Audit whether your Salesforce org is ready for AI agents
2026 buyers ask whether the org is ready for Agentforce / agentic AI. sf-intelligence already ships the kit — this page is the playbook, not a new product.
Short answer: Refresh a vault, then run debt → unused → risk → exposure tools in order. Every answer stays on your machine, read-only, with confidence labels. For safety framing see is AI safe on my org?.
run the audit pass on a demo org - no org, no auth
claude mcp add --transport stdio --scope user sf-intelligence-demo -- npx -y sf-intelligence demoThen walk the playbook against a synthetic org that has debt planted in it on purpose: "which Apex has governor-limit risk?" (a batch class queries inside a loop), "find hardcoded IDs in the Apex", and "which test classes don't really assert anything?". It shows you the report shape and the honest exclusions before you point it at an org anyone will read the numbers from. Fully offline: no Salesforce login, no sf CLI. more on the demo →
Why this framing
Industry commentary in 2026 treats technical debt as a top admin pain and "get AI-ready" as a budget line (Hubbl and Elements sell into it). The capabilities below already exist in the roster — packaging them as one narrative is the GTM move, not inventing new analysis.
Playbook (compose existing tools)
| Step | Ask / tool | What you learn |
|---|---|---|
| 1. Debt score | sfi.tech_debt_score | Weighted debt with honest exclusions when an extractor didn't run |
| 2. Org risk | sfi.org_risk_report | Synthesis of risk themes; optional gate: true for deploy readiness |
| 3. Unused surface | sfi.unused_fields_deep, sfi.find_dead_code | Cleanup candidates before agents inherit ghost metadata |
| 4. Hardcoded / quality | sfi.find_hardcoded_values, sfi.governor_limit_risks | IDs, injection-ish patterns, governor risk (heuristic) |
| 5. PII map | sfi.pii_inventory | Which fields look sensitive by name/type/description |
| 6. Guest exposure | sfi.guest_exposure_report | Experience Cloud / unauthenticated surface (when modeled) |
| 7. AI exposure | sfi.ai_exposure_report | What modeled GenAI / Agentforce-related assets touch |
| 8. Docs handoff | sfi.generate_onboarding_doc, sfi.generate_data_dictionary | Human-readable baseline for the cleanup backlog |
Example prompts
- "Score technical debt for this org and list the top cleanup themes."
- "Which unused custom fields are safest to retire, and what still references them?"
- "What PII could an agent see if we expose Account and Contact broadly?"
- "Run an AI exposure report — what GenAI assets are modeled and what do they touch?"
Honest limits
- Static / vault-bound — not a runtime agent monitor.
- Heuristic findings stay labeled; spot-check before executive claims.
- Coverage depends on retrieve + extractors; gaps are disclosed, not invented as "none in org."
- Commercial audits still win when you need a vendor logo on a board deck — see vs Hubbl and vs Elements.cloud.
MCP tools vs Agent Skills
Readiness is not just a metadata score - it is whether an agent can act on that metadata safely. sf-intelligence is agent infrastructure with two layers. The read-only MCP tools are the execution layer: grounded, offline reads that return cited, confidence-tiered evidence. Salesforce Agent Skills are the intent layer: portable instructions that decide when to run an audit tool, how to scope the question, and how to hedge the finding. Skills interpret the ask; the MCP tools do the grounded org read.
The two are complementary, not a migration - MCP stays the grounded execution surface while skills make the same read-only tools portable across Claude Code, Claude Desktop, Cursor, and other MCP clients. See how Agent Skills and MCP tools work together.
Next
Install · capability map · compare hub · repo POSITIONING.md.
Audit locally. Then widen agents.
Free on npm. One retrieve. Read-only by construction.