Ask HN: What trick of the trade took you too long to learn?

14 unsupp0rted 17 8/4/2025, 5:39:59 PM
Every week for the last 3 months I’ve learned a new trick when it comes to getting whatever LLM I’m using at the time to produce better output. That’s my trade, but lots of HNers have more interesting trades than that.

In my case, only recently I learned the value of getting an LLM to write and refine a plan.md architecture doc first, and for it to break that doc down into testable phases, and then to implement phase by phase.

Seems obvious in hindsight. But it took too long to learn that that should be my approach. I had been going phase by phase myself- no overarching plan.md for the LLM.

What Trick of the Trade took you too long to learn?

Comments (17)

oumua_don17 · 27m ago
Start as early as possible in investing (in index funds) and otherwise being financially savvy. It is very beneficial to realise early on that growing your hard earned money and spending it wisely is way more important as it will in the future lead to some unexpected benefits. Freedom of thought and action!
Terretta · 10m ago
Call Vanguard: 877-662-7447

An investing prof at Chicago puts this on the whiteboard at the start of semester, saying this is really all most people need to know and this class is unlikely to learn anything in his or any class that will let them, personally, do better.

yummypaint · 11m ago
Most problems (including analytically intractible ones) can be modeled with a relatively simple monte-carlo simulation. Most simple monte-carlo simulations can be fully implemented in a spreadsheet.

Using timing coincidences in particle physics experiments is incredibly powerful. If multiple products from the same reaction can be measured at once, it's usually worth looking into.

Circular saws using wood cutting blades with carbide teeth can cut aluminum plates.

You can handle and attach atomically thin metal foils to things by floating them on water.

Use library search tools and academic databases. They are entirely superior to web search and AI.

abetusk · 5m ago
I feel like the Monte-Carlo simulation modeling trick is one I picked up intuitively but only recently heard formalized. Do you (or anyone else) have a list of example problems that are solved in this way? Like a compendium of case studies on how to apply this trick to real world problems?
zappb · 1h ago
Writing tests first is a good way to end up with testable code. If you skip that, retrofitting tests is incredibly difficult.
atomicnumber3 · 21m ago
While it can be a useful forcing function, I find it also just fits into a higher velocity workflow in general, one I would describe like so:

1. Make PRs small, but not as small as you can possibly make them.

2. Intend to write at least one nice-ish test suite that tests like 50-80% of the LOC in the PR. Don't try to unit test the entire thing, that's not a unit test. And if something is intrinsically hard to test - requires extensive mocking etc - let it go instead of writing a really brittle test.

3. Tests only do two things: help you make the code correct right now, or help the company keep the code right long term. If your test doesn't do either, don't write it.

4. Ok - now code. And just keep in mind you're the poor sod who's gonna have to test it, so try to factor most of the interesting stuff to not require extensive mocking or shit tons of state.

I've found this workflow works almost anywhere and on almost any project or code. It doesn't require any dogmatic beliefs in PR sizes or test coverages. And it helps prevent the evils that dogmatic beliefs often lead you into. It just asks you to keep your eyes open and don't paint yourself into a corner, short term or long term.

AnimalMuppet · 3m ago
One more piece - if the test would be hard to write, use that to drive you to clean up the architecture of that piece of code.
ryoshoe · 24m ago
Even if you aren't going to do the complete test code just writing down the expected checks for each test makes things so much easier
MrDresden · 5m ago
Technical skills will only take you so far in most orgs. Learn to play the soft game.
WalterBright · 11m ago
That macros ruin everything they touch. I used them extensively for maybe 15 years, stopped adding them, and then a bit later removed them all from my code. My C/C++ code was a lot nicer without them.
vouaobrasil · 2h ago
I only learned this in the last five years: do less, automate less, do more by hand, and use the limited capability of the manual method to really choose projects that are worthwile, rather than aim for maximum efficiency.
juliansimioni · 11m ago
Similar to this: if you want to optimize your productivity*, do so on a timescale of at least weeks if not months or years.

Simple example: Can you get more done working 12 hours a day than 8? Sure, for the first day. Second day maybe. But after weeks, you're worse off in one way or another.

It's easy to chase imaginary gains like automating repetitive tasks that don't actually materialize, but some basics like sleep, nutrition, happiness, etc are 100% going to affect you going forward.

* I actually hate that word, and prefer saying "effectiveness". Productivity implies the only objective is more, more, more, endlessly. Effectiveness opens up the possibility that you achieve better results with less.

cwmoore · 11m ago
Observe as much mastery as possible at what you would like to do.
Terretta · 8m ago
Apprentice, journeyman, master: worked for millennia, will work for millennia more.
entrepy123 · 2h ago
How do you maintain tests, in order for LLM edits to not keep breaking things?

  - As a formal test suite in the program's own language?
  - Or using a .md natural language "tests" collection that must pass, which an LLM can understand?

To answer the OP, I learned use different models for reasoning vs. coding.
TMWNN · 15m ago
When I realized that a) `screen` exists and b) what it does, I felt like an utter fool for having gone for years—YEARS—without benefiting from it.
jshprentz · 10m ago
Drum cards