Eric retweeted
We need to reconsider our priors on code readability.
Humans, compared to agents, are pretty slow at understanding large quantities of text. So in the old days, readability was important so the human could efficiently grok what was going on.
So readability now comes down to:
1. Can a human, if needed, jump in and understand the code
2. Does the code use the same domain language as the human describing the code?
1 helps with observability and the occasional spot checks you need to do on important PR's. 2 helps the agent translate the words you use into code changes.
In other words, readability is now a nicety, not a necessity.
The way agents 'read' code is different - good agents are all about gathering data about how the code works. They'll prod around with it, execute parts of it.
So 'readability' to agents really means 'verifiability'. How easy it is to mess about with that code to see it working.
Let's take this scenario:
- You have a huge module in your codebase that uses all sorts of complex performance tricks
- It has 100% test coverage
A talented human would need an hour or two to get comfortable working in that module.
An agent needs zero time. They verify it with the tests, poke about a bit, and they're good to go.
The takeaways:
- When judging a module (or a codebase), don't think about how easy the code is to read
- Instead, think about how easy it is to verify
- Design your codebase so modules can be tested and verified independently
- Feel free to sacrifice a bit of readability for performance
- Cyclomatic complexity and function size don't matter to agents
Replying to @thdxr
@thdxr thebrewmap.app/
Taking all the product knowledge I gained as a startup engineer. I'm proud of %70 of the code, and it's mostly AI Gen. ADRs and specs and tests drive it.
A automated public changelog via release-please and GH actions are a sub tool within it.
14
Eric retweeted
Announcing mattpocock/skills v1
- Achieved a 63% reduction in token cost for skill descriptions
- Split skills into model-invocable and user-invocable skills, adding /codebase-design, /domain-modeling, and /grilling
- (UPDATED) /writing-great-skills - rewritten from the ground up, encoding my skill-writing best practices
- (UPDATED) /diagnose -> /diagnosing-bugs - now model-invocable, awesome for fixing hard bugs
- (NEW) /ask-matt: a router skill that teaches you how all the engineering skills work together
Replying to @rauchg
@rauchg thebrewmap.app/ - it's like a to-do list for breweries/distilleries.
Gpt4.* And some grunt work for setting things up, but I switched to Sonnet 4.5/4.6 for the major "production ready" lift.
Still in progress.
36
Replying to @antoinecojp @mattpocockuk
The flow seems similar to forking a session, then either summarizing or compacting that fork and re-ingesting back into the 'parent' session.
1
99
Replying to @kentcdodds
I'm maxing onto adaptability. Long held ritualistic patterns are drying up. If the only constant is change, then my staying power relies on flexibility. I know how good programs work, and I know I don't know a lot. This won't change. So I focus on idea communication.
11
I don't often see Reddit agree on things. But a small sample set apparently agree with you @tannerlinsley on directives reddit.com/r/javascript/s/Bo… tanstack.com/blog/directives…
Me? I'm just dealing with fomo here and my SPA 😂
5
308
Replying to @VicVijayakumar
I have stupid strong opinions about stupidly inane aspects of coding. When I started to recognize little contradicting patterns, I started to develop a generalistic rule: if I can't make the linter recognise it, I'll not enforce it on others.
19
MacDonalds is more Canadian than Timmies now days , atleast since the Brazilian majority RBI took over and MacDo scooped up their coffee supply.
1
1
30
Replying to @VicVijayakumar
> customers asked for X, product thinks they would be served better by Y, and here’s how to ship Z which accomplishes X ∩ Y without busting the budget or accumulating too much technical debt
This line of thinking, and ability to comm it with salient points, drives career growth
32
Replying to @housecor
If you write up an ADR, flip it to Copilot in Ask mode for evaluation, then use that ADR for Copilot in Agent mode you can get wildly successful results. In PR phase you can also batch feedback to fix up multiple things in one sweep.
As always with AI output is as good as input
1
19
Replying to @bduffez @VicVijayakumar
I've got myself a MotionGrey standing desk. They advertise supporting people up to 7 feet tall. motiongrey.com/products/moti…
I can attest that:
- the table is strong enough for me to stand on
- at full height I can almost stand under it
They have a 4 legged one too.
1
2
48
Replying to @housecor @james_mtc
I believe this is the way. As much as I use dev and local environments, (along with load/UAT/DevOps envs) I think in a 'pure" fashion we only really need Prod, Preview/Ephemeral envs, and temporary custom sliced release envs to verify code/app integrity.
2
130



