Shadowing Practice: Anthropic's CCA Exam as a Field-Guide for Agentic Engineering — Frank Coyle, UC Berkeley - Learn English Speaking with Video
בחר מצב תרגול
כל מצב פותח חדר תרגול במסך מלא.
263 משפטים
אם המשפטים קצרים או ארוכים מדי, לחץ על ערוך כדי להתאים אותם.
1
Frank Hoyle Reviewer Frank Hoyle Okay,
2
getting rolling and welcome aboard.
3
Just had a little technical issues, but we'll resolve them.
4
So my name is Frank Hoyle.
5
I am a computer science guy.
6
I've been teaching computer science for over 30 years, and I'm now teaching at Berkeley.
7
and one of the problems that all my students, past and present, are having is AI,
8
because computer science is no longer the magic pathway to a job.
9
So I've been trying to figure out ways to help them, come up with schemes to help them get ready for this world of agentic AI.
10
And one of the things that sort of dropped into my plate plate was the something called the Claude Certified Architect exam,
11
which I will be talking about today.
12
And it has a number of aspects to it.
13
And I think if you're interested in a career in agentic AI, to certainly take a look at least what the exam is about,
14
because I feel that Anthropic knows how people are using their system and what the issues are going to be.
15
So before we jump into that, I want to give a little bit of my philosophy.
16
May have to do this manually, getting stuck.
17
So this is a quote from a woman named Sister Corita Kent.
18
Nothing is a mistake.
19
There's no win and no fail.
20
There's only make.
21
The bottom line here is experiment, experiment, experiment.
22
Not only should you read, but you should do, you should make stuff.
23
Now, what happens when you make stuff?
24
A lot of times things don't work.
25
Thomas Edison said, I have not failed.
26
I've only found 10,000 ways that don't work.
27
And what I want to emphasize here is that what this shows us are something that in the design patterns movement,
28
which came around in the early 1990s with object-oriented programming, we had patterns for objects.
29
We now have patterns for agents.
30
But there's also anti-patterns.
31
And I think anti-patterns are a key to understanding what you should not do.
32
Because understanding what you should not do is the key to leading you to what you should do.
33
So a little bit about the CLAWD certified exam, released in March.
34
So it's brand new.
35
It is based on scenarios.
36
It is timed.
37
It is proctored.
38
It is available to companies in the CLAWD ecosystem, the anthropic ecosystem.
39
But individuals can pay $99 and take the exam once every six months.
40
And it's not just multiple choice questions.
41
It is multiple choice, but they are based on realistic constraints and realistic scenarios.
42
The five domains.
43
There are five domains that are covered and they give you the percentages of each.
44
So, agentic architecture, 27%.
45
Claude code, how to configure the Claude code system and workflow, 20%.
46
How to doing prompt engineering, structuring your output using JSON all over the place,
47
tool design, model context protocol integration.
48
These are topics that you should understand and know whether you're going to take the exam or not.
49
This is going to help you get ready for whatever the agentic world is going to throw at you.
50
And then there's going to be context management and reliability.
51
So these are the areas of the kind of questions you're going to run into.
52
Then there are, and they provide you with six production scenarios,
53
and the exam will randomly choose four, and all the questions will be centered around the four that they choose.
54
And what I'm going to do is walk you through the production scenarios
55
and give you some anti-patterns to be aware of because there's a number of ways you can solve the problem, but one of the big things is what not to do,
56
and that often can be the key to getting these questions right.
57
So number one, customer support resolution agent.
58
So we have agentic loops, control, something called stop reason, which is what Claude Code has.
59
Every time something happens, there's a stop reason, and you need to take a look at that because that can give you a lot of information about what's going on.
60
Scenario two, code generation.
61
Three, multi-agent research system, which we'll look at.
62
How do you distribute your agents?
63
Hub and spoke, who's the orchestrator, how much information should they know?
64
All these are important factors.
65
Scenario four, developer productivity with code.
66
So how do you do subtask isolation?
