A Top IPO Firm Turned Its Capital Markets Playbook Into Software

Cooley, one of the most active law firms in the US IPO market, has built a proprietary offering called GO Public on top of ChatGPT Work. According to a case study published by OpenAI, the product wraps an agentic system in what the firm calls a harness: a controlled workflow that specifies which steps agents may perform on their own, where lawyers must review and validate, and how the pieces assemble into a starting point for an IPO. It is a different pattern from buying legal software — the firm is productizing its own accumulated judgment.

The firm and the claim

Cooley says it advised on 180 deals globally in 2025, totaling more than $51.5 billion in deal volume, and describes a decade-long run at the top of the US issuer-side IPO market, having advised on more venture-backed IPOs than any other firm over the past 20-plus years. That track record is the asset GO Public is built to exploit. As David Wang, the firm's chief innovation officer, puts it in the case study, everyone understands that the leading capital markets firm knows what to do in an IPO; the project was about carrying that forward into a system.

The mechanism is a change in where the work begins. Before GO Public, teams typically started from a precedent filing for a comparable company and adapted it to the client's circumstances. Cooley says GO Public instead starts with the client itself, pulling together information the company provides, relevant public sources and curated precedents to assemble a more tailored first draft for lawyer review. Wang describes information curated from previous analyses as "baked-in know-how" inside the harness.

Dave Peinsipp, a partner and co-chair of Cooley's global capital markets group, frames the goal as "speed to quality" — not doing the same work faster, but arriving at a strong starting point sooner so lawyers spend more time applying judgment and challenging the disclosure. He also makes the client-side argument: IPO preparation competes for the attention of executives who are simultaneously running the business, and moving faster through preparation returns time to management.

What is genuinely new here

The interesting part is not that a law firm is using a language model. It is the shape of the artifact. GO Public is described as a proprietary product offering, built by the firm's legal engineers, innovation counsel and practitioners working with OpenAI's engineering team. Cooley is not consuming a vendor's legal AI product; it is assembling one and putting its own brand on the output.

That inverts the usual supplier relationship. Legal technology vendors have historically captured value by encoding workflow and selling it back to firms. Here the firm encodes the workflow itself and treats the model as infrastructure. If that pattern spreads, the firms with the deepest precedent libraries and the most repeatable practice areas have a structural advantage they can now express in software rather than only in headcount.

The harness concept is the operationally serious piece. Cooley describes it as laying out which steps agents can perform automatically, where lawyers must review or validate, and how it all comes together. For work of this consequence, that boundary is the entire safety argument. Wang is explicit that attorneys carry professional duties and obligations to reach the most legally defensible outcome, and the design keeps validation in human hands rather than treating review as optional.

What the case study does not say

This is a vendor-published customer story, and it is notably light on measurement. No time savings are quantified. No number of deals run through GO Public is given. There is no before-and-after on drafting cycles, no error rate, no indication of how many lawyer-hours a filing prepared this way consumes versus the traditional path. The only figures in the piece describe Cooley's practice generally — deal count and deal volume for 2025 — not the product's effect.

That absence is worth naming rather than filling in. "Speed to quality" is a well-constructed phrase precisely because it declines to commit to a speed claim. It is possible the firm has internal numbers it is not ready to publish, and it is possible the deployment is early enough that clean numbers do not exist yet. Either way, a reader should treat this as a description of an approach, not evidence of an outcome.

Several other questions go unaddressed. The case study does not describe how material nonpublic information is handled inside the system, which is the first thing any issuer's board will ask about a tool touching a draft registration statement. It does not say whether clients are told when GO Public is used on their deal, or whether they can decline. It does not describe what happens on review — whether lawyers are catching material errors, how often, and how those corrections feed back into the curated know-how. And it says nothing about pricing, which is the question underneath all of this.

The billing question nobody raises

A firm that compresses preparation work faces an obvious economic tension. If GO Public shortens intensive preparation and the firm bills for time, the firm has built a tool that reduces its own revenue on a matter. There are reasonable resolutions — fixed-fee or value-based pricing, capacity redeployed to more deals, or margin retained by charging the same and spending less — but the case study does not indicate which applies. Peinsipp's argument is about where lawyer attention goes rather than what it costs, and Wang's is about professional defensibility. Neither addresses whether clients see a lower bill.

This is not a Cooley-specific evasion; it is the unresolved question across the profession's AI adoption. It is also the one that will determine how fast the pattern spreads. Firms adopt tools that protect margin much faster than tools that compress it.

Peinsipp positions GO Public as the beginning of a broader shift, seeing potential not only for IPOs but for capital markets transactions more generally. That is a reasonable extrapolation — the work is document-heavy, precedent-driven and repeatable, which is the profile that suits this approach. It is also, for now, a statement of intent.

What to watch

• Whether Cooley or its clients publish any measured outcome — cycle time, cost, or error rates — rather than descriptions of the approach.
• How material nonpublic information and issuer consent are handled, and whether that gets documented publicly as more deals run through the system.
• Whether the firm-builds-its-own-product pattern spreads to peer firms with comparable precedent libraries, or stays confined to a handful with in-house engineering teams.
• What happens to legal technology vendors selling capital markets workflow if the largest firms build their own.
• Whether pricing models shift in response, since time-based billing and preparation compression pull against each other.