本站提供正體中文版。切換到正體中文本站提供简体中文版。切换到简体中文このサイトには日本語版があります。日本語で表示이 사이트는 한국어로도 제공됩니다.한국어로 보기Diese Website ist auch auf Deutsch verfügbar.Auf Deutsch ansehenEste sitio web también está disponible en español.Ver en españolQuesto sito è disponibile anche in italiano.Visualizza in italianoCe site est également disponible en français.Afficher en françaisEste site também está disponível em português.Ver em portuguêsDeze website is ook beschikbaar in het Nederlands.In het Nederlands bekijkenЭтот сайт также доступен на русском языке.Смотреть на русскомयह वेबसाइट हिन्दी में भी उपलब्ध है।हिन्दी में देखेंهذا الموقع متاح أيضًا باللغة العربية.عرض بالعربيةSitus ini juga tersedia dalam bahasa Indonesia.Lihat dalam bahasa IndonesiaBu site Türkçe olarak da mevcut.Türkçe görüntüleTa strona jest dostępna także po polsku.Wyświetl po polskuTrang web này cũng có phiên bản tiếng Việt.Xem bằng tiếng Việtاین وب‌سایت به فارسی هم در دسترس است.مشاهده به فارسی

When 80% of Drafts Aren't Written by a Patent Attorney, What's Left of This Profession?

AI2026.04

Let me confess something first: at the very moment of writing this article, this text too has AI’s fingerprints on it. All the more reason, then, for me to talk this problem through properly.

Today, roughly 80% of patent drafts are no longer typed out character by character by a patent attorney’s (a licensed patent practitioner in Taiwan) own hand. This isn’t the future tense; it’s the present continuous, unfolding right now.

When AI Writing Patents Becomes the Norm

As AI writing patents gradually becomes the norm, there’s one question that can no longer be sidestepped: what really is a patent attorney’s trump card?

The answer certainly won’t be “output.” On speed, AI will only grow more relentless, and a human can’t out-race it.

The real trump card is another, quieter ability—being able to see, the moment a claim is written down, what will happen a few years from now.

Outsiders Relax at “Allowed”; Insiders Only Start to Worry at “Allowed”

AI can produce responses with smooth logic and precise citations. An outsider sees the word “allowed” and relaxes; an insider only starts to feel uneasy once it’s allowed:

Which technical feature got quietly narrowed without anyone noticing? Which compromise made just to get past the examiner will, years later in litigation, turn into a fatal wound?

In the AI era, a patent attorney’s most valuable quality isn’t “finishing the draft,” but still being able to see that it’s “actually wrong” even when the words all look right.

AI will lay bricks, but it doesn’t know which wall can’t be torn down or which beam is straining under the load. It can help you win an allowance, but it can’t guarantee that this certificate will actually win a lawsuit. That battle-ready map of rights isn’t in the model—it’s in the patent attorney’s head.

From “the One Who Carves the Words” to “the One Who Defines the Boundaries”

The patent attorney’s role is quietly shifting: from “the one who carves the words” into “the one who defines the boundaries,” and ultimately “the one who answers for the outcome.”

So the question we should really be asking isn’t “Will AI replace patent attorneys?” It’s this:

When AI writes out a flawed scope of rights and not a single person can spot it—what got replaced, this one job, or the very capacity for “the value of judgment” itself?

That is the trump card, and it’s the one thing this era can least afford to outsource.