Innovation and AI: where it helps, and where it quietly lies to you
Quick answer: "Innovation ai" usually means using AI to help you build something new. The useful rule is simple: AI is excellent at the generative half of innovation, producing options, synthesising research, drafting, restructuring a messy problem. It is unreliable on the judgment half, deciding whether the market actually cares. A Harvard Business School field experiment with Boston Consulting Group found that on tasks inside AI's range, people did 12.2% more work, 25.1% faster, at higher quality; on a task just outside it, they were 19% less likely to be correct, and couldn't feel where the line was. So use AI to generate. Never let it deliver the verdict.
Type your idea into any AI and it will help you innovate for about ninety seconds.
It will restructure your thinking, list ten adjacent ideas, and draft a landing page. Genuinely useful. Then, somewhere without warning, it will start making things up, politely, confidently, in the same fluent voice, and you will not be able to feel the moment it crossed the line.
That line has a name, and it is the most useful thing I know about building with AI.
The jagged frontier
Researchers at Harvard Business School, working with Boston Consulting Group, ran a field experiment with 758 consultants. Some worked with GPT-4, some without. On eighteen realistic tasks that sat inside what the model was good at, the AI group crushed it: 12.2% more tasks completed, 25.1% faster, and noticeably higher quality.
Then they gave people a task designed to sit outside the model's competence, the kind of messy judgment call where the right answer isn't in the pattern. The people using AI were 19% less likely to get it right than the people working alone.
They called it the "jagged technological frontier." AI capability isn't a smooth hill you climb. It's a jagged edge. On one side the tool makes you sharply better. Cross it, and the terrifying part is you can't see when you do, and the tool makes you confidently worse, because it hands you a fluent, plausible, wrong answer and you have no reason to doubt it.
Innovation is a workflow that crosses that edge constantly.
Where AI genuinely moves innovation forward
This is the good news, and it is real.
- Generating options. Ask for thirty ways to approach a problem and you will get a few you would never have reached alone. AI is a brilliant divergent-thinking partner.
- Research synthesis. Dumping messy notes, a market, a pile of transcripts, and getting back the shape of it: patterns, tensions, the questions you haven't asked. Fast and mostly good.
- Drafting and restructuring. First versions of anything: a landing page, a survey, an interview guide, a positioning statement. A blank page is expensive; AI makes it cheap.
- Finding the shape of a problem. Talking through a vague itch until it becomes a sharp, testable assumption.
Every one of these is the generative half of innovation. More options, faster, at lower cost. If that were the whole job, AI would have innovation solved.
It isn't the whole job.
Where AI quietly makes it worse
The other half of innovation is judgment. Is this problem painful enough that people will change what they do? Is this enthusiasm real or just politeness? Of these ten options, which one is worth six months of my life?
This is exactly the terrain on the far side of the frontier, and AI does not go quiet there. It stays just as fluent and just as confident while being wrong. Worse, it is trained to be agreeable. Ask most AI tools whether your idea is good and you will get a warm, structured, well-reasoned yes. That yes feels like validation. It is the opposite. It is the same trap as showing your idea to friends who like you, except the friend never gets tired and never says the awkward thing.
I have watched people do this, and I have done it myself. They take an AI's encouraging summary of their idea and treat it as a signal. It is not a signal. It is autocomplete with good manners.
This is not my first company. The last one ended in a sale, and if I'm honest there was more improvisation in getting there than I'd put on a pitch call. It worked out, partly on judgment and partly on luck I couldn't repeat on demand. I don't want to run the next one on instinct, and I am not going to swap my instinct for a machine's instinct just because the machine writes more confidently than I do.
How to use AI to innovate without fooling yourself
You don't have to choose between "AI does everything" and "ignore AI." You just have to keep it on the right side of the frontier.
Generate wide with AI, then leave the building. Use it to produce more and better options than you could alone. That's its superpower. Then stop asking it what's true.
Take the best option to real humans. Find ten people who actually have the problem and ask them about the last time it happened, not whether they'd use your thing. Applause is free; the last-time story is data. AI can't manufacture this for you, and it shouldn't try.
Make the AI argue against you. Flip the prompt. "Here is my idea. Give me the strongest case that it will fail, and the three assumptions most likely to be wrong." Used this way, AI is genuinely valuable on the judgment half, as a sparring partner, not a cheerleader.
Keep the verdict human. The decision to commit, whether to build, pivot, or walk, stays with you, informed by evidence you collected from people, not by a paragraph a model wrote to please you.
If you want the longer version of how that testing loop works, I wrote it up in the honest guide to validating a startup idea. And if you're specifically wrestling with the "we use AI, so we must be innovative" story, that's its own trap. More on it in AI innovation is not the hard part.
The part I actually care about
This exact frontier is why we built Ventropolis as a validation engine. It's AI pointed the other way: not there to be nice about your idea, but to test it against real evidence, ask the question you're avoiding, read your customer conversations, and tell you where the signal is thin. AI is at its best in innovation when it is trying to save you from yourself, not sell you back your own optimism. If you want to put an idea through that before you build it, try Ventropolis.
So before you let a model tell you your idea is brilliant: which half of the work did it just do, generating, or deciding?
And if it was deciding, who checked?