67
Keep your tasks in their little universes.
68
And this harkens back to what we learned in computer science from doing multi-threaded programming.
69
When you have multiple threads operating and sharing memory, then you get into issues with synchronization.
70
You got to put locks in.
71
Keep the little threads independent it, keep your agents independent.
72
And then some Claude code for continuous integration.
73
And then we'll look at some patterns for structured data extraction.
74
Okay.
75
That's kind of where we're going to go.
76
Now, here's something that I like to point out.
77
Everybody's talking about loops, right?
78
The loop is the new thing.
79
Boris Churny says he doesn't write code, but his job is to write loops.
80
And Peter Steinberger, master of open claw, says, I don't code anymore.
81
I just design loops that prompt your ages.
82
So loops are the new big thing, right?
83
Well, no, they're not.
84
OK?
85
Back in the day, early days of computing, we had programming languages were exploding.
86
We had Fortran.
87
We had COBOL.
88
And there were big fights.
89
My programming language is better than yours.
90
It can do more.
91
No, it can't.
92
We can do this.
93
Beam and Jacopini, 1966, proved that if you want a language to be Turing complete,
94
which means can compute anything, that computers are possibly able to compute, then you need only three things.
95
The ability to write statements sequentially, okay, to have if-then conditionals,
96
and the third piece is the loop.
97
If you add the loop, you have Turing computability.
98
And now we are seeing this being resurrected in the agentic world with the focus on loops, because up to now we've had sort of sequences,
99
you have prompts, you have maybe if-then, but now we have a loop, and now this is what's giving us the power.
100
This is where the agentic stuff is getting very exciting.
101
Okay, start with scenario one, customer support resolution.
102
So here we have a loop operating, and I'm gonna jump to the anti-pattern.
103
What you don't want is just to let the agent go and do something and get the response back and use it.
104
What you want to do is you want to loop with something called the stop reason.
105
So I'm going to show you a little code here.
106
So here we have while loop.
107
It's a while true.
108
It's a loop.
109
We're looping right here.
110
So the first little block is where we call the model.
111
And we pass it the messages.
112
The messages are essentially the sequence of prompts that exist in the context window.
113
And we are asking the, and we have a prompt, and we have the context, and we have a tool.
114
And we're asking the LLM to do something with this tool and help us out.
115
The problem is the LLM can't do anything.
116
It is just a probabilistic next word predictor.
117
It can't execute tools.
118
So what it does, though, is it can figure out, if you point it to a tool,
119
it can figure out how to set things up so that you or your code can execute it.
120
So it's important to understand that the LLM is not executing these tools.
121
It can't do anything except talk back to you very intelligently sometimes, but all it can do is talk back to you.
122
So when it finishes this task and has a result,
123
which is basically here is, I know what you want, I know what the tool can do,
124
here's how it sets up the parameters that can then be, then use to actually execute the tool.
125
So this second block, you see, why did the LLM come back to us?
126
That's our stop reason.
127
Tool use.
128
Oh, OK.
129
We've stopped because the LLM wants to use the tool.
130
So let's just run the tool.
131
So that's what the second block is.
132
Run tool, the response is what the LLM said, and it's basically the parameters that it has extracted from the data that you provided it.
133
Then it executes that.
134
Then it goes back.
135
Then it continues.
136
Continues means the LLM sees it, says, oh, successful run.
137
So okay, come back down.
138
We're not running a tool anymore.
139
We're ended our loop.
140
Bingo.
141
Now, then we take the answer, and this is an opportunity for you to have a human in the loop potentially.
142
Check the confidence.
143
If it looks good, you keep it.
144
If you don't, then you escalate to a human.
145
So now there's another reason why you need to make sure you check your stop reason.
146
One of the stop reasons may be you have run out of tokens, and this response is based on partial when the LLM had to stop,
147
and it's going to give you a response, but if you have run out of tokens, then you need to take action.
148
Okay.
149
Next scenario.
150
Code generation with Claude.
