@patrickc Completely agree. It marks an inflection point and was a super complicated escape. It feels pretty preventable and maybe internally orchestrated by OAI.
Claude gets "governed" access, and Salesforce stock pops... HubSpot shipped its Claude connector in July 2025, over a year ago, and daily usage is still down 25%.
A chat interface on top of a CRM only works if the answer is already in the CRM. In complex B2B, it is not. Sales teams use their CRM less every year because the entry is manual and the record is stale by the time anyone queries it. Ask Claude a complex enterprise question against a pipeline where half the deals have no line items, no spec, no grade, and no freight assumption, and you get a confident answer built on nothing.
The work that moves revenue in my industry is CPQ. A customer sends a spec. Someone matches a grade, checks inventory across a dozen suppliers, pulls a price that moves with the resin market, adds freight from the right warehouse, and confirms regulatory fit. In plastics that quote takes four weeks. The work inside it is about two hours.
Claudeforce does not touch any of the data unification or complex workflows that drive value for the company. The four-week delays in customers seeing quotes live in ERP, in supplier PDFs, in a spec email, and in the head of the guy who has been quoting nylon for twenty years. Salesforce does not integrate to that, because it is not a connector problem. It is a data layer, and it is different for every industry.
For DTC, where the catalog is 200 SKUs and the price is the price, this is probably enough. For anyone selling engineered materials, chemicals, or industrial components, it is a nicer way to ask questions about a system nobody populated. Remember, CRM daily usage is down 25% over the last year. Sales teams need to get benefit from a CRM and not just be monitored.
We are building the system or intelligence that pulls together data, then runs the quote. Four weeks to 5 minutes.
#AgenticAI #CRM #CPQ #B2BSales #Plastics #ChemicalDistribution
@WilliamBryk Agents are going to eat the world. Very few people are looking at the fundementals of compute, memory and model size. Fast, small and smart models are here.