The legal industry just shipped an agentic system to production
Cooley, the law firm that's advised on more venture-backed IPOs than anyone else over the past two decades, built GO Public — a proprietary AI product on ChatGPT Work that fundamentally changes how IPO preparation works. Not "changes" in the press-release sense. Changes as in: they took a capital markets workflow involving thousands of manual tasks and built an agentic harness that synthesizes client information, public sources, and curated precedents into a tailored starting point for lawyer review.
In 2025 alone, Cooley advised on 180 deals totaling over $51.5 billion. When David Wang, Cooley's Chief Innovation Officer, says "when there's an IPO, there are thousands of things that need to be done constantly," he's not exaggerating. GO Public is their answer to that chaos — and it's one of the clearest examples yet of what agentic AI looks like when it ships to high-stakes production environments.
What makes this agentic (and why that matters)
GO Public isn't a chatbot wrapper. It's an agentic harness — a controlled workflow that defines which steps agents can perform automatically, where lawyers must review or validate work, and how everything synthesizes into an IPO preparation workflow.
The architecture matters because it solves the core problem with deploying LLMs in high-stakes domains: you need intelligence at scale, but you can't abdicate accountability. GO Public's harness provides what Wang calls "baked-in know-how" — information curated from previous analyses that guides agent behavior while preserving lawyer oversight at critical decision points.
This is the pattern we're seeing across production agentic systems in 2025: not autonomous agents running wild, but structured workflows where AI handles synthesis and pattern-matching while humans own judgment and validation. Cooley partnered directly with OpenAI to combine their capital markets subject-matter expertise with AI engineering to ensure "the right processes produce the right outcomes."
Speed to quality, not just speed
Dave Peinsipp, partner and co-chair of Cooley's global capital markets group, frames the value proposition clearly: "The point isn't simply to do the same work faster. It's to get to a strong starting point sooner, so our lawyers can spend more time applying judgment, challenging the disclosure and thinking strategically about the issues that matter most to the company."
Before GO Public, teams typically started with a precedent from a comparable company and adapted it. Manual, slow, and frontloaded with low-leverage synthesis work. GO Public flips this: it begins with the client itself, bringing together their information, relevant public sources, and curated precedents into a tailored starting point.
Wang's framing is worth quoting directly: "The amazing thing about integrating ChatGPT Work into GO Public is that it brings intelligence to this very large array of information that had to be manually sorted before. Once you do that first cut, you're able to really concentrate the human effort and expertise on the highest value surface areas."
This is the productivity unlock everyone's chasing: not replacing human expertise, but surfacing issues earlier so judgment gets applied where it actually matters.
Why IPOs are a perfect fit for agentic systems
IPO preparation is a fascinating target for AI because it combines high information density, time pressure, and massive downside risk. An IPO is "one of the most consequential moments in a company's life," Peinsipp notes, and "it can consume enormous management attention."
The process involves:
- Synthesizing huge volumes of company information, public filings, and comparable precedents
- Identifying disclosure issues that could derail the offering
- Coordinating across legal, finance, and executive teams under tight deadlines
- Producing legally defensible documentation where mistakes have real consequences
This isn't a domain where you can ship a chatbot and hope for the best. It's also not a domain where pure automation works — professional judgment and accountability are non-negotiable. That tension makes it a perfect use case for agentic systems: AI handles the information synthesis and pattern-matching that previously consumed lawyer hours, while humans own the strategic decisions and validation.
Peinsipp's point about giving "valuable time back to management" is crucial. IPO prep competes for executive attention while they're also running fast-moving businesses. If GO Public can compress the intensive preparation phase, management can focus on "the business, the story and the decisions that will ultimately shape the offering" — things only they can do.
The legal industry's AI inflection point
Wang's observation that "traditionally, the legal industry has been relatively change-averse" is diplomatic understatement. Legal is notoriously slow to adopt new technology, and for legitimate reasons: professional obligations, client confidentiality, regulatory constraints, and the fact that mistakes can end careers.
What makes GO Public significant is that Cooley didn't wait for the industry to change — they built the tooling themselves. Wang explains: "As attorneys, we have professional duties and obligations to make sure that the best, most legally defensible outcome occurs for our clients, and that just takes a lot of work." GO Public is designed to help lawyers "consider more information while directing their expertise toward the decisions that matter most."
This is the pattern we'll see in other risk-averse professional domains: firms with deep subject-matter expertise building proprietary agentic systems on top of foundation models, rather than waiting for generic AI tools to somehow understand their workflows.
Peinsipp sees GO Public as just the beginning: "We see enormous potential not only for IPOs but for capital markets transactions more broadly." Translation: they've validated the architecture on IPOs, and now they're thinking about where else to deploy it.
What Cooley got right
Three things stand out:
Partnership over prompting. Cooley worked directly with OpenAI to combine legal expertise with AI engineering. This isn't a "we threw some prompts at GPT-4 and called it a product" story. They built a proper agentic harness with structured workflows and validation points.
Starting with a bounded domain. IPOs are complex but well-scoped. There's a defined process, established precedents, and clear success criteria. That makes it easier to engineer the agentic workflow and validate outputs than trying to boil the ocean.
Preserving human judgment. The harness explicitly defines where lawyers review and validate work. This isn't AI replacing lawyers; it's AI handling synthesis so lawyers can focus on judgment. That's the only sustainable model in high-stakes professional services.
Open questions
Cooley hasn't published detailed metrics on GO Public's impact, which is understandable given client confidentiality. But there are obvious questions:
- How much time does GO Public actually save on a typical IPO? Wang and Peinsipp talk about redirecting lawyer time, but quantified benchmarks would be illuminating.
- What failure modes have they encountered? Any agentic system this complex will surface edge cases — how does the harness handle them?
- How much of the "baked-in know-how" is curated precedent versus learned patterns? The balance matters for transparency and auditability.
- What's the lawyer adoption curve look like? Even with clear value, changing professional workflows is hard.
These aren't criticisms — they're the natural unknowns when you're shipping something genuinely new in a complex domain.
The bottom line
GO Public matters because it's a real agentic system shipping in a high-stakes production environment, not a demo or a research prototype. Cooley took one of the most consequential (and stressful) moments in a company's lifecycle and built AI tooling that fundamentally changes how the work gets done.
The architecture — agentic harness with structured workflows and human validation — is the pattern we'll see more of as AI moves into professional services. Not chatbots, not full automation, but intelligence applied at scale with human judgment preserved where it matters.
Wang's closing point is the right one: "ChatGPT Work is changing legal work by bringing intelligence to every step of the process faster, more effectively, and more democratically." Whether that vision plays out across capital markets transactions broadly (as Peinsipp suggests) depends on how well GO Public performs in practice.
But the fact that Cooley built it, shipped it, and is publicly discussing it? That's already a signal that agentic AI has moved from labs to law firms. And the rest of the professional services world is watching.