An IPO is one of the few projects where a company’s management team and its lawyers are both on the critical path at the same time. Thousands of parallel tasks, most of them document-heavy, all of them consequential. Cooley’s answer, per OpenAI’s writeup published September 17, 2026, is a proprietary product called GO Public, built on ChatGPT Work.
What GO Public actually is
The firm’s own framing is worth reading closely. David Wang, Cooley’s Chief Innovation Officer, describes GO Public as an offering built on ChatGPT Work whose agentic harness analyzes information so lawyers can review and validate the work. The output is a tailored starting point for an IPO, not a finished filing.
That distinction matters more than the model choice. Before GO Public, the supplied material says teams typically started from a precedent at a comparable company and adapted it. GO Public starts from the client itself, combining information the client provides, relevant public sources, and curated precedents.
The harness is the product decision
Wang’s description of the harness is the part I’d steal. It lays out which steps agents can perform automatically, where lawyers must review or validate, and how the pieces come together. Information curated from previous analyses becomes what he calls baked-in know-how.
That is a workflow contract, not a prompt. It answers the question every regulated-industry builder eventually hits: where does the agent stop and the accountable human start? Cooley’s legal engineers, innovation counsel, and practitioners built it together, translating the firm’s capital markets experience into the system rather than bolting AI onto an existing checklist.
If you are designing something similar, the closest earlier post on this blog is how Anthropic’s consumer data opt-in changes what you build on Claude — same underlying concern about where professional accountability sits once a model is in the loop.
Speed to quality, not speed to output
Dave Peinsipp, a partner and co-chair of Cooley’s global capital markets group, draws a line the marketing usually blurs. The point, he says, is not doing the same work faster. It is reaching a strong starting point sooner so lawyers spend more time applying judgment, challenging disclosure, and thinking strategically.
Wang puts the same idea in operational terms: once the first cut of a large information set is handled, human effort concentrates on the highest-value surface areas. For management teams, the claimed benefit is attention returned to the business and the offering story.
Note what is and is not claimed. The supplied material does not give cycle-time numbers, error rates, or headcount effects. It also does not say how the harness handles a novel transaction type. Those are the questions I would ask before treating this as a template.
What a builder should take from it
Three things transfer beyond law.
First, the differentiator is the curated corpus plus the review boundaries, not access to a frontier model. Cooley’s decade of issuer-side IPO work is the asset; ChatGPT Work is the runtime.
Second, the product is deliberately incomplete. It produces a starting point for expert review. That is a safer shape than an autonomous pipeline in any domain where a human signature carries liability.
Third, the stated ambition is broader than IPOs. Peinsipp calls GO Public a vision for capital markets practice generally. Whether that holds depends on whether the harness generalizes or was quietly tuned to one deal type — and the supplied material does not say.
If you are scoping an agentic product for expert work, start by writing down the review boundaries before you write the prompts. Cooley’s harness is the part worth copying; the model underneath is replaceable.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.