Salesforce's AIforce Layer Under Scrutiny as Critics Demand Performance Metrics

Salesforce's AIforce layer was touted as a game-changer for CRM work when it was announced last week. According to Salesforce, Claude works better with AIforce than without it. However, Chris Pearson, who wrote about AIforce in salesforcedevops.net, is skeptical and wants to see the performance metrics.

Pearson has experimented with Claude and found that while it can handle simple queries, it struggles with more complex requests, especially in orgs with high tech debt. He notes that without AIforce, Claude relies on educated guesses rather than accurate information, which can lead to a 'doom loop' of token-hungry responses.

AIforce is supposed to address this issue by providing the model with existing permissions, business logic, and governance. However, Pearson argues that the absence of performance metrics for AIforce makes it difficult to trust the layer's claims. He proposes a test to compare AIforce's performance with and without the layer.

The proposed test would involve running CRM tasks against a reference org using three different configurations: a raw community-built MCP connection, an existing hosted 360 MCP Server, and Salesforce's new Headless 360 MCP Server powered by AIforce. The results would be graded blind to ensure objectivity.