Working With a Patent Attorney for the First Time

I assumed a patent worked like this: you describe the invention, you file it, you wait, you get a patent. That is not how it works. Months after we filed, we got an "office action," which is the examiner's formal response. Ours was a first, non-final office action, which sounds like a rejection and basically is one. Most applications get rejected on the first pass. I learned quickly that filing is the easy part. The real work starts when the examiner writes back.

Getting more of our claims onto the table

Our application had sixteen claims, but the examiner initially engaged with only the first four. The rest were sitting there, unexamined and unprotected.

Working with my attorney, we used a deliberate strategy: we folded the language of claim 1 into claims 5 through 8 and 13 through 16. It worked. The examiner expanded what they would consider to claims 1-8 and 13-16, and only set aside claims 9-12. That was the moment I understood something I had never thought about. How you structure and connect your claims decides how much of your invention actually gets protected. Coverage is not handed to you. You engineer it.

"Would be allowable if..."

The best sentence in the whole document said our claims would be allowable if we corrected two kinds of issues. That mattered more than it looked. It meant the examiner was not arguing that the invention was old or obvious. The fight was about language, not substance. Encouraging, and also a trap, because language is exactly where it is easy to be lazy.

The challenge I did not see coming: vague words

The examiner flagged what is called "relative terminology." Words like higher prior probability, diagnostically relevant, commonly affected, and high-gradient.

In plain English, I had used comparative words without defining the yardstick. Higher than what? Relevant by whose standard? In patent law a term is "indefinite" if there is no stated way to measure it, and indefinite claims do not hold up. So we replaced every soft word with a hard one. "Higher prior probability" became a defined probability threshold. "Commonly affected" became a specific frequency. "High-gradient" became a measurable intensity change, computed a specific way. Every fuzzy word had to become something you could actually put a number on.

The challenge that was tedious but mattered: antecedent basis

The second batch of issues was grammatical, and it showed me how exacting this world really is. In US patent practice, the first time you mention something it has to be "a" or "an," and every time after that it is "the" or "said." I had written "the masked image" before I had ever introduced "a masked image." Tiny. Pedantic. And enough to hold up a patent. We went through it claim by claim, term by term, fixing every instance.

Even the drawings got flagged. Some of our figures were in color, and they had to be redone in black and white. A reminder that the rules are the rules, all the way down.

The part I did not expect: the collaboration

The back-and-forth with my attorney was the most rigorous editing I have ever been part of. We argued over whether a "reward signal" should be singular or plural in one claim versus another, because it depended on whether the system was computing several signals or one combined one, and we had to point to the exact paragraphs in the original filing that supported each reading. My attorney caught ambiguities the examiner had missed and pushed to fix them anyway, on the logic that a claim which is clear today is a claim that is harder to attack tomorrow.

The hardest rule to absorb was this: you cannot add new ideas at this stage. You can only sharpen what you already disclosed. Every amendment had to be supported by language that was already in the application. If it was not in there originally, it was not coming in now.

What it taught me

My day job is reducing ambiguity, turning messy risk into a decision someone can stand behind. The patent process was that same instinct turned up to eleven. Precision is not pedantry. A vague word in a security policy, a contract, or a patent claim is a gap that someone can drive through later. The whole experience forced me to define out loud the things I had been comfortable leaving implicit.

After we filed our response, the application was allowed: a set of claims covering an AI system that detects disease in 3D medical images while protecting patient data. A second application is still working its way through examination.

I am proud of the invention. But the lesson I carry from it is about language. If you cannot define it, you cannot defend it.