Metadata
- Author: Benn Stancil
- Full Title:: An Very Obvious Deal
- Category:: 🗞️Articles
- URL:: https://benn.substack.com/p/an-very-obvious-deal
- Read date:: 2025-10-25
Highlights
we price software via dead reckoning: What’s fair today is what was fair yesterday. Outlook cost [12.50 a month](https://www.zdnet.com/article/microsoft-office-365-more-new-packages-and-prices-coming-in-november/#:~:text=Office%20365%20Small%20Business%20Premium%20will%20cost%20%2412.50%20per%20month%2C%20or%20%24150%20per%20user%2C%20per%20year.), so Yammer cost 15 a month, so Slack cost $12.50 a month.6 (View Highlight)
in the fourth model, this breaks down. When you’re selling a single unit of consumption, people seem much less willing to accept an arbitrary price. And they start asking one question in particular: “How much does this thing I’m buying cost you?” (View Highlight)
later, when we added computational middleware—“an in-memory compute engine”—that made running queries a plausibly expensive operation for us to perform, people still objected, but most customers were ultimately ok with (View Highlight)
And that’s the lesson, I’d argue. If you can sell subscriptions to your software, you won’t get asked about margins, but you only have so much flexibility about what you can charge. And if you sell consumption, you better charge for something that sounds expensive. Storing stuff works. Doing a bunch of hard math works. Generating an AI image works. But rendering a website, or calling an API, or managing a database of files and messages—well, “we don’t think that’s fair.” (View Highlight)
you can charge for usage—that is the trend, after all, and people use dbt a lot. But that doesn’t really work either, because dbt hasn’t historically done anything that looks like real work. It doesn’t store data; it offloads all the hard computation to a database (which is already charging people for that exact operation). And if you’re not doing the thing that feels like work, people get mad when you charge them for it. (View Highlight)
In doing this, Cursor noticed a pattern. A lot of the requests they were sending to Anthropic were relatively simple. They could be solved with simple models, and didn’t require giant state-of-the-art LLMs. (View Highlight)
Put a database underneath dbt Fusion. But a small one; one that’s not about competing with Databricks and Snowflake, but about saving customers’ money by choosing more efficient ways to run queries. (View Highlight)