September 2025
Our September meetup brought together an eclectic group around the intersection of IP law and AI - patent attorneys with 30+ years at a Big Law firm, US patent lawyers from boutique IP firms, in-house counsel from cross-border tech companies with a decade of patent experience, and startup founders building IP legal AI products, alongside engineers from major legal tech teams.
The mix of perspectives sparked some genuinely interesting discussions around "IP Law × AI."
01 | The Biggest Pain Point: Not Being Allowed to Use AI
I opened with a question: What do US patent attorneys see as the task they most wish AI could handle?
A patent attorney from a Big Law firm responded bluntly: "The biggest pain point is that we're not allowed to use AI." 😭
When lawyers try to experiment with AI tools, the most common resistance comes from data confidentiality requirements. Even though patent information is public, concerns persist.
Patent search behavior itself could reveal strategic direction.
Some firm systems directly display warnings: "Do not enter any client information into AI tools."
Some firms only permit AI use within self-built systems or IT-approved platforms.
02 | AI Accuracy? Where Are the Boundaries of Large Models?
Lawyers repeatedly raised concerns about AI's instability and accuracy in actual legal work:
For specific case analysis or compliance judgments, AI produces "hallucinatory outputs."
When using certain well-known legal databases with built-in AI for research, recommended cases are often irrelevant.
Some tested the same question at different times and received completely contradictory answers.
Especially for complex reasoning, AI struggles to replace human judgment.
03 | AI in Patent Practice
Lawyers concluded that the IP domain - with its relatively standardized, structured data - is where AI is "easier to deploy." But effectiveness varies by scenario:
✅ Well-suited AI tasks:
- Patent drafting and large-scale prior art searches
- Initial Claim Chart generation
- Hot document identification in patent litigation e-discovery
- Deposition summarization
⚠️ Requires human intervention:
- Claim strategy design based on existing patents
- Multi-patent portfolio analysis and valuation
- Overall litigation strategy construction and judgment
One patent attorney noted candidly: "Most of a lawyer's work is actually coordinating across different project timelines and managing client communication - areas where AI can barely help. These rely on the lawyer's own analysis and judgment."
04 | Startup Technical Paths and Market Positioning
The meetup also included founders building Legal AI products, who shared their technical implementation choices and market considerations:
Most companies use large model APIs (like OpenAI) rather than training custom models, primarily due to cost-effectiveness considerations.
Architecturally, there's stronger emphasis on RAG (Retrieval-Augmented Generation) systems to improve output relevance and stability.
For patent search, they use dual mechanisms of keyword + semantic search, supporting bilingual Chinese-English, classification restrictions, and other advanced search logic.
Market positioning sits between "traditional outsourcing services" (like Indian teams) and "premium US law firm services," aiming to establish cost-performance advantages.
One founder also emphasized: "Truly effective AI services hinge on controlling 'data source quality' and 'feedback loops,' not just the model itself."
In reality, many lawyers and firms worry that model training generates more concerns and alerts, fearing their client data might be used for training by others.
05 | Longer-Term Change: Not Replacement, But Restructuring
One lawyer proposed that AI's impact on the legal industry might unfold in three phases:
Task automation: Highly repetitive work gets replaced first.
Assisted judgment: AI participates in strategic discussions as a verification tool.
Business model transformation: AI drives service standardization, pricing logic shifts, and client relationship management changes.
Particularly in patent domains - given high data structuring and automatable components - many believe this will become a key breakthrough point for AI implementation.
Some also mentioned: "AI's greatest commercial value" might manifest in second-tier areas like patent prosecution, NPE responses, and inefficiency searches.
💬 Final Thoughts
If you're also interested in Legal AI practical applications, feel free to connect.
I'm Helen, organizing monthly legal tech coffee chats for this community. I also regularly share product insights and trend observations on LinkedIn.
Current discussion topics include: legal professionals in Web3, Immigration Law × AI... We keep groups intentionally small to ensure everyone can meaningfully contribute.
💬 Welcome to join our conversations.