Post

As I talk to business owners and fellow agency folk about how they're adopting/using AI, a few themes keep coming up: 1) Lots of people are diving in to experiment, relatively few feel like they know what they're doing, but that isn't stopping them from trying.

2) The % of people who are truly "AI-pilled" (following all the trends, experience with multiple models/versions) is incredibly small. Most AI adoption is, like, buying Claude for Teams and having everyone use the Claude desktop app to make work happen faster.

3) Almost no one had or has a good baseline for what "good" or "faster" looks like, so AI effectiveness is being measured on vibes, not hard data. More stuff is being produced, so it feels more productive in the short term, but outcomes are basically the same.

4) Likewise, everyone has some back-pocket tests for AI result quality, but almost nobody outside tech is doing benchmarking or structured evals to say with any confidence whether Opus 5.99 is actually better at writing than GPT-67 Andromeda, or whatever. Again, vibes.

5) Most businesses are chasing (in this order) quality, speed, and value. I can't honestly say I see huge differences in results from Opus 5.5, Fable 5.1, 6.1 Sol, or 6 Astra, but some folks are convinced [newest Anthropic model] is the only good model for what they do. (contd)

5.1) Quality = a lot of interest in whether models can go off and work autonomously for hours at a time with highest-quality output, with no human in the loop, or humans just involved at the end to review work and give notes

5.2) Speed = at DevDay, OpenAI put more emphasis on workflows where people are actively using AI to shape work in real time, in which case speed becomes super important (hence Ultrafast mode). Quality is important, but the feedback loops are tighter/faster

5.3) Value = Still seems like some heavy AI users see it as a badge of honor to spend thousands on AI. But more and more people want to get the most out of _one_ subscription, but value is a distant third in the discourse even though it's likely the most important thing to businesses as time goes on

6) Most of the above applies to a tiny fraction of users/businesses. The vast majority are using ChatGPT on the free tier, or Claude on the cheapest pro plan, mainly just getting answers or typing things out rather than giving AI agents real tasks yet.

0 reposts0 likes

An image that comes to mind is, if AI is a pool at a resort, most people are hanging with their kids in the shallow end or lounging poolside, with a few crazies like me crowded together at the deepest end — almost nothing in between.