Anthropic Made an Official Claude Certification — Everything on the Exam (CCA-F)

Author: Hyperautomation Labs · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Anthropic Made an Official Claude Certification — Everything on the Exam (CCA-F) · Channel: Hyperautomation Labs · 09:50 · Watch on YouTube

Topics: AI agents, Claude Code, Prompt and context engineering, AI governance and regulation

Summary

The video breaks down Anthropic’s Claude Certified Architect Foundations (CCAF), presenting it as a proctored, scenario-based credential for people who build production systems with Claude. It explains the exam format, pricing, passing score, and the five weighted domains. It also describes the six possible production scenarios and emphasizes that preparation should focus on hands-on judgment and timed scenario practice rather than memorizing definitions. The video closes with suggested study priorities and a promotion for the creator’s prep kit.

Key points

The Claude Certified Architect Foundations launched on March 12, 2026, as Anthropic’s entry-level architect certification.
The exam has 60 multiple-choice or multi-select questions, a 120-minute time limit, and proctoring.
Each attempt costs $99; the video says the first 5,000 employees at companies joining the Claude Partner Network can take it at no cost.
The passing score is 720 on a scaled score range of 100 to 1,000, and question difficulty affects weighting.
The five domains are agentic architecture and orchestration (27%), Claude Code workflows (20%), prompt engineering and structured output (20%), tool design and MCP integration (18%), and context management and reliability (15%).
The exam is aimed at solution architects with hands-on experience building and shipping Claude applications.
Candidates receive four scenarios, randomly selected from six: customer support, code generation, multi-agent research, developer productivity tools, Claude Code in CI/CD, and structured data extraction.
The video recommends studying agentic architecture first, followed by Claude Code and prompting, then MCP and context management.
Timed practice with scenario-style questions is presented as the highest-leverage preparation strategy.

Chapters

Start	Chapter	Summary
00:00	What the certification is and who it is for	The video introduces the CCAF, its launch, and the hands-on experience expected of candidates.
02:03	Exam format, scoring, and domain weights	The presenter explains the exam’s timing, price, scaled passing score, and the five weighted domains.
03:38	The five exam domains	The video details the tested skills, from agent orchestration and Claude Code to prompts, tools, and reliability.
06:45	Scenarios and how to prepare	The presenter outlines the six possible scenarios, recommends timed scenario practice, and promotes a preparation kit.

Quotes

A real proctored exam that says, “This person can architect production systems with Claude.” — 00:00

The passing score is 720, but not 720 out of a flat percentage. — 02:34

The exam cares less about clever wording and more about reliability. — 05:12

The exam is testing judgment under realistic constraints. — 07:46

The single highest-leverage thing you can do is practice on scenario-style questions under time — 08:48

Concepts and entities

Anthropic, Claude Certified Architect Foundations, CCAF, Claude Code, MCP, Claude Partner Network

Full transcript (original language)

00:00 For the first time in its history, Anthropic is putting a name on a credential. Not a course completion. Not a badge for watching videos. A real proctored exam that says, “This person can architect production systems with Claude.” It’s called the Claude Certified Architect. And in the next few minutes, I’m going to break down exactly what’s on it. Every domain, every scenario, how it’s scored, and the one thing most people

00:31 get wrong about passing it. Oh. Here’s why this is a bigger deal than it sounds. Until March 2026, there was no official way to prove you actually know how to build with Claude. Anyone could put AI engineer on a profile. This is Anthropic drawing a line in the sand. The same move AWS made years ago that turned cloud skills into a hiring filter. The Foundations exam is the entry point to that ladder.

01:02 And Anthropic has already said sellers, architects, and developer tracks are coming later this year. So, what is it precisely? The full name is the Claude Certified Architect Foundations. CCAF for short. It launched on March 12th, 2026. The same day Anthropic put a hundred million dollars into its new partner network. It is not a multiple guess trivia quiz about Claude’s features. It is a scenario exam.

01:33 You’re dropped into realistic production situations and asked to make the architectural call a senior engineer would make. This is aimed at one person. The solution architect who builds real applications with Claude and ships them to production. Anthropic’s guidance is roughly six months of hands-on experience. Meaning, you’ve actually wired up agents, written tool schemas, configured Claude code, and debugged prompts that misbehaved in the wild.

02:03 If you’ve only chatted in the app, this exam will feel hard. If you’ve shipped, it will feel fair. Let’s talk format because the numbers matter. 60 questions, 120 minutes. That’s about 2 minutes per question. Every question is multiple choice or multi-select. And it’s proctored, so no tab switching to the docs mid-exam. The price is $99 per attempt. One exception worth knowing. If your company joins the Claude Partner

02:34 Network, which is free to join, the first 5,000 employees get the exam at no cost. And here’s the part most people get wrong. The passing score is 720, but not 720 out of a flat percentage. It’s 720 on a scale track from 100 to 1,000. And the questions are weighted by difficulty. A hard agentic architecture question is worth more than an easy definition.

