🌏 中文版
This is a preparation path built from official material, not an exam-day account — I have not sat this exam, and at present nobody can. Every "what it tests" points back to the official certification page and the official Exam Study Guide. No leaked questions. Verified 2026-08-18; registration status, specs, and weights re-confirmed 2026-10-12.
NCP-AAI is one of the few professional-level credentials dedicated to agentic AI, a category it shares with Microsoft's AI-500. Only one of its ten areas (7%) is explicitly tied to NVIDIA products; the rest is general agentic engineering. Registration is not open yet, though, and that shapes how you should prepare.
For prices, validity, and gates across vendors, see What AI certifications engineers can take in 2026 — not repeated here.
Registration: Not Open Yet
Next to the "Register for Exam" button on the official certification page sits the label "(Coming soon)". Its sibling NCP-GENL is in the same state, while both associate exams (NCA-GENL, NCA-GENM) link straight to a Certiverse checkout.
NVIDIA publishes no opening date. So this article's purpose is: the objectives are public, so use them to audit your gaps and plan hands-on work — but don't put "passed" on a near-term roadmap.
Official Specs at a Glance
| Item | Detail |
|---|---|
| Fee | $200 |
| Length | 120 minutes |
| Items | 60–70 |
| Passing score | Not published — the FAQ says "pass/fail. You won't receive a score." |
| Validity | 2 years, renewable only by retaking |
| Language | English only |
| Delivery | Online, remotely proctored |
| Registration | Not open yet (Coming soon) |
The prerequisites are specific:
1–2 years of experience in AI/ML roles and hands-on work with production-level agentic AI projects. Strong knowledge of agent development, architecture, orchestration, multi-agent frameworks, and the integration of tools and models across various platforms. Experience with evaluation, observability, deployment, user interface design, reliability guardrails, and rapid prototyping platforms is also essential…
Note "production-level agentic AI projects" — not a gap you can read your way out of.
The Ten Topic Areas
NVIDIA slices this exam more finely than its others, into ten areas. The weights and descriptions both come from the table on the official web page:
| Topic area | Weight |
|---|---|
| Agent Architecture and Design | 15% |
| Agent Development | 15% |
| Evaluation and Tuning | 13% |
| Deployment and Scaling | 13% |
| Cognition, Planning, and Memory | 10% |
| Knowledge Integration and Data Handling | 10% |
| NVIDIA Platform Implementation | 7% |
| Run, Monitor, and Maintain | 5% |
| Safety, Ethics, and Compliance | 5% |
| Human-AI Interaction and Oversight | 5% |
| Total | 98% |
Two weights differ between the web page and the PDF. The PDF study guide prints Deployment and Scaling at 5% and Run, Monitor, and Maintain at 7%, with the rest identical; the web page totals 98% and the PDF 92%, so neither reaches 100%. When budgeting time, treat Deployment and Scaling as a 5–13% range and plan around 10%; the other eight areas agree across both versions and can be scheduled by their numbers.
Agent Architecture and Design (15%) — officially, "foundational structuring and design of agentic AI systems, focusing on how agents interact, reason, and communicate within their environments." Bullets include reasoning and action frameworks such as ReAct, agent-to-agent communication protocols, short- and long-term memory management, multi-agent workflow orchestration, logic trees and prompt chains for multi-step reasoning, and integrating knowledge graphs for relational reasoning.
Agent Development (15%) — "practical building, integration, and enhancement of agents." Bullets include prompt and dynamic prompt chains, integrating generative and multimodal models (text, vision, audio), building custom tools and APIs, error handling with retry logic and graceful failure recovery, and dynamic conversation flows with real-time streaming and feedback.
Evaluation and Tuning (13%) — "measuring, comparing, and optimizing agent performance." Bullets include evaluation pipelines and task benchmarks, cross-task and cross-dataset comparison, collecting and integrating structured user feedback, and tuning for accuracy-versus-latency tradeoffs.
Deployment and Scaling (13% on the web page, 5% in the PDF) — "operationalizing and scaling agentic systems." Bullets include production-scale multi-agent deployment and orchestration, MLOps for CI/CD, monitoring, and governance, performance profiling under distributed load, containerized scaling with Docker and Kubernetes plus load balancing, and cost optimization with high availability.
Cognition, Planning, and Memory (10%) — "core cognitive processes underlying intelligent agent behavior." Bullets include memory mechanisms for short- and long-term context, reasoning frameworks (chain-of-thought, task decomposition), planning for sequential and multi-step decisions, stateful orchestration, and adapting reasoning from prior experience.
