跟讀練習: /wayfinder: Nothing is too big to plan anymore - 通過影片學習英語口語
選擇練習模式
每種模式都會開啟全屏練習室。
212 句
如果句子過短或過長,請點選 Edit 進行調整。
1
I think I've figured out a way to plan any size of work with an agent.
我想我已經找到了一種方法,可以和智慧體一起規劃任何規模的工作。
2
The existing planning tools that I was using and playing around with, even ones that I've created, felt too constrained, too tied to a single session.
3
I felt like I wasn't able to be ambitious enough.
4
And because of that, I was kind of constraining the stuff I was building to fit AI, which doesn't feel right.
5
This new approach doesn't have that limit.
6
You can plan enormous chunks of work, and it will orchestrate the planning over multiple types of sessions.
7
It knows that you can't make your way cleanly to the destination.
8
You have to clear the fog of war.
9
It understands dependent decisions and it even allows you to plan in parallel.
10
And the best thing about this is this is based on software fundamentals.
11
This is based on the fundamentals of planning work that I learned when I was a real developer before AI.
12
And I've packaged this all up into a skill that's available right now on my skills repo called Wayfinder.
13
So the way that I was planning work before was really tied into a single session.
14
It's to do with the grill me or the grill with doc skill that's in my skills repo.
15
That's still a super important primitive, but it's really just tied into a single session.
16
Some work is bigger than what you can fit into the context window
17
and especially the smart zone of the context window of the agent.
18
And you know that going in.
19
So you'll often take time ahead of these AI agent sessions to break it down into smaller chunks say, oh, I'll just bite off this little bit.
20
I'll just bite off this little bit.
21
But then what you'll find is, okay, I'm working towards planning and this bit of grilling, and then you reach a question that you can't answer, or you just find yourself lost in fog,
22
and all the time you're managing the smart zone, you're trying not to spend too many tokens.
23
This has been out there for a while, and people are freaking loving this thing.
24
One-shotted a prototype, it kept starting again and again for months.
25
I really hate the phrase one-shotting, but I think what he means is it really helped him out.
26
John here even built his own freaking harness because he liked the Wayfinder approach so much.
27
It's got this gorgeous little star map on it that kind of lets you take tasks as you go.
28
So it's been out there for a little while and I'm finally making the video that people want me to make.
29
What is Wayfinder?
30
How do you best use it?
31
Well, let's start by looking at how big work typically gets planned.
32
You have a start point, a point where you need to start from, a sort of vague idea, not really how to get there,
33
and you're trying to get to some kind of destination.
34
You know vaguely where you want to end up, but the steps between are super foggy.
35
You've no idea how to get.
36
This is true, by the way, in engineering, but it's also true in many walks of life where you're planning something ambitious.
37
And so the first thing you should probably do is have a grilling session about it.
38
Get the AI to interview you and figure out the sort of basic premise of where you're going.
39
Now, for some work, that's sufficient and you'll be able to get straight to your destination.
40
But for a lot of work, that will still leave you in a lot of fog.
41
What you might find is based on that initial grilling session, you need to do more sessions.
42
So you might have a prototyping session or you might have another grilling session, or it might need to go off and do some research as well.
43
Conceptually, what we're looking at here is a map.
44
We are creating a map of how we're getting to our destination.
45
This is why it's called Wayfinder.
46
We are finding our way to the destination.
47
And each of these things on the map, they are tickets.
48
Each ticket requires its own individual session with the agent.
49
So you might have a prototyping session, a grilling session, and a research session.
50
And all of those things are created and managed by Wayfinder.
51
And just a note here, yeah, this is just a single skill doing all this, and it works with any coding agent.
52
On its map, Wayfinder gives you a frontier of tickets here.
53
In other words, the decisions that it knows about so far.
54
And it also keeps track of everything that's in fog.
55
So things that are not quite able to be decided upon yet
56
because we haven't done the research or we don't have a prototype to look at
57
or we haven't done enough conversation, enough grilling.
58
At some point, all of the fog will be resolved
59
and then you'll have finally made enough decisions to finally get to your destination.
60
Wayfinder can not only manage the research, but it can also do tasks here too.
