You're Measuring AI Productivity Wrong (And Why That Matters for Your Business)

The Question Everyone's Asking

Here's a question I hear constantly: "How much longer would it take you to do your work without AI?"

Maybe you've asked yourself the same thing. You're using ChatGPT to draft emails, Claude to analyze documents, or some other AI tool that's become part of your daily workflow. You try to imagine going back to the old way and it feels impossible. Five times slower? Ten times? The number feels huge.

That's exactly the problem. You're asking the wrong question, and it's giving you a wildly inflated answer.

A Researcher's Honest Assessment

Last week, Ryan Greenblatt published something unusual. Greenblatt is a researcher at Redwood Research, an AI safety lab that studies how to make AI systems safer and more reliable. He wrote what he calls a "scenario forecast for the present." Essentially, his best guesses about what's actually happening in AI right now, including things that aren't publicly confirmed.

This isn't a press release. It's not a vendor pitch. It's an insider-adjacent researcher trying to cut through the noise and measure what's real. And his central finding about AI productivity should change how you think about your AI investments.

The Two Questions

Greenblatt argues there are two ways to measure AI's productivity impact. Most people ask the first question. Almost nobody asks the second.

The wrong question: "How much longer would it take me to do the work I'm currently doing without AI?"

When you ask this question, you might answer 3x, 5x, even 20x. And you're not lying. But the number is inflated for a specific reason: you've already changed what you do.

Think about it. You've shifted your workflow toward tasks where AI helps most. You're doing work you couldn't do before (or wouldn't have attempted). You've stopped doing certain manual tasks entirely because AI made them obsolete. You're measuring against a baseline that no longer exists.

The right question: "How much would we have to speed you up (working faster or working more hours with no drop in quality) before you'd be indifferent between that speed-up and having AI tools?"

This question is harder. It forces you to compare apples to apples. It asks: what's the actual productivity gain when you account for the fact that you've already adapted your entire workflow around AI?

Greenblatt's answer for the best AI companies in the world (Anthropic, OpenAI, the places with the most sophisticated AI usage) as of April 2026: about 1.6x.

Not 5x. Not 10x. 1.6x.

And here's the kicker: those are the best AI companies in the world. For most businesses, the number is lower. But it's still real, and it's still worth pursuing.

Why This Matters for Your Business

If you've been reading LinkedIn or listening to AI vendors, you've probably heard claims that sound like the first question's answer. "AI made us 5x more productive." "We cut our content creation time by 80%." "Our team accomplishes in one day what used to take a week."

Those numbers aren't necessarily lies. They're measuring the wrong thing. They're comparing current AI-assisted workflows to hypothetical non-AI workflows that nobody actually uses anymore.

The 1.6x number is more honest, and honestly, more useful. Here's why.

First, it helps you set realistic expectations. If you're expecting AI to transform your business overnight, you're going to be disappointed. If you're expecting a meaningful but modest improvement that compounds over time, you'll be pleasantly surprised.

Second, it helps you budget accurately. A tool that genuinely provides a 1.6x speed-up to your team is valuable. But it's not worth the same price as a tool that would provide a 5x speed-up. When vendors promise transformation, you can translate that to reality: they're probably offering somewhere between 1.3x and 2x, depending on your team and your use case.

Third, it helps you avoid overpaying. The AI tools market is full of products that promise revolutionary results. Some are good. Many are overpriced. If you understand that even the best AI companies in the world are seeing 1.6x gains, you can evaluate vendor claims with appropriate skepticism.

Fourth, and this is important: the gains aren't evenly distributed. Greenblatt notes that some specific tasks see 3x to 10x reductions in human time required. Writing first drafts, generating test cases, analyzing structured data, these tasks can genuinely be much faster with AI. Other tasks see almost no improvement. Complex reasoning, creative strategy, anything requiring deep domain expertise, these still mostly require humans working at human speed.

The overall 1.6x average includes both types of tasks. Which means if you can identify the high-leverage tasks in your business and focus your AI adoption there, you might see gains well above 1.6x in those specific areas. But expecting every task to speed up by 5x is setting yourself up for disappointment.

The Quality Trade-Off Nobody Talks About

Here's something else Greenblatt observed that matches what I see with my clients: AI-produced work is generally sloppier and less reliable than human-only work.

This isn't AI-bashing. It's just reality. When you use AI to draft an email, write code, or analyze data, you're trading some quality and reliability for speed. Sometimes that's a great trade. Sometimes it's dangerous.

Greenblatt points out something particularly interesting: it's increasingly common for nobody, including the AI itself, to fully understand how a piece of AI-generated output works. You get a result that looks right, tests okay, and ships. But the deep comprehension isn't there.

For some tasks, that's fine. If you're using AI to draft routine customer service responses, a little sloppiness is acceptable. You review it, clean it up, and send it. The speed gain outweighs the quality loss.

