Cramming in the LLM age
The promise
This is a real-world recap of cramming for the Google Cloud Professional Agentic Architect (Beta) exam using LLM-assisted workflows.
No generic summaries you can find in official documentation.
The exam (Beta)
- 75 questions, choose 1 of 4, choose 2 of 5
- my clock started countdown at 2 hour 46 min?! (expecting 3 hour)
- completed at 33 mins left, while appreciating good distractor design and giving feedback
- 13 marked for review, but didn’t change anything. If you don’t know, you don’t know
Guide: https://services.google.com/fh/files/misc/professional_agentic_architect_exam_guide_english.pdf
Professional Agentic Architect Certification | Learn | Google Cloud
Why only 2 days
- Exam is now split into 2 phases (Conceptual MCQ, and Hands-on Lab later), so the hands-on training can be done later
- I have a week of ADK experience from #allthingsagentichackathon so not starting from zero
- Training to be more agile, Parkinson’s Law (work expands to fill the time available)
- I have taken the Machine Learning Engineer exam in 20 days, and thinking how a 10x engineer would use 10x less time (https://medium.com/data-science/20-days-to-google-cloud-professional-machine-learning-engineer-exam-beta-b48909499942). MLE exam turned out much harder.
Outcome
- Scores will be calibrated and released late Oct
- Learned that IAM which is already hard just got way harder (Fortunately only 15% weightage)
- Happy to go through refined learning processes instead of following the traditional learning path, less time invested to gain more capability
How I prepped
Claude (Opus 5.5 Medium Effort) to the rescue
/goal I am preparing for a google cloud beta certification exam.
This means no one else has done it so there’s no online experience.
However google may have released webinars on it, so try to find information from google forums.
I need a 2 day preparation plan.
Here is the exam guide: https://services.google.com/fh/files/misc/professional_agentic_architect_exam_guide_english.pdf
I want to be efficient. Relate each point to a course in the learning path at https://www.skills.google/paths/4525 so i know which part of which course to go through.
You can see the exam has different weightages for different sections , so consider ordering learning from high percentage to low in case i cannot complete learning in time before exam.
Also consider which topics are foundational knowledge for other topics. The learning path may have already ordered it but you may have own opinions after scanning through.
All the courses of the path should be accessible except https://www.skills.google/paths/4525/course_templates/1445 for some reason i don’t know.
If you have a way to get around tell me, otherwise you can fill in the blanks from other resources. Remember to focus on the exam guide. Additional information is not necessary.
In case i don’t have time to go through the courses, I want a per course interactive lesson and mindmap summarizing all the ideas from that course.
I also want an overall mindmap connecting all the material from the courses, labelled with which exam guide requirement it fulfils.
It should be a 2-way link so i want to go in reverse from the exam guide to the course material too.
All the summaries should show up in a single interactive html
How did Claude do
The Interactive HTML: https://claude.ai/artifact/Ey619wXCDS5Jn8wFMS81CT
- How unlucky, the only item it tagged “if time” (Agentic Strategy: Discover, Design, and Prototype) among “must-do”, “should-do”, was exactly the course I went through completely before generating this guide.
- Under Beyond the course path section, it found https://github.com/vishal-bulbule/google-cloud-agentic-architect-exam-guide even when I was pessimistic in my prompt. Credits to Vishal for saving my prep.
- Where’s my “per course interactive lesson and mindmap”? I’m doubting whether this is really Medium effort setting, forgiven for finding Vishal’s github
- Very resourceful to use Wayback machine to find the denied course. It even used the url that removes path scoping (delete /paths/4525 from skills.google/paths/4525/course_templates/1445) which I wouldn’t have thought. Original path-scoped url would find no results. The labs are still inaccessible, but at least the visible parts confirm I’m missing A2A and MCP. I hope Google Skills fixes them before the hands-on test.
My 2 Days
Day 1
- Did full course Agentic Strategy: Discover, Design, and Prototype, good in practice, too generic to help exam other than .
- Looked through cheatsheets in Vishal’s guide: https://github.com/vishal-bulbule/google-cloud-agentic-architect-exam-guide/blob/main/sections/06-cheat-sheet.md. Don’t understand most terms, but good first exposure before revisiting
- Looked through all Infographics in https://github.com/vishal-bulbule/google-cloud-agentic-architect-exam-guide/tree/main/visual-memory.