151
So Claude code has this concept of Claude MD file, a markdown file, where you put all the things you wanted to know.
152
What Anthropic recommends is you have three levels of Claude.
153
One that you have at the top level of your project, the other that you have inside your sort of the project folder,
154
and then within directories you can also specify.
155
So the idea is to have a hierarchical set of rules that could then control how the system is going to respond.
156
Moving right along, we have a multi-agent research system.
157
So here we're going to have the problem is how do I get my agents to go off
158
and do stuff and bring the answers back in a reasonable way?
159
The anti-pattern, you have one agent, and you load it up with tools.
160
So I like to think about you hire somebody to come to your house.
161
You hire a carpenter to come to the house, and the guy shows up with plumbing tools, carpenter tools, electrical tools.
162
He says, I can do anything.
163
Well, maybe you don't want this guy.
164
Maybe you want a professional carpenter.
165
So that's the kind of idea.
166
And this kind of takes us back to some of the functional programming ideas,
167
that functions should be do one thing.
168
And if you can get your agents to do one thing, with maybe one or two tools available to it, then that's going to be a win.
169
And that's going to help you with this exam. So specialize.
170
Don't overload.
171
The other part of this is don't let your agents' context spill over into the main context,
172
because context means tokens, tokens mean money, and the more context you have,
173
the more confused the LLM is going to be in giving you an answer.
174
So even though, oh, a million token context window, I can put everything in there.
175
No, no, don't put everything in there.
176
Limit what's going to go in there, because then you're going to get a much more accurate system.
177
So here's an example of a specialized subagent.
178
So this would be the critic.
179
So let's say you've run some stuff.
180
Now you want to get an agent to look at what's happened.
181
What you want to do is just give it what it needs to solve that critic problem.
182
I'm only giving it here the, we're passing it the claim and the evidence.
183
So your claim is sort of how we're going to solve the problem.
184
Here's the evidence.
185
But you're not giving it the thought processes that went in to creating this claim.
186
Why?
187
When you get a bunch of agents together, collaborating and talking to each other, there's a tendency to have group think.
188
And all the agents seem to kind of devolve into one idea.
189
I mean, it's like you're in a group, you're at a party, and everybody wants pizza except you.
190
But then people talk you into, you don't want to spoil the party.
191
So you go along.
192
And it seems that agents kind of work in the same way.
193
So you're going to return.
194
Basically, you're going to give each agent only a slice.
195
I didn't think about the pizza analogy, but yes, every agent gets its own slice and it should come through.
196
Okay.
197
Fourth scenario, developer productivity.
198
So the anti-pattern, let every subtask dump its full output into the primary thread, crowding out the context.
199
Again, this is what I was just talking about.
200
This is bad.
201
Let the context grow unbounded.
202
Bad, for the reasons we just talked about.
203
You want to isolate your subtask output, and you want to compact long sessions.
204
I'm going to take a second to talk about that.
205
So here's an example of a pattern.
206
You want to have your agent look at the logs and create a summary of where the problems are in the log.
207
So here's your task.
208
Scan all the logs for error.
209
Context fork.
210
So you're forking the agent into a separate thread, where whatever the agent does and thinks and adds tokens to,
211
does not come back and pollute the main context.
212
Now, you see here what happens.
213
Then you take this summation, and then you add that summation without all the other stuff into the overriding context.
214
Now, this last little block is kind of interesting, I think, because you can check your token count,
215
And you can determine how big the token count is.
216
And if you can set some limit, if you have more than 150,000 tokens, then what you want to do is you can run a compact.
217
So Anthropic and Claude have these compaction algorithms that take this giant context and compact it in some way, shape, or form.
218
Not quite sure how the implementation is of that.
219
But there is compaction.
220
Now, a little side effect, a little side channel.
221
I've been walking around.
222
When you walk outside, you see these guys handing out these books.
223
OK?
224
Anybody see these guys handing out these books?
225
Take them.
226
This is actually a pretty good little book.
227
In fact, I was looking at it last night.
