AI 赚钱的五层市场：从服务到后训练

Mike Fishbein @mfishbein原推发布于 2026.10.20

在 X 上打开原推

There are so many ways to make money from the AI boom. Almost every business is using AI now or wants to. But there's still a bigger market of companies who want to make more money than who are actively shopping for agents. Here's how I see the market (from biggest to smallest) and where I see cash flowing to next year. 1. AI-native services Every company wants more leads. Offer to grow revenue or profit for clients. Get AI to run your delivery behind the scenes. This is great for late-adopter companies in older industries who don't know about AI and don't want to. 2. Use AI chat apps Every company's AI journey starts with buying the team Claude or ChatGPT subscriptions. Many companies need training to help teams get more value from the tools, change management so teams actually use them, and/or consulting on use cases and operations. 3. Deploy agents Buy and configure off-the-shelf agents, like Harvey for law or Cursor for engineering. Get the agents working within the company's context, tools and workflows. I'm bullish on personal agents like Grok Bot. Companies will need help finding use cases and setting up their first set of more advanced bots. I think this will eat some of the market for level 2 above and some of the use cases for level 3 below. 4. Build custom agents AI engineers scope and build internal tools. Founder-led companies and forward-thinking execs love this offer. But it's a smaller market than companies who want to use off-the-shelf tools (2-3) or just have work done for them (1). 5. Post-train open models Coinbase did it for fraud prevention. Some big companies are doing it. But post-training is almost entirely done by application layer AI companies, not SMBs for internal use cases. I believe you can build a solid business in any of the five levels. There's a big enough market for all of them. And I see demand for 3-5 growing rapidly in the coming years. My company Atherial builds custom agents and tools (level 4) does AI strategy (e.g. finding use cases). I'm considering adding a level 3 offer. Here's why: I use Grok Bot every day. Personal agents have gotten incredibly powerful and easy to use. Some of what used to need a custom build can now be configured in an afternoon. So I'm thinking about offering Knowledge Work Factories built on Grok Bot. We'll still find high ROI use cases, but now instead of custom building with code, we'll configure in Grok Bot. Maybe still integrate with the tools they already use so as little training and change management is needed. What do you think? Where do you see the most demand and opportunity?