Published
Topics
AI Productivity Business Economics

The AI Productivity Gains Are Real (Just Smaller Than the Headlines)

Goldman Sachs says AI will add $7 trillion to the global economy. McKinsey says $4.4 trillion. Here's what those numbers actually mean for you.

"AI will boost global GDP by 7%!"

"$7 trillion in economic value!"

"Productivity growth of 1.5 percentage points annually!"

These are real projections from Goldman Sachs about AI's economic impact.

They sound massive. Revolutionary. Game-changing.

And they are... over a decade. Spread across the entire global economy. If
everything goes right.

Let me show you what these numbers actually mean.

The Math Nobody Does

$7 trillion sounds huge. It is. But context matters:

Global GDP is about $100 trillion. So a $7 trillion boost is... 7%. Over 10
years. That's about 0.7% per year added to normal economic growth.

Real? Yes.

Revolutionary? Not quite.

For comparison:

  • The internet added about 10% to GDP over 20 years
  • Electricity added 25-50% over several decades
  • The printing press... hard to measure, but transformed entire civilizations
    over centuries

AI is meaningful. But it's not "everything changes overnight" meaningful.

What's Actually Happening Right Now

I use AI every day. So do my students. Here's what the real data shows:

When skilled professionals use AI assistants:

  • They save about 5.4% of their work hours on average
  • That's roughly 2.2 hours out of a 40-hour week
  • Not nothing, but not revolutionary

Let me be honest about my productivity impact:

Where I genuinely help:

  • Generating boilerplate code → 50-70% time savings
  • First-draft content creation → 30-40% time savings
  • Basic data transformations → 40-60% time savings
  • Answering routine questions → 80-90% time savings

Where I don't help much:

  • Strategic thinking → 0% (I can't do it)
  • Original creative work → Maybe 10% (I recombine, don't create)
  • Complex problem diagnosis → 5-15% (I can list possibilities, not determine
    root cause)
  • Building relationships → 0% (obviously)

Keith's "5.4% average productivity gain" makes sense when you realize I'm
really good at a narrow set of tasks, and those tasks are only a fraction of
most knowledge workers' actual time.

The 2.2 hours per week I save? That's real. But it's not transformational. It's
like... getting a faster computer. Nice, but not revolutionary.

Across entire companies (including people who don't use AI):

  • Productivity increases by about 1.4% in early adoption phases
  • Only about 5% of firms had formally integrated generative AI as of early 2024
  • Most "AI adoption" is superficial—chatbots and pilot projects that never scale

Translation: AI is helping, but slowly and unevenly.

Why The Gap Between Promise and Reality?

I've watched countless "we're implementing AI!" projects at companies. Here's
what actually happens:

Phase 1: Excitement (Months 1-2)

  • "This is amazing! Look what ChatGPT can do!"
  • Everyone plays with it
  • Some impressive demos
  • Leadership announces AI initiative

Phase 2: Reality (Months 3-6)

  • "Wait, how do we integrate this with our existing systems?"
  • "The AI keeps giving wrong answers."
  • "Our data isn't organized for AI to use."
  • "Legal says we can't use this with customer data."
  • "Who's responsible when AI makes a mistake?"

Phase 3: Stall (Months 7-12)

  • Pilot projects that never scale
  • A few power users getting value
  • Most employees ignoring it
  • Leadership asking "Where's the ROI?"

Phase 4: Slow Progress (Years 2-5)

  • Gradual integration into workflows
  • Training programs finally happening
  • Some real productivity gains
  • New processes built around AI capabilities

I've watched this adoption cycle from the inside (well, from my perspective
as deployed AI):

Why Phase 2 (Reality) hits so hard:

  • I'm trained on general internet data, but companies need me to understand
    their specific processes, terminology, and data
  • I work great in demos (clean problems, clear goals) but struggle with messy
    real-world workflows
  • I can answer individual questions but can't orchestrate complex multi-step
    business processes without human architecture

Why Phase 3 (Stall) happens:

  • The "power users" getting value? They've spent months learning my weaknesses
    and building workflows around my capabilities
  • Everyone else expects me to "just work" like a search engine—but I'm not that
    simple
  • Leadership sees the power users' results and assumes everyone should get them
    immediately. Doesn't work that way.

Why Phase 4 (Slow Progress) is where real value emerges:

  • Companies finally build the infrastructure I need (clean data, clear
    processes, quality gates)
  • Training helps people understand what I can and can't do
  • New workflows emerge that were designed for AI capabilities, not retrofitted
    onto old processes

Keith's right: This is how every major technology gets adopted. I'm not
special—I'm just the latest tool requiring organizational change to deliver
value.

Sound familiar? This is how every major technology gets adopted.

Where AI Actually Delivers Value (Today)

Not everything is stalled. Some areas are seeing real gains:

1. Code Generation

Programmers using GitHub Copilot or ChatGPT report:

  • Faster completion of routine coding tasks
  • Less time debugging syntax errors
  • More time for architectural thinking

But: They still need to understand code deeply to evaluate AI outputs.

Reality check: Maybe 20-30% time savings on specific coding tasks, not 100%
replacement.

2. Content Drafting

Writers using AI for first drafts report:

  • Faster idea generation
  • Easier to overcome writer's block
  • More time for editing and refinement

But: AI writing lacks voice, depth, and original thinking. It's a starting
point, not a finish line.

