OpenAI Just Built a Lawyer. Marketing Still Ships Generic Agents.

OpenAI did not ship a prettier chatbot this week.

They shipped a vertical.

On September 17, OpenAI launched Astra for Law. GPT-6 Astra wrapped in a legal search index of more than 230 million URLs, plus instructions for legal analysis and writing. On the Vals AI legal research bench it scored 54 percent. The same model with web search only scored 38.7 percent. That 15-point jump is the whole story.

The model did not get smarter overnight. The stack got specific.

What actually shipped

Astra for Law is not an AI lawyer. OpenAI said that out loud. It is a configured platform for firms and legal tech builders: CourtListener coverage of more than 99.9 percent of published US precedential case law, 26 partner plugins, and 47 community plugins. Harvey, Legora, Thomson Reuters, Clio, iManage, Relativity. Early access sits with Latham, Ropes and Gray, Cooley, Sullivan and Cromwell.

Same week OpenAI also pushed ChatGPT for Financial Services. The pattern is not subtle. Frontier labs are productizing verticals. General chat is the demo. The billed product is the domain stack.

The three-layer stack

• LAYER 1: the frontier model. Everyone already has this.

• LAYER 2: a domain index plus writing rules. This is where accuracy actually moves.

• LAYER 3: plugins into the tools the client already pays for. Matter files. CRM. Catalog. Permissions stay intact.

BEFORE: a smart intern with a search bar. AFTER: a junior partner that can find the right passage, flag the hostile precedent, and drop the draft into the matter file. That is the gap agencies keep ignoring.

The marketing brief nobody asked for

Here is the part that should sting. Most agencies are still selling LAYER 1. We use Claude. We use Midjourney. We use ChatGPT. Cool. So does the intern at the client.

Law just proved the product is not the model. The product is the vertical agent. If we at Dzine Prodigy ran a brand the way OpenAI just ran law, the stack would look like this:

• a private corpus of the brand book, past campaigns, offer math, and competitor claims

• writing rules that refuse generic claims and force proof, price, and a next step

• plugins into Ads Manager, HubSpot, Shopify, and the CMS so the agent ships into the workflow instead of a Google Doc

That is AI-native marketing. Not 40 variations of a headline. A named agent that already knows the category, the objections, and the brand line it is not allowed to cross.

Do this next

Pick one client vertical this week. Wrap the model in their corpus. Give the agent a name. Plug it into one live tool they already use. Ship one brief through it end to end.

If the output still reads like a general chatbot, you built LAYER 1. Law just showed us LAYER 2 is where the work starts to pay.