If you're a RevOps leader, CRO, or founder at a B2B SaaS company who's already deploying agents and wondering why the ROI hasn't shown up yet, this guide was written for you.
What you’ll find in this guide
This guide is designed to move you from a Claude prototype that demos well to a revenue agent fleet you can trust in a board meeting.
- Why Claude needs context: What has to be in place before any agent you build will work in production
- How to connect Vasco to Claude: The exact commands, the right sequence, and what you get that a raw CRM connection doesn't
- Build, run, and scale: The step-by-step prompt structure, the full agent fleet, and how to go from one running agent to twenty
The 2026 priority: Get out of context debt
Most teams are past the "should we use AI" conversation. The new question is harder: how do you build an agent that doesn't just retrieve information but actually reasons on it, connecting a billing signal to a Gong call to a stage change to a quota gap and telling you what it means?
The answer isn't a better prompt. It's a foundation that encodes your definitions, resolves your identities, places your conversations correctly, and gives Claude something worth reasoning from. Once that's in place, the quality of what you can build changes entirely. Agents stop summarizing and start thinking.
The five components every reasoning agent needs
Most agent prompts fail in production for the same reason. They skip the structure that makes Claude consistent. Here is what every agent you build needs, in order.
Component 1: Role and job
Tell Claude exactly who it is, what it produces, and what done looks like. Not "you are a helpful assistant." "You are a WBR analyst. Produce the weekly business review for the CRO. Direct. Quantified. No softening." The more specific the role, the more consistent the output across every run.
Component 2: Organization context
Before any analysis runs, the agent calls Vasco's Organization Settings to load your company name, fiscal year, ICP definitions, segment thresholds, and motion definitions. This is the step that makes a generic template produce company-specific output. Without it, Claude reasons from its own defaults, not yours.
Component 3: Planning guidelines
Tell the agent how to sequence its work, what depends on what, and how to handle missing configuration. The rule that matters most: missing config is a note, not a blocker. The agent never stops because a definition is not set. It documents the gap and continues.
Component 4: The data chain
This is the component most builders skip. Before generating a single sentence of output, the agent calls three Vasco tools in sequence: the Metric Analyst for quantitative data, the Context Analyst for qualitative signals from call transcripts and account history, and the Domain Agent for entity-level detail. Every statement in the output traces back to one of these sources. If there’s no source, then there’s no statement.
Component 5: Output format
Specify the exact structure, section order, and delivery method. "Exactly 3 recommendations ranked by revenue impact" is a constraint. "Some recommendations" is not. The specificity of Component 5 is what separates agents that produce the same quality every Monday from agents that surprise you.
A preview of the revenue agent fleet
Every agent in this guide runs the same 5-component structure on Vasco's context graph. The foundation is what makes them reasoning agents rather than retrieval tools.
Start here: the WBR Builder
Build this one first. WBR prep drops from 4 hours to 20 minutes, and the first run almost always surfaces a metric that has been quietly wrong for months. When the room sees that, every conversation about trusting the agents changes.
Then: the Pipeline Analyst
Nine deals showing as active in HubSpot with zero engagement for 30 days. All nine flagged before the weekly review. Not because anyone checked, but because the agent connected CRM status, Gong silence, and Stripe signals into a single conclusion.
Then: the ICP Discoverer
Once you trust the reporting and the pipeline monitoring, you can ask the strategic question. The ICP Discoverer reads every won and lost deal and tells you exactly which accounts to chase next, validated against your actual outcomes rather than your original assumptions.
For teams ready to build something that lasts
This is not a framework for thinking about AI. It is a step-by-step manual with the actual commands, the actual prompt structure, and the actual deployment sequence.
Every section has something to do, not just read. Start at the connection step and do not skip the data chain.
Read the guide >