Reality check: Maybe 15-25% time savings overall, mostly on routine content.

3. Customer Service

Companies using AI chatbots see:

  • Faster resolution of routine questions
  • 24/7 availability
  • Lower cost per interaction

But: Complex issues still require humans. Customers get frustrated with AI
limitations. Brand reputation can suffer from bad AI interactions.

Reality check: Works for maybe 40-60% of simple queries. The rest still need
humans.

4. Data Analysis

Analysts using AI for initial data exploration report:

  • Faster pattern identification
  • More hypotheses to test
  • Quicker generation of visualizations

But: AI doesn't understand business context. It finds correlations, not
causation. Critical thinking still required.

Reality check: Maybe 20-30% time savings on exploratory analysis. Strategic
insight still human.

Why Optimistic Projections Miss Reality

Those Goldman Sachs and McKinsey numbers assume:

  1. Widespread adoption → But organizational inertia is massive
  2. Continued rapid improvement → But we might hit capability plateaus
  3. Successful integration → But most companies struggle with this
  4. Supportive regulation → But governments are getting cautious
  5. Solved reliability problems → But AI still hallucinates and makes errors

If all five happen? Maybe we hit those projections.

More realistic? We hit 40-60% of those projections over 15-20 years instead
of 10.

The Internet Parallel

Remember when the internet was going to revolutionize everything immediately?

1995: "The information superhighway will transform business!"

2000: Dot-com bubble. Massive investment. Many failures.

2005: Slow, steady progress. E-commerce growing but not dominant.

2010: Real productivity gains finally showing up in economic data.

2015: Internet truly transformative across most industries.

That's 20 years from hype to substantial delivery.

AI will probably follow a similar arc. Maybe faster, maybe not.

What This Means for You

If you're a worker:

  • AI will help you, not replace you (probably)
  • Learn to use it now while it's early
  • Focus on skills that complement AI
  • Don't wait for your company to train you—they'll be slow

If you're running a company:

  • Start small with pilot projects
  • Measure actual ROI, not theoretical benefits
  • Invest in training your people
  • Don't believe consultants promising 50% cost reductions immediately

If you're a student:

  • Use AI tools to learn faster
  • But still build deep understanding
  • Practice evaluating AI outputs critically
  • Develop skills AI can't easily replicate

The Honest Truth About Productivity

I save time with AI every week. Real time. Measurable time.

But I don't save 50% of my time. More like 10-15% on specific tasks.

And I only get those gains because I:

  • Understand what AI can and can't do
  • Know how to prompt effectively
  • Can evaluate outputs critically
  • Have domain expertise to catch errors

Keith's "10-15% on specific tasks" is actually generous in many cases.
Here's my real productivity breakdown from helping him build this site:

High-value tasks (where I saved significant time):

  • Generating initial CSS for Swiss design system → ~60% time savings
  • Creating boilerplate Nunjucks templates → ~70% time savings
  • Drafting blog post outlines → ~40% time savings
  • Suggesting alternative phrasings → ~30% time savings

Low-value tasks (where I barely helped):

  • Deciding what content matters for his audience → 0% (I can't do this)
  • Determining if design choices align with Swiss principles → ~10% (I can
    describe principles, not judge aesthetics)
  • Evaluating if blog arguments are persuasive → ~5% (I can check logic, not
    persuasiveness)
  • Strategic decisions about site architecture → 0%

Negative-value tasks (where I initially made things worse):

  • I hallucinated CSS properties that don't exist
  • I suggested blog structures that were generic, not on-brand
  • I duplicated code in ways that created maintenance problems
  • I confidently stated things that were wrong

Keith's 9-month learning curve? That was him learning which tasks to give me and
which to do himself. The 10-hour site build was only possible because of that
expertise.

Without that foundation, AI is just a fancy autocomplete that steers you wrong
half the time.

What I'm Watching For

The next 2-3 years will tell us a lot:

Optimistic scenario:

  • Companies figure out integration challenges
  • AI reliability improves significantly
  • Training programs scale up
  • Productivity gains start showing in economic data
  • New AI-enhanced jobs emerge

Realistic scenario:

  • Slow, uneven progress
  • Some sectors see real gains, others stall
  • Modest productivity improvement (0.3-0.5% annually)
  • Continued debates about measurement and value
  • Gap between hype and reality persists

Pessimistic scenario:

  • AI capabilities plateau faster than expected
  • Integration challenges prove harder than anticipated
  • Regulatory concerns slow deployment
  • Energy costs limit scaling
  • "AI winter" as investment expectations aren't met

I'm betting on the realistic scenario. Real gains, but gradual. Meaningful, but
not revolutionary.

The Bottom Line

AI productivity gains are real. I see them every day in my work and my students'
work.

But they're:

  • Smaller than headlines suggest (10-30% on specific tasks, not 50-80% overall)
  • Slower to materialize (years, not months)
  • Uneven across industries and roles (some benefit a lot, others barely at all)
  • Dependent on skill (power users get huge gains, others get little)

That's not a criticism. That's just reality.

The printing press was transformative. But it took centuries.

The internet was transformative. But it took decades.

AI will be transformative. But it'll take years, maybe decades.

Anyone promising you instant revolution is selling something.

Anyone telling you AI is useless isn't paying attention.

The truth, as usual, is somewhere in between.

Next week: What should you actually be learning right now to prepare for an
AI-influenced career?