Immediate thoughts on Claude Fable

With today's release of Claude Fable, I thought it would be good to jump in and compare its power to other Anthropic models. The first thing I noticed is that Fable, when given instructions to build things, immediately jumps to MCPs much faster than other AI models we've used with Claude. For example, back in the 4.5 era of Sonnet, it would jump straight into writing some code and then work back toward APIs and ingestion. In 4.8 we saw more MCP focus, but building a small agent this morning to help our sales team, I noticed the immediacy to search for MCPs across our stack. This is powerful in terms of speed for us, as the deployed solution was assembled much quicker than if we had built a lot of code first and then wrapped it through APIs. That's a huge bonus for us and our customers who want to move fast; the data is already available, but servicing that data remains a challenge.

The elephant in the room is the cost right now, which is a bit unknown after June 22 when it moves from the account plan to tokenised usage. The caveat is that if the speed of the product is faster than the others because the quality of output and the solutions found are better, then the cost increase is largely negated in my mind by the quality. Yes, it might be 2x the cost of Opus 4.8; however, if it delivers better code faster and doesn't require rewrites and provides better solutions sooner and doesn't need a rebuild, then overall you might end up using fewer tokens in the long run, which is something everyone should be aware of with any model.

Other initial feelings are that the thought and scheduling side seems much more in-depth, and, from running it through my IDE, seems faster when getting to a resolution (i.e., completed product),

To give visibility on what I was actually building here: it was an agent that takes Gemini call notes and automatically syncs them through to HubSpot against the contact and account record.

Yes, this is a simple task for sales teams to do, but our job in the deployed team is to make life easier for the people around us, and sometimes these things get forgotten. If meeting notes don't get logged against accounts and records, then we can't agentically run over the context that we have with customers and the content that we are speaking to people about, including prospects. This severely limits the quality of data that we're capturing. Although this is a simple tool that runs twice a day, it should mean that no data is ever getting missed and is in sync with our CRM. 

Dictated with Wispr Flow