JavaRush/Courses/Claude Code SELF/Why a fresh perspective matters

Why a fresh perspective matters

Open

1. Review is separate work, not finishing up

Tommy still has something holding him back. He has some doubts that everything is so simple. And he starts with:

“Listen, why do you even need a separate subagent? The game is done, it is in the gallery. What is there to inspect—and why not Sam himself? Why is he still not able to get it right?”

“And here are two different jobs that are easy to confuse.” Alex does not rush. “When you write code, your goal is to add. A new feature, a new screen, another piece. But understanding code is not about ‘adding more.’ It is evaluating what is already there. Finding what is wrong, what is inconvenient, what is broken. In English this is review, a review. And it is a different mode of thought: not ‘what else can I slap on,’ but ‘what is wrong here.’”

“So it is kind of… not finishing up?”

“Not at all, and it is easy to mix them up. Take construction. Some people lay bricks—they build. Then the inspector comes and checks whether everything is level: did the wall warp, does the roof hold? He is not building. He is evaluating what was built. And if the inspector starts adding bricks as he goes, he will fail both jobs: he will not finish the build properly, and he will not check it. Because these are different heads.”

On the left, Sam in a hard hat lays bricks while a blue subagent holds a level to the leaning wall and marks a cross; on the right, the subagent has dropped the level and builds alongside Sam — the wall is falling apart.

Left: Sam builds, the subagent checks — and immediately sees the wall has warped. Right: the subagent has started building alongside Sam — and there is no one left to check.

“So there is no way to do it with one head?”

“You can, but it is bad. When you are building, you are fully in the ‘what else can I add’ state—and against that backdrop you simply do not see that that wall over there is crooked; you have already gotten used to it. And acceptance is a deliberately different pass: not ‘what else can I slap on,’ but ‘what is wrong here.’ Professionals separate these on purpose: one thing is writing, another is reviewing. So ‘yeah, I will take a look as I go’ does not work—when you go as you go, you are exactly not looking, because your head is busy adding.”

Writing and reviewing are two different modes of the head. Writing: the goal is to add, the question is what else to slap on; on a construction site that's the bricklayer laying bricks; the result is a new feature. Reviewing: the goal is to evaluate what's already there, the question is what's wrong here; on a construction site that's the inspector checking whether it's level; the result is what's crooked, awkward or broken. “I'll take a look as I go” doesn't work: the head is busy adding Writing and reviewing: two different modes Writing Reviewing Goal add evaluate what's already there Question in your head “what else can I slap on?” “what's wrong here?” On a building site the bricklayer lays bricks the inspector checks it's level Result a new feature, a new screen what's crooked, awkward, broken “I'll look as I go” doesn't work: your head is busy adding

2. Why another person’s eye sees better: cleaner context

“Okay, review is a separate job,” Tommy agrees. “But why should SOMEONE ELSE do it, not Sam? He spent today disassembling his Snake himself—and he did not find the tail bug, even though I kept poking him in the face about it. He just missed it.”

“There, you answered it yourself.” Alex snorts. “And not because Sam is stupid. But because he built this game himself. Every line he wrote—and all of it is still in his head, in that very conversation where you were putting it together. The eye runs through the code and at every spot nods: ‘yes, that is how it was intended’—and slips past the place where it is actually crooked. He sees not what is in the code, but what he wanted to put there. A familiar thing: you reread your own email a hundred times and do not see a typo, but a colleague notices it in a second. Is he more literate? No. He just reads with fresh eyes.”

“And the fresh subagent does not have that ‘as intended’ picture in its head?”

“Not at all. It does not remember what Sam wanted. It did not see you assemble Snake. It looks exactly at what is actually lying in front of it. So it sees a crooked wall right away—for Sam it is familiar and native, and for someone else it is just crooked.”

Tommy slowly nods and sums it up himself—crooked, but in his own words:

“So the trick is not that the nitpicker is smarter than Sam. It is that his head is clean. Not clogged with our own build.”

“Exactly. Not more brains — cleaner context. Write that in gold letters; it is worth a lot. That is why ‘ask Sam to be stricter’ does not work: strictness does not clean the same blurred map. And a separate subagent comes to Snake without that tired map—for him it is just a game he sees for the first time. What for you is ‘yeah, that is how it was supposed to be’—for him is ‘wait, why does it jerk like that?’”

The subagent sees only what's in front of it. A fresh subagent sees the Snake for the first time and looks only at the code. Everything you and Sam stewed in — the whole conversation about the Snake, “we decided it this way because…”, “that's how it was intended”, the crooked wall you got used to — stays outside the frame; it doesn't have any of that. So it asks right away: “Wait, why does it jerk like that?” — bug found The subagent sees only what's in front of it Fresh subagent sees the Snake for the first time Everything you and Sam stewed in — it has none of it the whole Snake conversation “we decided so because…” “that's as intended” the crooked wall feels normal what the subagent sees { } the Snake code exactly what's in front of it “Wait, why does it jerk like that?” bug found Not more brains — cleaner context: the crooked stands out because there's nothing to get used to

And he adds, and there is something more serious in his voice:

“And note: while Snake was lying unfinished, the blurred eye was your internal workings. And as soon as it went to the gallery and became an external thing that other people open, a fresh look stops being a perfectionist’s whim. It is the normal way to support something already released. A release is not the finish line. It is the start of a thing’s second life, where it gets fixed and refined. And you should look at it not only through the eyes of the one who assembled it.”

flowchart LR
  A["Built it"] --> B["Release"]
  subgraph L["A thing’s second life"]
    C["A fresh look
through other eyes"] --> D["Fix and refine"] end B --> C D -->|"again"| B

At the end Alex drops the second side of the medal—casually, in passing:

“A subtle point for the future. A clear head is a plus. But it also means the checker does not know everything you and Sam were steeped in. Your whole context, all your ‘we decided it this way because…’—he does not have that. Today that will work in your favor: lack of smudging matters more. But the very idea that a separate helper may not take something into account—just put it in a far pocket. You will need it later.”

“In a pocket then a pocket.” Tommy nods, but puts the thought about “not taking something into account” away without looking at it; he does not need it right now. Instead, something else, something joyful, grabs him. “Listen, it turns out something interesting. I thought you could delegate only work: do this, write that. But it turns out you can hand off not only hands, but also a gaze. Not ‘do it for me,’ but ‘look for me with eyes I do not have.’”

“Now you have grown up.” Alex rejoices like a father.

“And further.” Tommy is already picking up speed. “If you can set up a separate fresh perspective, then I set up my own head for each task. The main agent writes. And a separate nitpicker subagent tears everything apart.”

1
Task
Claude Code SELF,  6 level,  2 lecture
Locked
Review findings: what I fix, what I reject ⚖️
Review findings: what I fix, what I reject ⚖️
Комментарии
  • популярные
  • новые
  • старые
Для того, чтобы оставить комментарий Вы должны sign in
This page has no comments yet