61
So if you need to set up some configuration
62
or you need to go out and talk to someone and actually go and run an errand, then Wayfinder can figure that out for you as well.
63
In other words, all of the complicated stuff that you might need to do while you're planning something big, Wayfinder orchestrates it all for you.
64
It keeps track of everything that's been done and it measures the fog of war for you, keeps track of all the frontier of things you can decide right now.
65
How does it keep track of it?
66
Well, it does it in your issue tracker.
67
In my public course video manager repo, here are all of the Wayfinder maps that I've done recently.
68
And you notice that if we look at this one, this is the big old map here.
69
And underneath it are 12 subtasks or sub-issues.
70
And these are the decision tickets.
71
So we can zoom down here and we can understand all of the decisions that have been made.
72
As decisions get made, then obviously they get resolved inside the ticket.
73
So in this one, this is a sub-issue, close the clips during publish race, and we resolved it with a discussion a couple of weeks ago.
74
That resolution also gets written back up to the parent map.
75
So if we look back up here, we can see that a small version of that also gets written in the map.
76
And so Wayfinder is keeping track of all the decisions that have been made, all the prototypes that have been created, all the tasks that have been done.
77
And by the way, even though I'm using GitHub for this, my skills are issue tracker agnostic.
78
so you can use it with any issue tracker you like.
79
You just need to do a little bit of configuration via setup matpocot skills.
80
Use it with linear, use it with JIRA, use it with literally whatever you like.
81
The very first thing you'll need to decide when you kick off a new Wayfinder session is the destination.
82
For instance, in this one, I was adding a command palette with a bunch of new actions into my application
83
and what I ended up wanting was a buildable spec.
84
So I wanted a specification for this command K command palette in the CVM diagram window.
85
So I started it off like this.
86
I invoked the Wayfinder skill and then I gave it a description of what I wanted.
87
I would like the ability in the CVM to add an icon picker.
88
Not only that, I want the ability to search other diagrams.
89
I want the ability to copy things from the diagram and save them as, you know, big old chunk of work.
90
It went through and explored the repo and it invoked the grilling skill and it grilled me about what I wanted.
91
It first asked me what done looks like, whether I wanted a spec and it recommended a spec.
92
That's good.
93
And then it asked me a few initial questions before then going and creating some tickets and the first map.
94
And it created the other tickets as sub issues.
95
So we kicked off with seven tickets immediately.
96
However, only three of those tickets were takeable right now.
97
So figure out where icon names come from, component storage schema and palette information, architecture and grid keyboard.
98
I don't remember that one.
99
And so what I did was I then worked through each of those tickets in a new session.
100
The way I did that was I just called Wayfinder on that ticket name.
101
I did it in a slightly fancier way where I actually have a handoff skill
102
that automatically wrote me a prompt and spawned a clawed subagent.
103
But what it was essentially doing is just calling the Wayfinder skill on this map
104
and on the specific ticket wherever it was.
105
Yeah, here it is.
106
Here's your ticket.
107
Transpar Lucid SVG Geometry to Path Builder and it just mentions the full ticket name.
108
So this is how you work through a Wayfinder map.
109
You do an initial Wayfinder prompt just to chart the map and figure out the next ticket.
110
And then for each ticket, you say Wayfinder with the ticket URL.
111
So you use Wayfinder for both, both for charting the map initially and then walking through each ticket.
112
As you can probably see from this diagram, tickets can have different types.
113
And there are four types.
114
And these ticket types are actually brought into the issue tracker themselves.
115
So we actually have Wayfinder Research, which is a ticket type.
116
Research tickets are where the agent needs to go off and find some information and bring it back.
117
it usually kicks it off immediately.
118
So you don't actually need to watch it.
119
It does it in a sub agent and then reports back.
120
Prototype tickets, which are the next type here,
121
create a prototype, which is so unbelievably invaluable for really seeing things come to life as you're planning.
122
I've done a whole extra video on this, on how important prototypes are, and it reuses the prototype skill from that video.
123
Some folks look at Wayfinder and they think, God, that's a lot of planning.
124
Doesn't that look like waterfall?
125
And the prototypes are the way that you prevent it from becoming waterfall.
