This article is a deep-dive from JudyAI Lab — an AI engineering playbook series with 100+ published guides, 5,000+ weekly readers across 60+ countries, focused on the practical side of running AI agents, trading systems, and content pipelines in production.
McKinsey’s 2025 report dropped an awkward number — close to 90% of companies have adopted AI, but 94% of them feel they “haven’t seen significant value” [Source: https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/where-ai-will-create-value-and-where-it-wont].
But the same set of research from McKinsey also keeps saying that AI adoption could add up to 3.4 percentage points of annual productivity growth [Source: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/the-economic-potential-of-generative-ai-the-next-productivity-frontier].
On one side, tools that are hyped to death. On the other, ninety percent of companies can’t make them work. So what’s going on in the middle?
For solo founders, this hits closer to home. You don’t have a “company” to restructure workflows for AI — you are the workflow.
The Research Side: AI Really Does Work, and There’s Controlled Studies to Prove It
That frequently cited MIT study from 2023 [Source: https://news.mit.edu/2023/study-finds-chatgpt-boosts-worker-productivity-writing-0714] put 453 white-collar workers through writing tests. The conclusion was clean: groups using ChatGPT wrote 40% faster and produced 18% better quality output. The full paper is in Science [Source: https://www.science.org/doi/10.1126/science.adh2586].
That’s not marketing hype. It’s a controlled experiment.
But there’s a detail from that study that often gets glossed over — the participants were using ChatGPT-3.5 at the time, which is several generations behind today’s mainstream Claude 4 / GPT-4 family. Follow-up comparative studies at similar scale are still rare, but the tool’s capability ceiling is clearly a whole level higher than where it was in 2023.
McKinsey’s 2025 business-building report added another angle: the time for new ventures to reach significant revenue shrank from 38 months in 2023 to 31 months in 2025 [Source: https://www.mckinsey.com/capabilities/business-building/our-insights/how-to-build-businesses-faster-and-better-with-ai]. A full seven-month difference, compressed into two years.
The tools really do work. Let’s get that settled first.
So Why Can’t 90% of Companies Extract Value? The Key Is “Process Redesign”
This is where it gets interesting.
McKinsey’s conclusion is actually counter-intuitive — “AI’s productivity gains come from process redesign, not just plugging AI into existing workflows” [Source: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work].
Here’s what that means. If you just slot ChatGPT into your current “first draft → revise → publish” workflow, the improvement is limited. But if you redesign the whole workflow into “AI writes first draft → human only checks viewpoints and facts → tone check → publish,” that’s where the 40% comes from.
Big companies can’t pull off process redesign because it requires coordinating dozens of departments, changing SOPs, training people, and navigating political landmines.
Here’s where solo founders have the edge — there’s no one blocking you. You alone can rebuild your entire workflow from scratch. Taking McKinsey’s “process redesign is the key” conclusion and applying it to solo founders basically means: you lack the internal friction of big companies, which puts you in the best position to actually get AI right.
But One Thing Tools Won’t Tell You: First Figure Out What to Keep
Tools help you do already‑clarified tasks faster. They can’t help you figure out what you actually want.
A lot of solo founders get stuck in this loop: subscribe to five AI tools, try each one briefly, and end up concluding “AI doesn’t work.”
The problem isn’t the tool — it’s not deciding first what are the things only you can do, that you want to keep? Everything else goes to AI.
Get the sequence wrong, and even the most powerful tool won’t save you.
Three Criteria for Solo Founders
By this point, here’s three takeaways for you.
First, map your workflow before picking tools. If you can’t list what you actually do every day, buying more AI subscriptions just becomes a collection. Track your time for a week, break down what each block does, then decide what’s worth delegating.
Second, go deep on one or two — don’t sample everything. MIT’s 40% boost came from a controlled experiment using a single writing task; to amplify that into your daily workflow, you need long-term usage so you know the tool’s boundaries. Sampling five beats deep-diving two every time. My own process at JudyAI Lab came from exactly this — use one tool for half a year before you know its real ceiling.
Third, preserve what only you can do. AI saves you time, but what you do with that saved time determines whether it’s worth anything. If AI saves you two hours on copywriting and you spend those two hours writing more copy, that’s not a savings. This connects to what AI night shift workflow tries to solve — it’s not about how much AI can do, it’s about whether you can redirect the time AI frees up toward things only you can do.
Most of McKinsey’s 94% who didn’t see value fell into the third trap.
Tools work. Whether you’re using them meaningfully is a whole different story.
FAQ
Q1: Is MIT’s 40% study still relevant?
The study itself used ChatGPT-3.5 in 2023, but model capabilities have improved massively in these two years — most practical observations suggest the gains are even bigger now. What you should watch out for isn’t outdated data — it’s that 40% came from a controlled single writing task. To scale that to “a whole workday,” you need process redesign.
Q2: Should solo founders learn AI tools first or build products first?
Build products first, and weave AI in as you go. Without a concrete task as a vehicle, learning tools becomes collecting. The fastest way is “this week I need to do X — let’s see how much AI can help,” not “let me master the tool first.”
Q3: How many AI tool subscriptions are enough?
Two is plenty: one conversational (Claude or ChatGPT), one vertical (depends on your business — coding gets Cursor, design gets Figma AI, data analysis gets either). Go deep for six months before adding a third.
Q4: How do I know if I’m using AI right?
Simple check — has the time I saved turned into things only I can do? If I save two hours and spend them on something AI could also do, I haven’t saved anything. That’s the loop most of McKinsey’s 94% who couldn’t extract value got stuck in.
Q5: How do I actually redesign my workflow?
Start with one task. Pick something you do every week (like writing weekly reports, replying to clients, organizing meeting notes), break it into 5 steps, and ask for each step “can AI do this?” Delegate what you can. What’s left is “what only I can do.” Get one task optimized, then move to the next.