Durant l'horari laboral faig de tech lead de backend en @KaveHome. En el meu temps lliure faig cosetes a @festivaleswiki 🐍 #python #django #beer
València
Joined April 2022
- Tweets952
- Following404
- Followers91
- Likes4K
Enrique Soria retweeted
The AI agent fatigue is real. Unfortunately, software development has historically been measured by lines of code. And now that coding agents can generate 100x more code than humans, software engineers are expected to do 100x more in the same amount of time.
But managing AI agents is much more exhausting than writing code or even managing human developers.
The context switching drains your mental energy. You’re constantly working on System 2 thinking, which much more exhausting than the flow of implementing a well thought feature with your headphones on. You have to keep tabs on all your agents and get them unstuck when they fall in stupid deadlocks.
And why can’t we slow down? Because the company next door is shipping 10x with Claude Code or Codex.
In the end, you reach a state of fatigue and psychosis that pushes you toward just accepting whatever the AI agent says and throw more compute and LLM tokens at the problem when they get stuck.
Fantastic for the business model of companies that sell AI tokens. Bad for your mental health and your product’s reliability.
AI was supposed to make us work less
I've never been this tired in my life
most days I have up to 10 agents running at the same time. one fixing a bug, one writing pages, one refactoring something I already forgot I asked for, 3 other I have no idea what they are doing but i suppose i ask it?
and I jump from conversation to conversation all day
compared to 5 years ago, I probably context switch 100x more
by the end of the day my brain is fried. completely toasted. nothing left, not even for the people I love
I call it the fried brain effect
thought it was just me but almost every founder I talk to lately says the exact same thing
we got superpowers, and somehow the pressure only went up. if you can ship 10x more, you feel like you have to ship 10x more. every single day
I don't have a solution
honestly I have no idea where this goes from here
anyone else feeling this?
Enrique Soria retweeted
Python 3.15 adds frozendict as a built-in type! 🐍
It works like a dict, but it's immutable. Once you create it, you can't change it. That also makes it hashable, so you can use it as a dict key or put it in a set.
Great for config data you don't want changed by accident.
Will you use it?
Here's a quick example:
Enrique Soria retweeted
Mucho avance esta tendiendo a darle peor calidad al Software, la calidad no solo es el código, la calidad viene en cada proceso, y se está tendiendo a la no revisión, se está forzando el crear Software lo más rápido posible, dejando de lado el tener un sentido para hacerlo.
Enrique Soria retweeted
“the danger isn’t that ai writes bad code- it’s that it makes writing unnecessary code 100% free” — @HarivanshRathi
Enrique Soria retweeted
With AI, it’s tempting to create big tickets because AI can handle big tasks, and it generates code so quickly.
But the benefits of small, focused tickets remain relevant in the AI era.
- Get feedback quicker.
- Avoids context bloat / compacting.
- Less chance of spending a bunch of time going in the wrong direction.
- Leads to small PRs, which are easier to review.
- I’m more likely to understand a small ticket, and to notice gaps, inconsistencies, and logical issues in the requirements.
- Small tickets lead to smaller releases, which reduces risk.
Enrique Soria retweeted
Getting people to care about what you built remains arguably the hardest part of building software
Enrique Soria retweeted
"Some programmers have aesthetic preferences about code and they need to get over it because aesthetics don't matter."
Yes, sometimes they don't matter. But aesthetic judgements are also fast. If they're calibrated to match useful properties of code, that's useful.
Enrique Soria retweeted
stop making ai agents memorize your entire codebase.
people thought that throwing more markdowns file into the codebase and calling it SDD whatsoever will help. or installing some shitty memory plugins will.
trust me, I've been there...
every markdown you add is another potential source of stale truth.
want an agent-friendly codebase?
- shallow, feature-based modules
descriptive filenames
- comments explaining why, not what
- tests as executable specifications
- explicit, enforceable boundaries
- code search before vector search
your agent doesn’t need a better memory.
it needs a codebase that’s easier to understand, navigate, and verify.
sometimes the best context engineering is just good software engineering.
what a plot twist 😁
something that keeps bothering me is how some people see slop as something that is happening to them, like they're some kind of helpless bystander powerless to do anything about it
excuse me but you are a software engineer, you literally learned how to make sand think and make computers do your bidding?! and now you can't do anything about it? please
the 2 things you should know to cut through all the noise:
1. you need to give your agent a way to verify its own work (i just posted about this so won't rehash)
2. your codebase is memory
the 2nd is not new either, we have always had lint rules, compilers with good diagnostics and strong type systems, LSPs, etc to help us write better code. there is no difference when agents write the code instead, you just need more of them, and stronger ones.
forget rules and skills. you can add those later.
yes, your codebase is a form of memory to agents! agents love extending what already exists. so every nit and every annoyance you've ever seen, every mistake and bad pattern you observe from agents, needs to turn into a constraint, or the agents will just keep repeating those same mistakes like a virus in your code. then you keep observing and repeating this process until your codebase is locked down enough that even agents with little context can one shot a high quality PR.
an even better direction is to refactor/rewrite your code such that agent anti-patterns become categorically impossible. use type systems to your advantage, use libraries/frameworks that reduce the architectural search space, or heck, create your own framework like we did for grok bot and design high quality primitives and well thought out APIs
Enrique Soria retweeted
Esto no es algo nuevo, al negocio siempre le ha interesado tener nuevas funcionalidades, no si la arquitectura escala o si es clean code.
Lo único que logró la AI fue darnos la capacidad de subir código espagueti y acumular deuda técnica a velocidad x100
Enrique Soria retweeted
do we have a a word for when you're enjoying someone's takes on coding with AI, then try the app they're building and learn it's complete vibecoded slop?
Enrique Soria retweeted
Code review comments are the only feedback where "this function does too much" lands harder than it would on the author.
Enrique Soria retweeted
I do not understand why the aggressive tech people don't understand this. The failure rate needs to be zero. Not 1%, not .1%. I am not letting copilot email my boss without oversight. The cost of the tail risks are way too high.
Enrique Soria retweeted
My main hesitation on turning it loose is the possibility of catastrophic mistakes, particularly having seen the hallucinations before. It's probably 99% accurate - is that good enough if it botches something massive?
On the small potatoes stuff, it's great.
Enrique Soria retweeted
how do i explain to my PM that i can't show any results yet because i spent the entire week making better architectural decisions, while my coworker Chad one shotted a comparable task in an hour and satisfied the business immediately?
i can't say "ask me in a year when you have to maintain Chad's code / when you see the CVEs", my performance review is today.
not implying anything here, it's a genuine question