228
And one of the things it had in it was this is by this guy Sam Bagua.
229
I have no connection.
230
I don't even know Sam.
231
But there's on line page 32.
232
It says his company provides custom logic for compression of context.
233
And you can write your own.
234
You can extend his base class and have your own compression of your data, whatever you think is important.
235
So I think that's kind of an interesting spin on this whole thing.
236
OK.
237
Claude Code for continuous integration.
238
Anti-pattern.
239
Always have interactive modes in a pipeline.
240
No, no, no, because interactive modes mean Claude will stop and ask you, you want to do this, you want to do that, can I have permission for that?
241
So there are ways to set it up so that it'll just run straight through.
242
The other tip that I'll give you here is there's something called the batch.
243
So you can take your prompts, you can take your work, and you can put them in a batch,
244
and for 50% fewer token costs, you will get the result they promise in at least 24 hours.
245
So if you're going to go take a nap, you're going to go on vacation, you're going to go out, take a day off, run your stuff in batch mode,
246
and you're going to have less to pay. Where am I here?
247
All right.
248
I've only got a few minutes left, a few seconds left, but I want to conclude with this.
249
Remember, nothing is a mistake.
250
There's no win.
251
There's no fail.
252
There's no exam.
253
Only make.
254
You do it, and you make it, and you're going to succeed.
255
If If you want to reach out to me, reach out to me, COIL at Berkeley.
256
Look at my websites.
257
I got a website, Code Supreme AI.
258
I'm a big jazz fan, and I named this website after John Coltrane, Love Supreme, if you know that song.
259
Great.
260
Anyway, that's my story, and I'm sticking to it.
261
And I'm at the zero time.
262
OK.
263
Thank you.
הורד אפליקציה
כל מה שאתה צריך כדי לדבר בשטף
הגיית AIציון לכל משפט
תרגול IPAשלוט בכל צליל
אוצר מיליםבנה את מאגר המילים שלך
משחק אוצר מיליםלמד תוך כדי משחק
אותו ערוץ
C11
What Does an AI Engineer Do Now? - Aparna Dhinakaran, Arize AI
Daily ConversationUniversity StudentsWorking Adults
C11
Your AI Agent Starts Every Session Like a New Hire - Killian Carlsen-Phelan, Sonar
Daily ConversationUniversity StudentsWorking Adults
C15
The Multi-Agent Architecture That Actually Ships — Luke Alvoeiro, Factory
Business EnglishUniversity StudentsWorking Adults
C1Why Software Factories Fail
Daily ConversationUniversity StudentsWorking Adults
C12
How I deleted 95% of my agent skills and got better results — Nick Nisi, WorkOS
Business EnglishUniversity StudentsWorking Adults
C28
Beyond Code Coverage: Functionality Testing with Playwright — Marlene Mhangami, Microsoft
Business EnglishUniversity StudentsWorking Adults
C19
"Software Fundamentals Matter More Than Ever" — Matt Pocock
Business EnglishUniversity StudentsWorking Adults
C1Don't Build Agents, Build Skills Instead – Barry Zhang & Mahesh Murag, Anthropic
Business EnglishUniversity StudentsWorking Adults
FOR YOU
המלץ על סרטון
C13
153 משפטים
Lesson 3: AI as a learning partner | AI Fluency for students
Business EnglishInterview EnglishUniversity Students
C13
200 משפטים
CAP Theorem in System Design Interviews
Business EnglishInterview EnglishUniversity Students
C11
309 משפטים
Claude Code: The Advanced Guide (99% of Devs Skip These Features)
Business EnglishUniversity StudentsWorking Adults
B25
140 משפטים
Mock Interview Preparation: Common Questions with Feedback!
Business EnglishInterview EnglishUniversity Students
C13
80 משפטים
What is clean code? - Uncle Bob
Business EnglishUniversity StudentsWorking Adults
B24
85 משפטים
How To Dominate 2026 as a Software Developer
Business EnglishInterview EnglishUniversity Students
C12
176 משפטים
3 Questions That Reveal What Your Life Is Missing - Arthur Brooks
Business EnglishUniversity StudentsWorking Adults
C13
51 משפטים
Business Partner C1. Unit 1. Video 1.1.1.