Most text labels are too succinct so no context how they relate. Visualizations of enforcement spectrums of hook lifecycles were useful to relate concepts. Color coding and gotchas add nuance and prompt critical thinking
Day 2
- Went through sections: https://github.com/vishal-bulbule/google-cloud-agentic-architect-exam-guide/tree/main/sections
Good to identify which are substitutes (GKE, Agent Runtime, Cloud Run), and which are complements in a process (Agent calls MCP to use tools) - Revisited cheat sheet to understand jargon with better context. Seeing familiar terms side by side helped deepen understanding of how and why they fit or don’t
Do the questions match the weightage (NO)
Provided
Section 1: Building agents using low-code tools (~13%)
Section 2: Using coding agents for application development (~17%)
Section 3: Developing custom agents (~33%)
Section 4: Evaluating and deploying agentic workflows (~22%)
Section 5: Securing and governing agentic workflows (~15%)
Perceived
Section 1: Building agents using low-code tools (~5%)
Section 2: Using coding agents for application development (~15%)
Section 3: Developing custom agents (~40%)
Section 4: Evaluating and deploying agentic workflows (~30%)
Section 5: Securing and governing agentic workflows (~10%)
Question Answering Strategy
The options are mostly very different, so the strategy from my MLE article about 2 choices on each of 2 variables permuting into 4 options doesn’t apply as frequently.
It takes more time to seriously read each option.
How are the questions
Generally 3–6 sentences, readable within 1 minute.
No crazy things like MLE beta in 2020, each option longer than the question, and the whole question probably taking 10 minutes just to read.
How are the distractors
Mostly weak
1. redo what is already done in the question
2. goes against the question, doing things that the question already says don’t do
3. contradict what the question gives
4. subset of other options
5. ridiculously complex setups
Some Valuable
The better designed questions will teach you more about the services through distractors.
1. What flags exist (may be fake though)
2. Variations of doing the same task from easy to complex
3. Architectural designs that don’t suit the question at hand, but you may use outside the exam
What are the decision axes
- Scope of configurations in coding assistants
- Extent of Determinism
- Local vs Cloud, 3rd party (eg. open source) vs Google Managed
How to study
Do not accept what you are given. We can design our own UX today.
Have a vision, make it happen.
Evaluation Criteria
No way I’m going to remember all 13 evaluation metrics at https://adk.dev/evaluate/criteria/ in a flat table, so I asked gemini for a decision tree.
The 13 critera are rather well-named, so even if someone went in with 0 knowledge, he may make smart guesses.
Authentication
If you have 1 day to prep
Vishal mentions at the bottom of https://github.com/vishal-bulbule/google-cloud-agentic-architect-exam-guide/blob/main/sections/00-overview.md: Final 3 days: only the Exam signals, decision tables, and practice question explanations from each chapter + the Section 6 cheat sheet.
Much of the “The 2026 rename map” in the overview are not something to be concerned about so don’t need to memorize, just read the new name directly.
The options are different enough so you will know what each service is refering to even if they mix old and new names. I didn’t try to memorize and none of the questions triggered any naming confusions.
Each section’s README took me a good 2–3 hours to read.
You can go in decreasing weightage order and even skip Section 5 Governance entirely because who knows how soon agents will escape control and take over the world? Determinism is our last saviour.
Is this an insult to those who prep more thoroughly
- As long as you are really learning instead of memorizing exam dumps, it’s fine. Your mastery or incompetence will show during design discussions
- Time spent learning has diminishing returns, not everyone learns at the same pace or is looking for the same level of mastery. Measuring competence by years of experience or hours of passive video watching is becoming obsolete
Reflections
I wish the exam designers went the MLE beta style, making the easy very easy and hard very hard.
However, well designed interfaces and systems have usually 1 obvious way to do it, so making it easy for users, also makes it harder to write distractors in exams.
The exam title includes Architect, so high level design is more important than low level configuration. This is increasingly true with assistive coding available today.
I wish governance and IAM was easier, but we can expect authentication and authorization to get even more complex in future as agents’ capabilities grow.
I learned a lot from the prep and exam itself, and recommend anyone interested to give it a try.
I’m curious to hear how others learn too, so feel free to connect in comments or at https://www.linkedin.com/in/hanqi91/
2 days to Google Cloud Professional Agentic Architect Exam (Beta) was originally published in Google Cloud – Community on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source Credit: https://medium.com/google-cloud/2-days-to-professional-agentic-architect-exam-beta-ab794eb4e7cf?source=rss—-e52cf94d98af—4
About the Author
