Every business that buys legal help, marketing, research, or consulting should pay attention to a quiet shift happening in how that work gets priced and delivered. Generative AI has made it cheap enough for clients to do in-house what they used to pay outside experts to do. And the outside experts are now scrambling to figure out how to stay needed.
The clearest example is happening in law. Goldman Sachs, Morgan Stanley, and Citi have started telling their outside law firms that fees should come down, because AI has made routine legal work faster to produce. This matters because the entire law firm business model runs on billing by the hour. If the hours needed to do a task drop, and the price does not, clients notice fast.
This is not actually a new fight. Companies have swung between doing work in-house and paying outside firms for decades, most notably in the 1990s when cheaper communication and software made it worth sending back-office work out. What is different now is the direction: AI is often pulling work back inside the company, not sending it out.
The interesting part is how the smartest service providers are responding, and it is not by arguing their staff is smarter than a client's AI tools. That argument is losing. Instead, firms are turning their expertise into a product a client can buy instantly, rather than a service they have to commission and wait for.
Moody's is a good example. Instead of waiting for a bank to call and ask about a company's credit risk, Moody's now delivers its ratings and research directly inside the Excel spreadsheets that banks already use through Microsoft's AI tools. The answer that used to take a phone call and a day now shows up in the same window the client was already working in.
A law firm, A&O Shearman, took this further. Working with an AI company called Harvey, it packaged its senior lawyers' reasoning on tasks like reviewing loan documents into AI agents, and now sells access to them by subscription, including to other law firms. A law firm is now also a software vendor. That is a genuinely new business model for an industry that has sold hours since it existed.
None of this means bringing work in-house is free. Someone still has to check the AI's output, fix mistakes, and keep it compliant, and those costs pile up quietly over time. That is exactly the opening providers like Thomson Reuters are chasing, betting that clients will pay to avoid owning that headache themselves.
For any company that buys these services, the practical move is simple: ask your provider directly whether AI has already lowered their cost to deliver, and if it has, ask why your invoice has not changed. The firms that answer with a cheaper, faster, fully managed option will keep your business. The ones that only talk about the quality of their people are the ones to watch closely over the next year.