Business EnglishUniversity StudentsWorking Adults
C12
210 משפטים
What Software Architecture Should Look Like • Dave Farley • GOTO 2022
Business EnglishUniversity StudentsWorking Adults
C13
260 משפטים
【TED】優れたリーダーはどうやって行動を促すか/サイモン・シネック （日本語字幕）
Business EnglishDaily ConversationUniversity Students
C11
402 משפטים
FULL DISCUSSION: Google's Demis Hassabis, Anthropic's Dario Amodei Debate the World After AGI | AI1G
IELTS PrepBusiness EnglishUniversity Students
אוצר מילים והערות הגייה לשיעור הזה
שיעור הדיבור הזה ברמה C1 מבוסס על הסרטון “Anthropic's CCA Exam as a Field-Guide for Agentic Engineering — Frank Coyle, UC Berkeley”. המילים שחוזרות הכי הרבה: agent, tool, loop, context, code. בסרטון הזה יש 263 משפטים ו-2767 מילים לתרגול Shadowing. הדיבור נמשך 19:52. הדובר שומר על קצב יציב של כ-139 מילים בדקה, קצב נוח לתרגול Shadowing. 85% מהמילים נמצאות בין 3,000 המילים הנפוצות ביותר באנגלית; כדאי ללמוד את השאר לפני שמתחילים.
אוצר מילים מרכזי בסרטון הזה
15 המילים המתקדמות ביותר בסרטון, עם ההגייה והמשמעות שלהן:
| מילה | הגייה | משמעות |
|---|---|---|
| token שם עצם | /ˈtoʊkən/ | אֲסִימוֹן, זִ׳יטוֹן |
| execute פועל | /ˈɛksɪˌkjuːt/ | להוציא להרוג |
| carpenter שם עצם | /ˈkɑɹpəntɚ/ | נַגָּר |
| compact שם תואר | /kəmˈpækt/ | דחוס |
| certify פועל | /ˈsɜː.tɪ.faɪ/ | וידא |
| critic שם עצם | /ˈkɹɪt.ɪk/ | מְבַקֵּר, מְבַקֶּרֶת |
| slice שם עצם | /slaɪs/ | פְּרוּסָה, נֵתַח |
| fork שם עצם | /fɔɹk/ | מַזְלֵג |
| escalate פועל | /ˈɛs.kə.leɪt/ | הִסְלִים |
| excite פועל | /ɪkˈsaɪt/ | הלהיב, עורר |
| collaborate פועל | /kəˈlæb.ə.ɹeɪt/ | שיתף פעולה |
| claw שם עצם | /klɔ/ | טוֹפֶר, טֹפֶר |
| bingo שם עצם | /ˈbɪŋ.ɡoʊ/ | בינגו |
| folder שם עצם | /ˈfoʊldɚ/ | תִּיקִיָּה |
| distribute פועל | /dɪˈstɹɪbjuːt/ | חילק |
Phrasal verbs שתשמעו
| מילה | משמעות |
|---|---|
| figure out פועל | הֵבִין, מצא תשובה |
| come back פועל | שָׁב |
| go out פועל | יָצָא |
משפטים שכדאי לחזור עליהם
משפטים קצרים ושלמים מהסרטון שאפשר להשתמש בהם בשיחה יומיומית:
- Three, multi-agent research system, which we'll look at.
- How do you distribute your agents?
- That's kind of where we're going to go.
- We're not running a tool anymore.
- If you don't, then you escalate to a human.