03:06 So, you can’t just memorize 40 answers and coast. You have to be strong where the weight is, and that brings us to the five domains. The exam is split into five domains, and the weighting tells you exactly where to spend your study time. Memorize these percentages. Agentic architecture is 27%. Claude code workflows, 20. Prompt engineering, 20. Tool design and MCP, 18. Context and reliability,

03:38 15. More than a quarter of your score lives in one domain, so let’s start there. Domain one, the heavyweight at 27%. Agentic architecture and orchestration. This is about designing systems where Claude acts, not just answers. Multi-agent setups. Task decomposition. Breaking a big job into pieces a sub-agent can own. Hub and spoke orchestration. Where one coordinator delegates to specialists.

04:09 The agentic loop itself. Plan, act, observe, repeat. And the hard one, reliability. When do you let an agent run on its own? And when do you force it to stop and escalate to a human? Get comfortable here. This domain alone can pass or sink you. Domain two, 20%. Claude code, configuration, and workflows. This is the practitioner’s domain. The Claude.md hierarchy. How project, directory, and user

04:40 instructions stack and override each other. Custom slash commands. Persistent project context that survives across sessions. And plugging Claude code into CICD. So, it’s not just a chat window, but part of your pipeline. If you use Claude code every day, this is your free points. If you don’t, this is where the gap shows. Domain three, also 20%. Prompt engineering and structured output. And notice the second half of that name,

05:12 structured output. The exam cares less about clever wording and more about reliability. JSON schema enforcement. Few short examples that lock the format. Validation and retry loops. When the model returns something malformed. And the key distinction they test, programmatic enforcement versus prompt-based guardrails. When do you trust the prompt? And when do you trust the code around it? The answer is almost always trust the code.

05:43 Domain four, 18%. Tool design and MCP integration. This is where Claude reaches out and touches the world. Designing model context protocol servers and clients, drawing clean tool boundaries. One tool, one job. The tradeoffs of SAE transport. Authentication patterns, so an agent can act on a user’s behalf without holding the keys to everything. And writing tool schema so clear

06:14 that Claude picks the right tool every time. Weak schemas are the number one cause of an agent calling the wrong function. And the exam knows it. Domain five, 15%. Context management and reliability. The quietest domain. And the one that separates people who’ve shipped from people who’ve demoed. How do you keep an agent coherent across a long task without blowing the context window? What do you keep? What do you summarize? What do you throw away?

06:45 And reliability patterns. Retries, fallbacks, graceful failure. So your system degrades politely instead of collapsing. Small weight, big real-world payoff. Now, those domains don’t show up as dry definitions. They show up inside scenarios. Anthropic built a pool of six realistic production scenarios. And on exam day, you get four of them drawn at random. So you can’t predict which four. You

07:16 have to be ready for all six. They are a customer support resolution agent, code generation with Claude code, a multi-agent research system, developer productivity tools, Claude code running inside CICD and structured data extraction. Read that list again because it’s a map of exactly what to practice. Inside each scenario, you’re not asked, “What is an agent?” You’re asked things like This support agent keeps escalating

07:46 tickets. It should handle itself. What do you change? This multi-agent research system is returning contradictory answers. Where’s the orchestration flow? The exam is testing judgment under realistic constraints. Which is exactly why memorizing flash cards fails. And why practicing on scenario-style questions works. So, is $99 and a couple hours worth it? If you are a freelancer or consultant, this is the first credential that lets

08:18 you say Claude certified and means something real. It’s a differentiator while almost nobody has it. If you’re employed, it’s a clean signal to your team that you can own production AI work. And if your company’s in the partner network, it might cost you nothing but time. The window where being early actually matters is open right now. Here’s how I’d prepare. Don’t read docs cover to cover. That’s the slow way. Map your study time to the waiting.

08:48 Agentic architecture first, then Claude code and prompting, then MCP and context. And the single highest-leverage thing you can do is practice on scenario-style questions under time so the format stops being a surprise. Two minutes a question goes fast. That’s exactly why I built the Claude certified Architect Foundations complete prep kit. It’s a full set of practice questions and timed mock exams that mirror the real scenario format.

09:19 Organized by the five domains and their exact weightings, so you spend your time where the points are. If you’re going to sit this exam, do it once. The link to the prep kit and mock tests is in the description and pinned in the comments. Go in ready. And if this breakdown saved you an afternoon of digging, subscribe. I go deep on the practical side of building with cloud every week on YouTube, Facebook, and Instagram. The certification is just the start of the ladder. I’ll see you on the next

09:50 one.