I wrote to Jordana Goodman a few days ago. She teaches intellectual property law at Illinois Tech and her research is on inventorship, meaning who actually ends up named on a patent and who gets written out. I asked her a question I have not been able to answer on my own. If AI hands every company the same build capability, what is left to compete on.
Her answer came back on 19 August and it reframed the question better than anything I have read this year. It is worth walking through slowly, because there is a valuation consequence at the end of it that almost nobody is pricing yet.
The test every patent has to pass
To get a patent, an invention has to be non-obvious to what the law calls a person having ordinary skill in the art. Lawyers shorten it to PHOSITA. It is an imaginary competent professional in your field, and the whole system hangs off them. If your invention would have been obvious to that person, you get nothing.
So the question that decides whether your filing survives is a strange one. What does ordinary skill actually mean.
Knowledge is not skill
The lazy answer is that skill equals knowledge. Everything written down is prior art, so the sum of the published record is what our imaginary professional knows. Goodman pushes back on that, and her line for why is the best explanation of the difference I have heard:
"Though books have knowledge, this isn't the equivalent of skill. I can read a LOT of weight lifting books and still not be an Olympic athlete."
Skill is the application of knowledge, not the possession of it. Which matters enormously right now, because the thing we have just handed every company on earth is knowledge at infinite scale.
Invention is two jobs, and AI has only taken one
Goodman splits inventing into two parts.
The first is combining known things into an unknown new thing. A new filament in a lightbulb. A new engine in a car. This is what most people picture when they picture inventing. On this half, her view is direct:
"With AI as a tool, you can brainstorm a lot more ideas in a short amount of time."
And she expects that to raise the bar for obviousness.
The second part is the one she does not think AI has reached. Recognising a problem worth solving. Her example is a shoe. She has known since she was five that it is annoying to crush the heel of your shoe when you cannot be bothered to untie it. Somebody else built the spring loaded heel that solved it. She has a child who spills every drink. Somebody else invented the sippy cup. Neither is a hard problem once it is named. Naming it was the whole invention.
"People who can recognize problems worth solving are going to be incredibly valuable workers in the next decade..."
She flags that as a hypothesis rather than a finding, and I would flag it the same way. But I think she is right, and I think the money follows it faster than most boards expect.
What this does to your patents
Here is the part that lands on a balance sheet.
Obviousness is not a fixed line. It moves with what an ordinary professional can do. If ordinary professionals now sit next to a machine that generates a thousand combinations before lunch, then combination inventions become easier to call obvious. Not immediately, and not by statute. It happens quietly, through examiners and through courts, over a few years.
Three consequences follow, and they point in different directions.
Filing gets harder from here. A patent whose entire inventive contribution is a clever combination of known parts is a weaker filing in 2028 than the same disclosure was in 2022. If your roadmap assumes a steady rate of grants on incremental work, that assumption needs testing now, not at the next renewal cycle.
What you already hold gets more valuable. Anything granted with an early priority date was examined against a world where the bar was lower, and priority does not move. A portfolio filed before the bar rises is an asset that cannot be recreated later at any price. Very few finance teams have looked at their own filings this way.
Counting patents stops working as a proxy. If the obviousness bar is moving, a portfolio has to be read at claim level to know what is actually enforceable. Patent count was always a poor measure of strength. It is about to become an actively misleading one, particularly in a diligence room where the buyer has done the claim reading and the seller has not.
What this does to your people
Goodman's second skill, spotting the problem worth solving, is not a job title anywhere. There is no hiring line for it, no course that teaches it, no way to screen for it in an interview loop. It sits inside people who have spent years close enough to a real customer to be irritated by the same thing repeatedly.
Those people are usually not the ones being protected in a restructure. They are support leads, field engineers, long serving account managers, the operator who has run the same process for nine years and can tell you exactly which part of it is stupid. If problem recognition is the scarce input to invention, then a company that cuts that layer to fund its AI spend has traded its only remaining edge for a tool its competitors bought too.
Where I have landed
Every company will soon be able to build the same thing. That is close to settled. What is not settled is who knows which thing is worth building, and who owns the rights to it once it exists.
That is the whole competition now. Problem selection on one side, owned intellectual property on the other, with a shrinking gap in between where execution advantage used to live.
My thanks to Jordana Goodman for the answer, quoted here from our correspondence with her words unchanged. Her research on inventorship and attribution is worth your time if this is your area.