דקדוק בסרטון הזה
המבנים שהדובר משתמש בהם הכי הרבה, עם המילים המדויקות מהסרטון:
| מבנה | בסרטון |
|---|---|
| הווה מושלם ממושך (Present perfect continuous) have/has been + -ing — פעולה שהתחילה בעבר ועדיין נמשכת עכשיו | I've been teaching · I've been trying · I've been walking |
| הווה מושלם (Present perfect) have/has + past participle — פעולה שקרתה בעבר ועדיין יש לה חשיבות או השפעה עכשיו | have not failed · I've only found · we've had |
| סביל (Passive voice) be + past participle — הדגש הוא על מה שקורה, לא על מי שעושה את הפעולה | is based · is timed · is proctored |
הגייה ששווה לשים לב אליה
הדובר משתמש ב-64 קיצורים וצורות מקוצרות, כמו you're, don't, I'm. אמרו אותם בצורה הקצרה, כפי שאתם שומעים אותם.
- הצלילים של “th”: anthropic /ænˈθɹɒp.ɪk/, pathway /ˈpæθˌweɪ/, algorithm /ˈælɡəɹɪðm̩/
- הצלילים של “sh” ו-“zh”: compaction /kəmˈpækʃən/, specialize /ˈspɛʃəˌlaɪz/, summation /sʌˈmeɪʃən/, compression /kəmˈpɹɛʃ.ən/, conditional /kənˈdɪʃ.ə.nəl/
- מילים ארוכות — שימו לב להטעמה: parameter /pəˈɹæm.ə.tɚ/, ecosystem /ˈikoʊˌsɪstəm/, interactive /ˌɪn.tɚˈæk.tɪv/, hierarchical /ˌhaɪəˈɹɑːkɪkəl/, probabilistic /ˌpɹɑ.bə.bɪˈlɪs.tɪk/
צלילים שדוברי עברית מתקשים בהם:
- /w/ — שפתיים מעוגלות, לא /v/: sequentially /səˈkwɛntʃəli/, workflow /ˈwɝkfloʊ/, pathway /ˈpæθˌweɪ/
- /æ/ — פתוח יותר מה-“אֶ” העברית, לא כמו “a” ב-“cup”: anthropic /ænˈθɹɒp.ɪk/, batch /bæt͡ʃ/, compact /kəmˈpækt/, compaction /kəmˈpækʃən/, parameter /pəˈɹæm.ə.tɚ/
- /r/ — הלשון לא נוגעת בשום מקום, לא ה-“ר” הגרונית: prompt /pɹɑmpt/, anthropic /ænˈθɹɒp.ɪk/, carpenter /ˈkɑɹpəntɚ/, parameter /pəˈɹæm.ə.tɚ/, compression /kəmˈpɹɛʃ.ən/
איך לתרגל עם הסרטון הזה
- הקשיבו לכל הסרטון פעם אחת בלי לדבר, ורשמו את המילים שאתם לא מכירים.
- תרגלו Shadowing משפט אחר משפט במהירות רגילה, וחזרו על כל משפט עד שהקצב שלכם תואם את הדובר.
- הקליטו את עצמכם והשוו למקור, תוך תשומת לב למילים כמו token, execute, carpenter.
מהי טכניקת השאדואינג?
שאדואינג היא טכניקת למידת שפה מבוססת מדע שפותחה במקור להכשרת מתורגמנים מקצועיים ופופולרית על ידי הפוליגלוט ד"ר אלכסנדר ארגואלס. השיטה פשוטה אך עוצמתית: אתה מקשיב לאודיו באנגלית ילידית וחוזר עליו בקול רם מיד — כמו צל העוקב אחרי הדובר עם עיכוב של 1–2 שניות בלבד. בניגוד להאזנה פסיבית או תרגילי דקדוק, שאדואינג מכריח את המוח והשרירים בפה לעבד ולשחזר דפוסי דיבור אמיתיים בו זמנית. מחקרים מראים שזה משפר משמעותית את דיוק ההגייה, האינטונציה, הקצב, הדיבור המחובר, הבנת השמיעה ושטף הדיבור — מה שהופך אותו לאחת השיטות היעילות ביותר להכנה לIELTS Speaking ולתקשורת באנגלית בעולם האמיתי.