126
Huge amounts of low fidelity upfront planning.
127
A prototype is a high fidelity way to get feedback on what you're actually building.
128
And the fact that Wayfinder encourages you to build so many prototypes means that the output is unbelievably good.
129
So, so far we've got research, prototype, obviously there are grilling ones as well.
130
So, grilling sessions.
131
And this is just where you need a discussion over maybe an implementation detail, over a particular aspect of the plan.
132
And the final type of ticket are tasks.
133
These are things that need to be done in the real world, stuff that the agent can't quite do itself, or possibly sometimes the stuff agent can do itself,
134
but is scheduled behind other work.
135
One really cool thing about Wayfinder is the way that it establishes blocking relationships between tickets, because some decisions can only be made once other decisions are made.
136
And so what you end up with is here, we've got 14 out of 17 done on this map.
137
So a lot of work done, but we've still not built the skill that this whole map is built around.
138
And once we built the skill, then we actually need to revisit some other stuff based on how the skill works and how it actually improves things.
139
And so what you're doing a lot of the time when you're working through a Wayfinder map is going, OK, I've resolved that ticket.
140
Let's see how this opens up new tickets.
141
What has the frontier moved to?
142
So then once the map is complete, what do you then go and do with it?
143
Well, this one, because its detonation was a spec, the Wayfinder map is probably a little bit too dense to create a spec.
144
So what I like to do is create a spec from the map.
145
This was the spec that I created from it.
146
And you can see it's basically the same setup as I've had before.
147
I literally just called to spec on the Wayfinder map and it pulled in this enormous document
148
with basically all of the decisions that have been pulled from the Wayfinder map into this GitHub issue.
149
The initial draft was actually too large for GitHub's character limit.
150
So that kind of tells you how big it was.
151
And from there, I turned it into tickets using my usual approach, which is to spec and then to tickets.
152
In other words, Wayfinder fits in just in exactly the same place that grill with docs does in my usual approach.
153
So instead of doing grill with docs and then doing to spec and to tickets, You're spending a lot more time in Wayfinder, creating this enormous map, and then taking that map,
154
turning it to spec, turning it to tickets, and then implementing each ticket, and then running code review at the end.
155
The really cool thing about the Wayfinder setup is that the specs that it creates are so dense, and they all link back to the original decision tickets.
156
So you can actually go and the agent can go and view the primary source if it's confused about anything.
157
That was always a kind of weakness with Grill with Docs, which is that you were really relying on the spec to be the source of truth,
158
but the spec is always just a summary of what was actually said in the meeting.
159
Whereas now with Wayfinder you've actually got access to that primary source, which is amazing.
160
So that is Wayfinder.
161
It's a way of mapping huge chunks of work by planning things out really in detail ahead of time.
162
It can handle prototyping, can handle research, can handle arbitrary tasks, can handle discussions too.
163
Let's jump into an FAQ now of frequently asked questions that I get when people ask me about Wayfinder.
164
One is, this is way too much process.
165
This way too heavy for the kind of work that I do.
166
When should I actually use it?
167
Well, the answer to this is if you think the work
168
that you're doing can be completable and planable in a single session, then plan it in a single session.
169
If you kind of already know the way to your destination, then there's no need to use Wayfinder
170
because you can just path your way there in a single session and just figure it out.
171
Wayfinder is for the cases where you have the fog of war.
172
You have no idea quite where to go, and you just need to start and then see where you get to.
173
By the way, I've actually been using Wavefinder for non-coding tasks, so I've been meaning to put up a garden office in my garden, and I've been using Wavefinder for that.
174
So it's commissioning a site survey, figuring out all that stuff, figuring out who to contact, doing all the research, finding the different firms that could build it.
175
It's awesome.
176
Another response people have to Wavefinder is, this is SDD, this is spec-driven development, and I don't want to do spec-driven development.
177
I don't want to spend all this time putting together a spec.
178
This seems bananas.
179
Well, the way I think of specs is really just a destination for a multi-session piece of work.
180
In other words, we have a huge task down here, let's say task number four, that we're trying to schedule over multiple agent sessions because it's just too big.
181
And what we want to do is we need a spec so that we can, when we get to the end, figure out where we were going.
