Free AI Agent Ops Cheatsheet for RevOps

What is the difference between an AI agent and a workflow, and how do I know which one to build?

A workflow follows a fixed rule — if X happens, do Y, every time, no variation. An agent reads context, interprets it, and decides what to do based on what the data actually says. If the right action is always the same regardless of the deal, build a workflow. If the right action depends on stage, activity history, close date, or any other combination of signals, that is an agent job.

Do I need Weflow specifically to use this, or can I apply it with native Salesforce Flow?

The framework — Trigger, Context, Action, Output, Feedback loop — works regardless of your automation layer. The cheat sheet uses Weflow for the workflow diagrams, but the same logic maps to native Salesforce Flow or any other orchestration tool that can fetch records, call an LLM, and write back to Salesforce objects. What you cannot skip is the layer that fetches context and writes output back — Claude handles the reasoning in the middle, but it does not trigger, pull data, or update fields on its own.

What Salesforce data needs to be in good shape before any of these agents are worth building?

At minimum: Stage, close date, amount, and owner populated on all open Opportunities, field history tracking enabled, and email and calendar sync writing to native Salesforce objects rather than a virtual layer. The cheat sheet is direct about this — if Activity History is incomplete, the deals-at-risk agent and the rep nudge agent will produce outputs that look correct but are not. Check field completion rates on the objects each agent reads before you build anything.

Which of the four agents in the cheat sheet should I build first?

Start with the deals-at-risk agent — it has the clearest trigger logic, reads fields that most teams already populate, and produces output that is immediately useful in a pipeline review. The meeting brief agent is high-value but depends on call transcripts being in Salesforce, which is a data dependency many teams have not solved yet. Get one agent running cleanly and audited before you stack more.

How do I know if an agent's output is actually accurate and not just plausible-sounding?

Pull five random records after every run and manually verify the agent's output against the raw Salesforce data it read — this is the per-agent audit step in Section 8. The specific signal to watch is false positive rate: if reps or managers are dismissing more than 20% of agent flags, your trigger threshold or prompt needs recalibrating. Output that looks structured and confident but does not match the underlying data is the most common failure mode, and it only surfaces if someone is checking.

How often should I review and update the prompts powering these agents?

Run a weekly health check on output quality and false positive rates, and do a full prompt review every quarter. Prompts written for last quarter's pipeline motion may not reflect how your team sells today — deal stages get renamed, qualification criteria shift, and what counted as a risk signal six months ago may not be the right threshold now. Treat prompts like code: version them, log what changed and why, and never overwrite a working prompt without saving the previous version first.