Who Owns The Next Move
A deleted Apex test class and lessons from monkey management changed how I hand work to agents and decide when they should ask.
Divings by Sangam
Gathering the notes.
Please wait while the page loads.
Divings by Sangam · An independent publication
Models tested.
Ideas explored. People heard.
Field notes on AI, Salesforce, and the work of building.
Subscribe on Substack
Join 65 subscribers on Substack.
Short introductions on Substack. Full articles here.
Read the archive
By Gella Sangamesh Gupta Building Salesforce tools and writing from practice. Explore the work.
The Model Scores, My Code Decides
Three Salesforce demos with Jev exposed the difference between a model’s score and the decisions my code still has to own.
I Talked. ChatGPT Did the Salesforce Work.
I used ChatGPT voice mode to inspect Salesforce flows, compare metadata with documentation, and steer the work through conversation.
Keep exploring
More from the field
Vibe Coding is Real Now
Testing Salesforce’s Claude plugin against a missing field, real org metadata, and user-mode permissions changed what I expected from AI coding.
The Hour I Had ⏳
I exhausted an agentic Salesforce trial in under an hour—and lost the chance to measure the tool before deciding what it was worth.
The Context Was Always the Problem
Salesforce can give an agent the org context it needs. Deciding when to stop, what to verify, and whether to write code is still a method I have to own.
Ideas into practice
Built in practice
Engineering
Salesforce Compound Engineering
Salesforce-aware engineering workflows for Claude Code, Cursor, Codex, and other AI coding tools. Plan, build, review, and capture knowledge for the next iteration.
View project (opens in a new tab)Discovery
Hidden Automation Detector
See which Salesforce flows, triggers, and validation rules run when you save a record. A Chrome extension for visibility into automation execution.
View project (opens in a new tab)