182
That's all a spec is in this context.
183
It's just a destination document to handle this multi-session work.
184
And then each session is done in an implementation ticket.
185
Also, this is people get confused when they first use Wavefinder because they go, right, it's creating some tickets.
186
Aren't we supposed to do the tickets later?
187
These are kind of implementation tickets versus decision tickets.
188
So in Wavefinder, you have decision tickets.
189
These are implementation tickets.
190
So the difference between my approach and most other approaches is that people, when they get to the end of this, they will keep that spec around somewhere.
191
For me, I close the issue you continue the spec and the spec is gone.
192
It's gone from my repository.
193
I rarely, if ever, refer to it again.
194
Once the spec is present in the code, then you can just delete the spec.
195
Whereas people who do spec-driven development go back to the spec and edit it and modify it.
196
There are lots of approaches to spec-driven development, so I'm probably annoying someone with that.
197
But what I'm essentially trying to say is that these specs are non-persistent.
198
With that, folks, I recommend you go off and you chart your own awesome, foggy idea.
199
I have found Wayfinder just so liberating in
200
that it just lets me get started and it handles all of that difficult decision for me.
201
I've been using it to plan courses, been using it to do engineering work, been using it to build a garden office.
202
It is just awesome.
203
The cool thing about it is that the destination is totally up to you, whether you want it to create a spec that you then run through an AFK agent, which is what I do,
204
or if you just want it to implement the work for you in tasks, then it totally can.
205
There is no more fun feeling than starting a new Wayfinder session and knowing that you're going to see something awesome, but not quite knowing how you're going to get there.
206
If you're into this stuff and if you want to keep up with my skills, then you should check out this seven lesson free course that I've put out on AI skills for real engineers,
207
which is on my AI hero site.
208
I'll add the link below.
209
This lets you build up a repeatable workflow that you can ship great work and it's all built on solid software fundamentals.
210
Thanks so much for hanging out.
211
It is always fun filming these sessions and I'm so glad that people are enjoying Wayfinder so much.
212
Cheers pals, I will see you in the next one.
同頻道
FOR YOU
推薦影片
本課的詞彙與口語要點
這節 B2 級別的口語課以影片“/wayfinder: Nothing is too big to plan anymore”為素材。 影片中反覆出現的詞有:ticket, Wayfinder, spec, session, map。 這段影片共有 212 個句子、3154 個單詞可供跟讀。 講話部分時長為 15:08。 說話人語速較快,每分鐘約 208 個詞,會出現較多連讀和弱讀。 84% 的單詞屬於英語最常用的 3,000 詞,其餘的詞建議在練習前先學一下。
影片中的重點詞彙
影片中最難的 15 個單詞,附發音和釋義:
| 單詞 | 發音 | 釋義 |
|---|---|---|
| grill 名詞 | /ɡɹɪl/ | 烤架 |
| prototype 名詞 | /ˈpɹəʊtətaɪp/ | 原型, 雛形 |
| chunk 名詞 | /t͡ʃʌŋk/ | 塊 /塊 |
| diagram 名詞 | /ˈdaɪ.ə.ɡɹæm/ | 圖 /圖, 圖表 /圖表 |
| resolve 動詞 | /ɹɪˈzɑlv/ | 解決 /解決 |
| spectacles 名詞 | /ˈspɛktəkl̩z/ | 眼鏡 /眼鏡, 眼鏡兒 /眼鏡兒 |
| palette 名詞 | /ˈpælɪt/ | 調色盤 /調色盤 |
| frontier 名詞 | /ˈfɹʌntɪə/ | 邊界 /邊界, 邊疆 /邊疆 |
| orchestrate 動詞 | /ˈɔɹ.kəstˌɹeɪt/ | 編管絃樂曲 |
| unbelievably 副詞 | 不可信地 | |
| fidelity 名詞 | /fᵻˈdɛlədi/ | 逼真度, 保真度 |
| waterfall 名詞 | /ˈwɔːtəfɔːl/ | 瀑布 |
| ambitious 形容詞 | /æmˈbɪʃ.əs/ | 有雄心, 遠大 /遠大 |
| configuration 名詞 | /kənˌfɪɡ.əˈɹeɪ̯.ʃən/ | 編排 /編排, 配置 |
| dense 形容詞 | /ˈdɛns/ | 密, 稠密 |
影片中出現的短語動詞
| 單詞 | 釋義 |
|---|---|
| figure out 動詞 | 弄清楚 |
| end up 動詞 | 到達 |
| check out 動詞 | 退房 |
| go out 動詞 | 出, 出口 |
| go through 動詞 | 通過 /通過 |
| put together 動詞 | 拼合, 組合 /組合 |
值得反覆跟讀的句子
影片中簡短而完整的句子,可以直接用在日常對話裡:
- I'll just bite off this little bit.
