What AI changed between in-house teams and their outside counsel

Panel 2 was supposed to be about expectations. It became about trust within the first two minutes, and stayed there.

The premise was straightforward enough. In-house teams and their outside counsel are redefining what each expects of the other, and AI is the reason. The in-house voice on the panel took the question from the client side. Andreas Lucke, German and European Patent Attorney and partner at Boehmert & Boehmert, took it from the firm side. Neither described a technology problem.

Trust was already the question. AI moved it.

The problem predates AI. The old concern was whether the partner had done the work personally, or passed it to a candidate and signed it off without a real review. The signs were familiar: elegant claims, and problems buried deep in the description.

AI did not remove that concern. It made it harder to see. The output now arrives polished, well phrased, and plausible on first reading. The problems sit further down. The case the head of IP worries about most is a technical objective written so that it already contains part of the solution, or points toward it. That is a small flaw on the page and a serious one years later.

He sorts firms into three groups. There are those who say they do not use AI, and given the age profile of their people, he thinks many of them are telling the truth. There are those who say they use it, name the tools, invite questions, and price accordingly. And there is the large middle, which says nothing at all and simply delivers the file.

That middle is where the cost lands. The in-house team now spends more time checking work it used to accept, and pays the same fee for it. Asking directly does not help much either. Voeckel finds that outside counsel are uncomfortable with the question, and that the conversation gets harder still when the follow-up is whether the output was checked, given that the writing style changed.

The number in the room

On what AI should be used for, the head of IP was unambiguous. It should improve quality, not save time. His team builds playbooks rather than prompts, closer to the way attorneys handle standard clauses in a license or R&D agreement, then tests outside counsel's work against them. They run the same check on text a human wrote. And they are explicit that no fee reduction is being asked for in return.

Then the room made it sharper. One participant put it directly: when AI vendors sell a 70% time reduction, that expectation makes trust harder to build and the conversation about fees close to impossible.

The honest answer is that the range is real and wide. François-Xavier Leduc (DeepIP CEO) shared that clients report anywhere between 15% and 70%, and the spread is explained by the use case and by how proficient the user is. What a single headline number leaves out is what happens next. Time recovered on one task tends to be reinvested somewhere else, usually into better work downstream, which is the conclusion Panel 1 reached from the other direction. The technology returns the time. It does not decide what the time is for.

What actually repairs it

Asked what a firm would have to do to earn the trust back, three steps and no hedging were given.

1. Tell the client you use AI. Openly, including for the unglamorous parts, the emails, the processing of prior art.

2. Then set expectations in both directions: efficiency and quality both improve, and once in a while something will slip through. Do not present yourself as the one firm that catches every hallucination. Every head of IP has the same problem inside their own department, with human drafts as much as machine ones.

3. Third, talk about money. Ask what you are permitted to use, and whether the client expects fees to move.

Lucke described the same relationship from inside a firm, and said he had hesitated to raise trust at all because it sounded old-fashioned. Mid-size clients rely on him completely. They do not read the claims and often could not. They rely on him to keep them out of trouble. What that rules out is handing over something he has not thought through because a machine produced it and it left his desk. He has also become selective about who he works with, for a reason worth sitting with: in the past, someone taking a shortcut would have been visible in the work.

Both rejected the idea that AI levels expertise. Lucke gave the cleanest formulation anyone offered all evening. If you cannot draft an application or file an opposition without AI, you cannot do it with AI. There is far more work you can get through, which makes judgment more decisive rather than less. The process only holds up if it moves step by step, and every step is a decision. What AI is good for, in his description, is arguing back. You can test your own reasoning against it, the way you would with a sparring partner.

He was equally precise about where it earns its place. Reading large volumes of documents for a concept rather than a keyword. Finding the disclosure that suggests something a court would treat as obvious, which otherwise costs hours of looking. Making sense of prior art that is badly written and worse translated, where the document is all you have. Ask it a legal question, though, whether a claim is genuinely novel, and it makes mistakes.

That is also where both drew the line on collaboration. The client side does not want to co-draft section by section. He wants a complete claim set to review, then a complete application with figures. Lucke agreed from the other side: one person on one document, and the worst outcome is both parties generating long files and passing them back and forth. Collaboration belongs on strategy. Production does not improve by committee.

Which leaves the question of why in-house teams use firms at all, once the knowledge gap closes. Answer was capacity, not expertise, with something near an even split between internal and external work as a matter of risk. People leave. People fall ill. Deadlines do not move.

Matthias Hofmann offered the sentence the profession will end up using. Two years ago large clients were instructing their representatives not to use AI. A few months ago, one of the largest sent that same question to its European firms, and the one that answered no made the client angry. Telling your attorney which tools to avoid is starting to look like telling a physician which instruments to pick up.