Who It Is For
The CAAP is the professional credential for practitioners who design, build and operate agentic AI systems. It is designed for AI engineers building agentic systems for production use, software developers moving into agent development, solution architects designing multi-agent and tool-using systems, and technical leads accountable for agents already in operation.
Exam Specification
| Credential | Certified Agentic AI Professional (CAAP) |
|---|
| Track | Professional Track |
|---|
| Assessment format | Examination (40%) and an applied agentic build project (60%) |
|---|
| Contact hours | 40 contact hours across 5 days |
|---|
| Certification fee | USD 2,400 |
|---|
| Validity | 3 years from award date |
|---|
| Renewal | 45 CPD hours per 3-year cycle, logged with AICA |
|---|
| Retake policy | Reattempt after a 14 day waiting period, up to 3 attempts in any 12 months |
|---|
| Delivery | Through Authorized Training Partners, online or center-based |
|---|
| Verification | Cryptographic registry entry, QR-verifiable digital badge, Open Badges 3.0 |
|---|
Exam duration and question counts are set in the Candidate Handbook and are not published here. The full proctoring, retake, appeals, renewal and revocation terms are on the exam policies page.
Course Learning Objectives
- Build a multi-step agent that plans a task, calls tools through well-defined contracts and completes work end to end.
- Orchestrate multi-step and multi-agent workflows, and justify when orchestration is warranted over a simpler deterministic pipeline.
- Design an agent's context and memory architecture across short-term, long-term and episodic stores so it stays grounded over long tasks.
- Instrument an agent with tracing, structured logging and evaluation suites that measure quality objectively before and after deployment.
- Harden an agent against known failure modes, including runaway loops, hallucinated tool calls and cost runaways, with layered controls tested adversarially.
- Deploy an agent through a staged rollout, with authentication, state handling and integration into existing systems.
- Optimize the token spend and latency of a working agent and prove the gain with before-and-after measurements.
- Document the complete build as an operations-ready pack covering architecture, evaluation results, safety controls and a deployment plan.
Learning Units
LU1: Agent Frameworks, Orchestration & Tool Use
9 contact hours
- LO 1.1Build a working agent loop, using a framework or model APIs directly, and defend the architectural choice.
- LO 1.2Define tool and function-calling contracts with typed schemas, explicit error returns and input validation.
- LO 1.3Orchestrate a multi-step or multi-agent workflow and identify when a plain pipeline is the better design.
LU2: Context Engineering & Memory Design
7 contact hours
- LO 2.1Design a context strategy for an agent: what enters the window, in what structure, and what gets compacted or evicted.
- LO 2.2Implement short-term, long-term and episodic memory, selecting storage and retrieval appropriate to each.
LU3: Evaluation, Monitoring & Observability
7 contact hours
- LO 3.1Build an evaluation suite for a non-deterministic system, with a purpose-built dataset, rubric scoring and pass thresholds.
- LO 3.2Instrument an agent with tracing and structured logging so any failure can be attributed to a specific step.
- LO 3.3Detect and quantify regressions when models, prompts or tools change.
LU4: Safety Controls & Failure-Mode Design
6 contact hours
- LO 4.1Enumerate the failure modes of a given agent and design a layered control for each.
- LO 4.2Test controls adversarially and document the residual risk honestly.
LU5: Deployment Patterns & Integration
6 contact hours
- LO 5.1Take an agent from local prototype to a production-like deployment through defined stages.
- LO 5.2Integrate an agent with existing systems, handling authentication, state and error recovery.
LU6: Cost & Performance Optimization
5 contact hours
- LO 6.1Profile the token spend and latency of a working agent and identify the dominant cost drivers.
- LO 6.2Apply optimizations such as model routing, caching and prompt slimming, and prove the gain with measurements.
Assessment Blueprint
| Learning Unit | Weighting |
|---|
| LU1: Agent frameworks, orchestration & tool use | Exam 8% + project 12% (20%) |
| LU2: Context engineering & memory design | Exam 7% + project 10% (17%) |
| LU3: Evaluation, monitoring & observability | Exam 7% + project 12% (19%) |
| LU4: Safety controls & failure-mode design | Exam 7% + project 11% (18%) |
| LU5: Deployment patterns & integration | Exam 6% + project 9% (15%) |
| LU6: Cost & performance optimization | Exam 5% + project 6% (11%) |
| All units, both components | Exam 40% + project 60% (100%) |
Weightings are indicative of emphasis across the two assessment components. The applied agentic build project is assessed as one integrated system. Certification decisions are made independently by the AICA Certification and Standards Authority.
Credential Terms
| Credential validity | 3 years from award date. |
|---|
| Renewal | Via Continuing Professional Development: 45 CPD hours per 3-year cycle, logged with AICA. |
|---|
| Retake | Reattempt after a 14-day waiting period; maximum 3 attempts in any 12 months. |
|---|
| Appeals | To AICA's Certification and Standards Authority. |
|---|
| Proctoring | Examinations are proctored, online or center-based through Authorized Training Partners. |
|---|
| Conduct | Certification requires agreement to the AICA Code of Professional Conduct. |
|---|
Verification
Every CAAP credential is issued with a unique credential identifier recorded in the AICA verification registry, a QR-verifiable digital badge, and an Open Badges 3.0 export. Each credential is signed with the AICA registry key and recorded in a public transparency log, so any employer can check it against its live registry record. Verify any credential at aicauthority.org/verify.