- How do you best use it?
- You've no idea how to get.
- When should I actually use it?
- Aren't we supposed to do the tickets later?
影片中的語法
說話人最常用的結構,並附上影片中的原話:
| 結構 | 影片中的用法 |
|---|---|
| 現在完成進行時 have/has been + -ing — 以前開始、現在仍在繼續的動作 | I've actually been using · I've been using |
| 現在完成時 have/has + 過去分詞 — 過去發生但與現在仍有關聯的事 | I've figured · I've created · I've packaged |
| 被動語態 be + 過去分詞 — 強調發生了什麼,而不是誰做的 | is based · are created · be decided |
| 定語從句 who / which + 從句 — 補充說明人或事物 | Research, which is · tickets, which are · prototype, which is |
需要注意的發音
說話人用了 64 次縮略和弱讀形式,例如 you're, I've, I'm。請按聽到的簡短形式來說。
- “sh” 和 “zh” 音: speculation /ˌspɛk.jəˈleɪ.ʃən/, ambitious /æmˈbɪʃ.əs/, configuration /kənˌfɪɡ.əˈɹeɪ̯.ʃən/, detonation /dɛtəˈneɪʃən/, publish /ˈpʌblɪʃ/
- 長單詞——注意重音位置: speculation /ˌspɛk.jəˈleɪ.ʃən/, configuration /kənˌfɪɡ.əˈɹeɪ̯.ʃən/, conceptually /kənˈsɛp.tjʊə.li/, detonation /dɛtəˈneɪʃən/, repeatable /ɹɪˈpiːtəbəl/
中文母語者容易讀錯的音:
- /v/ — 不要讀成 /w/,上齒輕觸下唇: resolve /ɹɪˈzɑlv/, invoke /ɪnˈvoʊk/, revisit /ɹiːˈvɪzɪt/, vaguely /ˈveɪɡli/, invaluable /ɪnˈvælju(ə)bl̩/
- 詞尾子音 — 要讀清楚,後面不要加母音: prototype /ˈpɹəʊtətaɪp/, chunk /t͡ʃʌŋk/, resolve /ɹɪˈzɑlv/, spectacles /ˈspɛktəkl̩z/, palette /ˈpælɪt/
- /r/ 和 /l/ 的區別: grill /ɡɹɪl/, resolve /ɹɪˈzɑlv/, liberate /ˈlɪbəˌɹeɪt/, repeatable /ɹɪˈpiːtəbəl/, linear /ˈlɪn.i.əɹ/
如何用這段影片練習
- 先完整聽一遍影片,不要開口,記下不認識的單詞。
- 先用 0.75 倍速逐句跟讀,熟練之後再回到正常速度。
- 錄下自己的聲音並與原聲對比,特別注意 grill, prototype, chunk 這類單詞。
什麼是跟讀法?
跟讀法 (Shadowing) 是一種有科學依據的語言學習技巧,最初開發用於專業口譯員的培訓,並由多語言者Alexander Arguelles博士普及。這個方法簡單而強大:您在聽英語母語原聲的同時立即大聲重複——就像是一個延遲1-2秒緊跟說話者的影子。與被動聽力或語法練習不同,跟讀法強迫您的大腦和口腔肌肉同時處理並模模擬實的講話模式。研究表明它能顯著提高發音準確性,語調,節奏,連讀,聽力理解和口語流利度——使其成為雅思口語備考和真實英語交流最有效的方法之一。























