Kilpatrick's Alton Absher, Charles Gray, Tyler McAllister, Patrick Njeim, and Tra Unger, joined by Spencer Thevenin
of Dish, presented “Bring Us Your Hardest IP Problems: An AI-Powered Partnership for Building What Comes Next”
at the 21st annual KTIPS (Kilpatrick Townsend Intellectual Property Seminar). The session opened with a live
audience survey repeating questions first asked a year earlier, turned to the professional responsibility framework
now governing a lawyer's use of Generative AI, and then demonstrated six agentic tools the firm built in-house,
running from invention intake and quality assurance through portfolio analysis and the assertion lifecycle, from
developing a demand letter to preparing for oral argument.
The panel provides these key takeaways:
1. The lawyer, not the tool, remains accountable. Competence under Model Rule 1.1 and its
Comment 8 now reaches the benefits and risks of relevant technology, and the USPTO Rules
of Professional Conduct mirror that duty. Ethics authorities treat Generative AI as a
nonlawyer assistant subject to the supervisory obligations in Model Rules 5.1 and 5.3, and
guidance issued in 2026 scales that obligation to the autonomy of the system. In practice,
what courts have sanctioned is not the use of AI. It is the failure to verify, and the failure of
candor once an error surfaces.
2. Practitioner use has moved from casual to committed, and from convenience to
substance. Our 2025 and 2026 survey results reveal an evolution in how legal professionals
leverage Generative AI in patent prosecution. We are witnessing a clear transition from
casual experimentation to daily, deeply integrated reliance. Practitioner usage has matured
beyond basic entry-level tasks (such as search and summarization) toward substantive
workflows, including novelty evaluations, quality checking, and document drafting.
Furthermore, the motivation behind adoption has shifted from a fear of being left behind to
top-down organizational mandates focused on turnaround speed and enhanced work
product quality. Ultimately, Generative AI is evolving from an ad-hoc efficiency hack into an
integrated component of modern patent practice.
3. For work that is specific to a practice, building beats buying. The firm's AI development
group is creating tools that span the IP lifecycle, all of them developed internally, to close the
gap between what practitioners need day to day and what the available products actually do.
The reason to build is narrow and worth stating plainly: the tooling has to understand
prosecution practice, and general-purpose products do not. Each tool starts from a workflow
the firm already runs and is built around it. The transferable question for an in-house team is
not build versus buy in the abstract, but which parts of your workflow are specific enough
that a generic tool cannot know them.
4. The step change is agentic workflows, not chat. What was demonstrated were multi-step
systems that chain tools together and return a work product, such as an assertion workup, a
portfolio analysis, or a ranked set of predicted questions from the bench, rather than a chat
window returning prose. What makes them usable is where they stop. Each workflow pauses
at defined points and requires a lawyer to approve before anything consequential happens.
When evaluating any tool, ask what it will not do on its own, not only what it can produce.
5. Put the guardrails in the build system, not only in a policy document. Every tool goes
through a written development lifecycle approved by information security: no engineer
releases their own work, matter content runs only where nothing is retained and nothing
trains an outside model, the system may read but cannot send or file without a person, and
each tool is scoped, inventoried, and monitored. The tools also report their own limits,
labeling a conclusion low confidence where the underlying record is thin. A policy depends
on everyone following it every time. A build system does not.
For more information, please contact:
Alton Absher, aabsher@ktslaw.com
Charles Gray, cgray@ktslaw.com
Tyler McAllister, tmcallister@ktslaw.com
Patrick Njeim, pnjeim@ktslaw.com
Tra Unger, tunger@ktslaw.com
