Opinionated software > platforms that just hand users a box of Legos to let them "figure it out".
Most new AI “platforms,” or AI-native versions of legacy products, give customers primitives: data objects, a wiki to populate with context, memory files to maintain, a skill library to assemble. Then they bolt an agent builder on top.
“Now go build whatever you need.”
Candidly, I think this is the Salesforce philosophy repackaged for AI. The customer buys software to do a job and inherits a "project" to make it useful.
The hard part of building a product is being opinionated. You have to understand the job deeply enough to make decisions on the customer’s behalf. And take responsibility when those decisions are wrong.
Giving an agent access to all the data doesn’t solve that. Which context matters for which decision? What should it remember? When should it act, and when should it leave something alone?
Those questions don’t disappear because you gave the customer an agent builder and somewhere to write instructions.
A customer should tell a sales product who to sell to and their discounting philosophy. They shouldn’t have to teach it not to cold-email a buyer already negotiating with a teammate. And when a buyer spends three calls asking “Why should we switch?” then says “Too expensive,” the product should know that a discount doesn’t answer their question.
We debate this daily at Monaco. Our guiding principles are:
1/ Packaged for USE, not CONFIG. Customers should start with agents that do the job, not a blank agent builder and a setup project.
2/ Customers should teach us what’s different about their business. They shouldn’t have to teach us how sales works.
3/ If a good salesperson would know better, our product should too. When it doesn’t, that’s a gap in our product, not a missing paragraph in the customer’s wiki.
“You can build that on our platform” is not the same as “our product does that.”