Most agentic AI pilot failures are not AI failures, they are architectural failures. That is the central argument behind a new framework by John Taschek, executive vice president and chief market strategy officer at Salesforce, who has laid out 12 rules for deploying AI agents in enterprise environments.
Inspired by Edgar Codd’s 12 rules for relational databases, Taschek’s framework is organized into four layers: foundation (data and context), core (agency), operations (work), and apex (engagement).
Foundation, data and context
1. Unified data lineage, Every piece of data must have a traceable history: origin, changes, permissions. No mystery data.
2. Grounded real-time data access, Agents must work with live data, not stale snapshots. Acting on outdated information is a design flaw.
3. Semantic metadata, Agents need to understand the meaning of data. Concepts like “at-risk customer” must be formally defined, not guessed by the model.
Core, agency
4. Observability and behavioral traceability, Every decision must be logged and explainable for debugging and improvement.
5. Continuous adversarial validation, Agents must be tested constantly against edge cases and bad inputs in a permanent red-team exercise.
6. Multi-step reasoning and goal decomposition, Agents must break complex goals into steps and adapt when circumstances change, not just follow a script.
7. Hybrid deterministic governance, AI reasoning is probabilistic, but some rules (legal, financial, safety) must be hard-coded and architecturally impossible to violate.
Operations, work
8. Agnostic orchestration, Agents from different vendors and models must coordinate without custom plumbing, avoiding orchestration lock-in.
9. Human-agent synergy and empathy mandate, Agents collaborate with humans, not replace them. When confidence is low or emotional context is detected, hand off gracefully with full context.
10. Sovereign agency, The enterprise stays in control of data residency, model choice, identity, and policy. External agents get scoped, auditable access only.
Apex, engagement
11. Outcome-based parity, Measure agents by business outcomes, revenue influenced, issues resolved, time saved, not task completion counts.
12. Trusted agency, The highest-weighted rule. Agents earn the right to act through algorithmic fairness, content safety, consent and data permissions, hallucination prevention, explainability, stakeholder value, and vendor accountability.
The framework is supported by Salesforce’s data from more than 20,000 agent deployments. “Most AI pilots focus on capability and speed, and skip the hard work of earning trust from the business,” Taschek noted. A Salesforce study found that more than half of US desk workers remain AI skeptics, citing generic outputs, insufficient training, and low trust.
Sources: 12 rules of agentic AI for successful enterprise transformation (ZDNet, June 24, 2026)