new subscription tiers to justify, and employees are eager to make sense of it all.

                  ┌───────────────────────────────┐
  What We Train:  │   Copilot  •  Prompts  •  AI  │  ◀── 90% of Mindshare
                  ├───────────────────────────────┤
  What We Use:    │  Excel • Word • PPT • Teams   │  ◀── 8 Hours a Day
                  └───────────────────────────────┘

The software didn’t vanish. The learning ecosystem supporting it did.

And in that vacuum, organizations are quietly accumulating what I call competence debt.

What Is Competence Debt?

The dynamic mirrors technical debt in software development. Technical debt occurs when engineers take expedient shortcuts to ship features faster today, deferring structural rigor until the accrued interest creates systemic drag tomorrow.

Competence debt works the exact same way with human skills:

• We skip teaching how structured tables work in Excel because Copilot can analyze the range.
• We gloss over paragraph styles and heading hierarchies in Word because an LLM can draft and reformat the narrative.
• We stop teaching visual hierarchy in slide design because an AI can assemble a full deck on command.
• We bypass information architecture because users can just query natural-language search to locate files.

In the short term, this looks like pure velocity. Work gets generated faster, and complex tasks that once required deep application familiarity are suddenly accessible to novices.

There is credible research confirming these initial gains. Landmark studies on customer-support agents and enterprise software developers demonstrate that generative tools increase raw throughput significantly, delivering the highest relative lift to newer, less-experienced workers.

That capability is remarkable, but it introduces a subtle trap: being able to prompt an answer is fundamentally different from understanding the answer.

AI Hides Incompetence Dangerously Well

Before generative AI, a lack of application literacy was self-evident. If you didn’t understand how to build an Excel model, the formula errored out, the sheet broke, and the problem remained visible.

Today, an employee with minimal spreadsheet literacy can ask Copilot to generate a calculation, analyze a messy dataset, or forecast a trend. The resulting output arrives formatted, polished, and immediately plausible.

That surface-level polish is precisely what makes it dangerous:

• What happens when the formula is subtly flawed, but formatted convincingly?
• What happens when the calculation runs without errors, but solves the wrong business question?
• What happens when a slide deck looks visually balanced, but completely obscures the core message?
• What happens when ungrounded source files lead an AI to summarize outdated policy with absolute confidence?

We have spent years training people on how to generate. We need to spend just as much energy teaching them how to evaluate.

The Clues in the Data: Judgment as the Core Skill

The data increasingly supports this shift. The Microsoft Work Trend Index highlights how human responsibilities evolve as AI becomes a permanent layer in the operating environment:

These numbers tell a clear story. High-value work is moving away from mechanical generation and toward editorial verification. But verification requires baseline domain knowledge:

• You cannot quality-control an Excel formula if you don’t understand basic data modeling and referencing logic.
• You cannot evaluate a presentation’s flow if you don’t understand structural storytelling.
• You cannot judge whether a Word document is durable and accessible if styles, sections, and metadata are unfamiliar concepts.
• You cannot catch a hallucination if you don’t know what the truth looks like.

The research also profiles a segment termed “Frontier Professionals”—the top tier of power users who consistently extract outsized business value from AI. What sets them apart isn’t that they send more prompts. It’s that they refuse to outsource their thinking. They use AI as an accelerator while maintaining tight editorial control over the underlying logic.

“The Basics” Aren’t Basic

Part of the problem lies in the label itself. Calling foundational digital skills “the basics” makes them sound trivial, when they are actually structural concepts in disguise:

• Pressing Ctrl + T in Excel is a keystroke; understanding relational data integrity and structured references is a core business skill.
• Opening the Styles pane in Word is a feature; understanding semantic hierarchy, document accessibility, and machine-readable structure directly dictates how well AI models can ingest that content.
• SharePoint isn’t merely file storage—it is an information architecture framework. Teams isn’t just persistent chat—it is a collaboration protocol.

These platforms encode intentional models for information flow, data hygiene, and organizational governance. Natural language interfaces do not render those architectural concepts obsolete; they make them critical.

The Fallacy of AI as Remedial Education

The most common strategic misstep is treating AI licenses as an instant workaround for organizational digital literacy.

The playbook has become familiar: provision Copilot seats, host a prompt engineering webinar, distribute a prompt cheat sheet, and expect measurable productivity gains.

       Broken Process  +  AI  =  Faster Chaos
  Messy Data Structure +  AI  =  Polished Hallucinations

If underlying business practices and data habits are messy, AI simply accelerates the dysfunction. A poorly normalized spreadsheet remains unreliable, an unstructured intranet remains unsearchable, and a confusing slide deck is merely generated in seconds instead of hours.

This is why training cannot treat AI adoption as a disconnected silo. We shouldn’t teach Copilot instead of Excel; we must teach Excel through the lens of Copilot. Not prompt writing instead of document design, but structured authoring augmented by AI.

The Silent Erosion of Hallway Knowledge

There is an institutional cost here that rarely appears on a training dashboard: the slow erosion of peer-to-peer expertise.

For decades, enterprise competence relied on informal “hallway experts”—the colleague who mastered Power Query, the teammate who knew presentation layouts inside and out, or the peer who understood metadata inheritance. These individuals raised the baseline competence of entire departments through informal mentorship.

When organizations position every challenge as a problem to hand off to an AI chat box, fewer people build the foundational fluency required to become those internal guides. Over time, an organization ends up with an entire workforce generating plausible material, with vanishingly few people who understand how any of it works under the hood.

That is when the competence debt comes due.

Redefining Productivity

Speed and volume are vanity metrics when decoupled from correctness:

• Generating ten decks in the time it used to take to build two isn’t productive if the narrative misses the strategic mark.
• Producing complex financial models in minutes is counterproductive if a hidden circular logic error slips past unreviewed.

AI has driven the marginal cost of content creation close to zero. As a result, human judgment, domain literacy, and structural verification are now the scarcest, highest-leverage assets in the enterprise.

The goal isn’t to push back against AI adoption. The capabilities across modern productivity suites are transformative, and we are only scratching the surface of what agentic workflows will enable. But we must stop pretending that adding an intelligence layer eliminates the need for fundamental competence.

We don’t need to choose between digital literacy and AI fluency. We need an integrated model where one reinforces the other.

In Part 2, we’ll turn this diagnosis into an operational strategy. We’ll map out a three-tiered enterprise training framework that moves from baseline structural fluency, through real-world scenario workflows, to advanced AI augmentation—ensuring your workforce knows what to delegate to AI, what to own, and how to verify the difference.