Knowledge Integration and Data Handling (10%) — bullets include retrieval pipelines (RAG, embedded search, hybrid), configuring and optimizing vector databases, enterprise ETL, data quality checks and augmentation, and real-time reasoning over structured and unstructured knowledge.
NVIDIA Platform Implementation (7%) — the only NVIDIA-specific area: integrating NeMo Guardrails for compliance and safety, deploying NIM microservices for high-performance inference, optimizing workflows with the NeMo Agent Toolkit, using TensorRT-LLM and Triton Inference Server to cut latency, and managing multimodal input pipelines on NVIDIA hardware.
Run, Monitor, and Maintain (5% on the web page, 7% in the PDF) — monitoring dashboards and reliability metrics, log and anomaly tracking, continuous benchmarking against prior versions, automated tuning, retraining, and versioning.
Safety, Ethics, and Compliance (5%) — system security and audit trails, compliance guardrails, bias and toxicity mitigation, layered safety frameworks (filters, escalation protocols), and licensing and regulatory compliance.
Human-AI Interaction and Oversight (5%) — user-in-the-loop interfaces, structured feedback loops, transparency mechanisms (explainable reasoning, decision traceability), and human oversight and intervention.
How It Differs From Microsoft's AI-500
Both are professional-level agentic AI exams, so they belong side by side:
| NCP-AAI | Microsoft AI-500 | |
|---|---|---|
| Fee | $200 | $165 (plus the required $165 AI-103 first) |
| Status | Registration not open | Open; the page no longer carries a beta label |
| Heaviest areas | Architecture 15% + Development 15% | Development 30–35% |
| Platform lock-in | Only 7% is explicitly NVIDIA product content | The whole exam revolves around Microsoft Foundry |
| Prerequisites | 1–2 years AI/ML with production agentic work | Must hold the AI-103 certification |
| Official training | Five DLI courses, all paid | Learning paths not live |
NCP-AAI's platform lock-in is surprisingly low — one area out of ten (7%) explicitly tests NVIDIA products; the other nine are general agentic engineering. That makes it closer to vendor-neutral than most vendor certifications, and it means preparing for it transfers better to your actual work.
The Five Recommended Courses
NVIDIA lists recommended training on the page with a price on each; none is free:
| Course | Format | Price | Hours |
|---|---|---|---|
| Building RAG Agents With LLMs | Self-paced | $90 | 8 |
| Evaluating RAG and Semantic Search Systems | Self-paced | $30 | 3 |
| Building Agentic AI Applications With LLMs | Self-paced | $90 | 8 |
| Adding New Knowledge to LLMs | Instructor-led | $500 | 8 |
| Introduction to Deploying RAG Pipelines for Production at Scale | Self-paced | $90 | 8 |
The four self-paced courses total $300; adding the instructor-led one takes it to $800.
Selection advice: Building Agentic AI Applications With LLMs ($90) maps most directly onto the 15% of Agent Development, and Evaluating RAG and Semantic Search Systems ($30) covers Evaluation and Tuning's 13% at the lowest price. Those two, at $120, are the best-value combination. The $500 instructor-led Adding New Knowledge to LLMs maps to relatively little of the blueprint — skip it unless your employer is paying.
What to Do Now
Since you cannot register, the right move is not a study plan but a gap audit against the objectives:
- Use the ten areas as a checklist and ask, for each, "have I done this in production?" — that is precisely what the stated prerequisites demand
- Close the platform 7%: NeMo Guardrails, NIM microservices, the NeMo Agent Toolkit, TensorRT-LLM, and Triton are the only things general agent experience will not transfer to
- Wait to buy courses until registration opens: DLI courses have no expiry pressure, but the blueprint may shift when the exam opens, and buying early risks learning material that gets cut
On this site, the harness layer of agent security and RAG evaluation frameworks map directly onto the Safety 5% and Evaluation 13%.
What May Change (Re-check Before You Register)
| Item | Status as of 2026-08-18 | When to re-check |
|---|---|---|
| Registration | Coming soon, not open | Monthly |
| Two weights that differ | Web totals 98%, PDF totals 92% | When registration opens |
| Fee and item count | $200, 60–70 items, 120 minutes | When registration opens |
| DLI courses and prices | Five, $30–$500 | Quarterly |
| Language | English only | Every six months |
References
- NCP-AAI certification page (specs, blueprint, recommended training)
- NVIDIA certification overview and FAQ (scoring, retakes, recertification)
Related on this site
Loading...