Certifying Agents

Outcome: trusted, explainable certification that scales across thousands of agents.

The Agent Gallery now carries agents that plan, reason, call tools, and act with autonomy. Without certification, it is hard for enterprises to:

• trust what an agent will do in production or what data it will touch,
• compare agents that claim similar skills but deliver different results,
• deploy safely in regulated environments where controls and auditability matter.

Our certification framework addresses these gaps with two dimensions:

• Dimension 1: Agent Capability Maturity — how agentic it is, how much autonomy and agency.
• Dimension 2: Agent Readiness — how operationally ready it is across security, effectiveness, and interoperability.

Dimension 1: Agent Capability Maturity

Five levels describe how agentic a workflow is and how the user interacts with it:

1. Level 1 — User as Operator, Agents as Task Executors (narrow task): helper bots, fixed scripts, minimal governance.
2. Level 2 — User as Collaborator, Agents as Trustworthy Assistants (multi-task): scoped delegation with human supervision.
3. Level 3 — User as Consultant, Agents as Team Players (pre-coordinated): coordinated agents, orchestration, identity-aware workflows.
4. Level 4 — User as Approver, Agents as Goal-Driven Professionals (dynamic planning): autonomous planning with continuous feedback.
5. Level 5 — User as Observer, Agents as Autonomous Workers (fully autonomous): self-managing digital employees with adaptive optimization.

Example prompts we use to place an agent:

• Interaction: does the user provide step-by-step instructions for a single predefined task?
• Autonomy: does the agent make independent decisions or only follow a fixed script?
• Scope: can the agent chain commands and handle ambiguity or is it limited to one narrow purpose?

Dimension 2: Agent Readiness

Seven levels describe operational readiness and deployment posture:

1. Concept & Roadmap Tier — requirements defined; design/build in progress.
2. Prototype Tier — standalone prototypes validated on intended outcomes with mostly synthetic data.
3. Trusted Agent Huddle Tier — prototypes onboarded to the AI refinery and orchestrated across systems.
4. Pilot Tier — full functionality with limited robustness, tuned on production data for pilot groups.
5. Production Tier — General Usage — agents in scaled internal production for general purpose use.
6. Production Tier — Customer Usage — agents with high functionality, tuning, and controls for customer-facing production.
7. Production Tier — Regulatory Usage — agents with regulatory-level controls for highly regulated production systems.

Readiness measures

Readiness is scored across three measures:

• Performance — effectiveness, correctness, reliability, and latency under real and synthetic workloads.
• Security — risk-based grading with static/dynamic analysis, supply-chain scanning, and prompt-injection detection.
• Functionality & Interoperability — ecosystem readiness, tool contracts, and documentation quality.

Readiness gating by tier

---

Next steps

• Read Functionality Certification to understand static analysis & ecosystem checks.
• Read Performance Certification to see how we validate correctness & speed.
• Read Security Certification to learn the risk framework and grading pipeline.