A legal plugin from AI company Anthropic set off a sell-off that wiped $300 billion from enterprise software stocks. Tech analyst Tae Kim went through the plugin’s public code and found each legal “skill” is only a prompt, 200 lines of instructions for the AI. Those instructions can mark up NDAs, flag deviations from standard terms and draft summaries for a person to review. The plugin is not perfect, but it proves small text instructions can replace chunks of expensive enterprise software.
That threatens how enterprise software makes money. Vendors have long priced it by the seat, meaning one paid license for each user. Seat pricing works while people are the limit on output, because more work means more people, and more people means more licenses. If output can rise without adding users, the price no longer matches how the business runs. That is what unsettled investors and corporate buyers.
Contract review shows why. Much of it is structured checking, where a reviewer compares the contract against standard terms, flags the risky areas and formats the result for approval. If a prompt can produce a credible first pass, charging more because more people open the same screen becomes harder to defend. Prices come under pressure first where the work is predictable and the first draft is most of the labor.
An AI agent, software that carries out tasks on its own, can draft and summarize many such documents at once. A small team reviews the results and signs off, so buyers judge spending by the volume of work handled and the time it takes, instead of by headcount.
Premium tools have some protection. Many hold their place because of content gathered, sourced and checked over years, and no prompt can conjure that out of thin air. Even so, the business model can weaken while the content holds. If the work becomes “instructions plus an agent”, vendors lose the ability to grow revenue by selling more seats.
Established software companies face a hard transition. They have to keep the existing product stable for customers who demand reliability, security and support. They also need workflows built for AI agents as well as people, with pricing to match. The same engineers have to do both, and the risk of getting it wrong rises as customers start negotiating harder on renewals.