For other tasks, it's not fine. If you're using AI to draft legal documents, analyze financial data, or make strategic decisions, sloppiness can be expensive. You might save 30 minutes on the initial draft and spend three hours cleaning up the errors, or worse, miss an error that costs you real money later.

The smart approach: categorize your tasks. Which ones are high-volume, low-stakes, and benefit from speed? Use AI aggressively there. Which ones are high-stakes, require deep accuracy, and can't afford errors? Use AI cautiously or not at all.

The Bigger Economic Picture

Greenblatt also estimates some bigger economic numbers that help contextualize where we actually are with AI.

Current annualized revenue from general-purpose AI tools: around $100 billion globally. That's real money. But it's a tiny fraction of the global economy.

AI's share of US GDP: roughly 0.5%. For context, healthcare is about 18% of US GDP. Construction is about 4%. AI is real and growing, but we're not living in an AI-dominated economy yet.

Labor market effects: Greenblatt doesn't see large, widespread impacts yet, though he notes that junior software engineering hiring has notably decreased. If AI were truly causing 5x or 10x productivity gains across the economy, we'd see massive labor market shifts. We're not seeing them because the gains are real but modest.

What does this mean for you? You haven't missed the window. We're in the early innings, not the late ones. If you're a small business in Las Cruces or El Paso (or anywhere) and you haven't started adopting AI yet, you're not hopelessly behind. You're probably right on time.

The businesses that will win aren't the ones that adopted AI first. They're the ones that adopted it thoughtfully, measured it honestly, and integrated it into workflows where it actually helps.

What to Do With This Information

Stop measuring AI productivity by asking how much longer things would take without it. That question gives you an inflated, misleading number.

Instead, ask: "If I took away all our AI tools tomorrow, how much would I need to speed up my team to maintain the same output?"

Be honest. Account for the fact that you're already doing some things you wouldn't do without AI, and that's okay. Focus on the core work you were doing before AI and are still doing now.

For most businesses, I'd guess the honest answer is somewhere between 1.2x and 1.8x. That's your real number. Use it.

When a vendor tells you their tool will "10x your productivity," translate that. They probably mean it will speed up certain specific tasks significantly (which is great), but the overall business impact will be more modest (which is still valuable).

When you're budgeting for AI tools, use realistic multipliers. A 1.5x speed-up to a $60,000/year employee is worth about $30,000/year in value. If the tool costs $10,000/year, that's probably a good investment. If it costs $40,000/year, maybe not.

And most importantly: focus on the high-leverage tasks. Don't try to apply AI everywhere. Find the repetitive, high-volume tasks where even a 3x speed-up makes a meaningful difference, and start there. Build out slowly, measure honestly, and adjust as you go.

Moving Forward

I work with businesses in Las Cruces and El Paso that are trying to figure out where AI actually fits in their operations. Most of them have tried ChatGPT or Claude. Many are using AI in at least one part of their workflow. Almost all of them are wondering if they're doing it right.

The honest answer is that AI is useful, real, and growing. But it's not magic, and it's not going to 10x your business overnight. A 1.5x to 2x improvement, applied to the right tasks, with honest measurement and realistic expectations, is genuinely valuable.

If you want to talk through what realistic AI adoption looks like for your specific business, I'm happy to help. No hype, no buzzwords, just practical advice on what works and what doesn't.

FAQ

Is AI really only a 1.6x speed-up? That seems low.

The 1.6x number is specifically for the best AI companies in the world, measuring overall productivity gains when you account for workflow changes. For most businesses, the number is probably lower (1.2x to 1.5x). But specific tasks can see much bigger gains. The key is that the overall average is more modest than the hype suggests, and that's useful information for planning.

Should I wait to adopt AI since the gains are smaller than advertised?

No. A 1.5x productivity improvement is still significant, especially when it compounds over time. The point isn't that AI isn't worth adopting. The point is that you should adopt it with realistic expectations, focus on high-leverage tasks, and measure honestly. Businesses that wait for AI to become "revolutionary" will fall behind businesses that are getting 1.5x gains right now.

How do I measure AI's actual impact on my business?

Don't measure it by asking how much longer things would take without AI. Instead, track specific metrics: time to complete tasks, cost per unit of output, error rates, customer satisfaction. Compare those metrics before and after AI adoption for the same types of work. And be honest about workflow changes. If you're now doing tasks you wouldn't have attempted before, that's a benefit, but don't count it as pure speed-up.

What kinds of tasks benefit most from AI right now?

High-volume, repetitive tasks with clear patterns. First-draft writing (emails, reports, social media). Data analysis and summarization. Image and content generation for marketing. Customer service responses. Code generation for routine programming tasks. These tasks often see 3x to 5x speed improvements. Complex strategy, creative work requiring deep expertise, high-stakes decisions, and anything requiring nuanced judgment still mostly require human-speed human work.