Blog
Salesforce architecture, Agentforce, and what actually holds up once the demo is over.
Technical debt was the number one Salesforce admin challenge for the second year running, and only 2% call their org clean. Agents turn that from a slow tax into an invoice. Here is the rubric I score orgs on before anyone grounds an agent in them.
Every Salesforce estimate given before someone has opened the org is a guess dressed up as a number. Here is what half a day of looking actually turns up, and why the findings are so consistent.
One invoice generating records across seven Salesforce objects looked sophisticated in the design review. Then Agentforce arrived and turned that design into a recurring bill.
Anthropic gave three agents conflicting goals and write access to the same project. They fought. I read that as an architecture review, and I'd seen the same failure in a Salesforce org years earlier.
A prompt can make a model sound like a Salesforce architect. A harness is what lets it operate like one. This is why the demos look great and the deployments feel fragile.
I was told a Salesforce archive couldn't be done because the data was too tightly integrated. The same mistake that made it look impossible is the one teams are now making with AI agents.