Gustavo Ayala

[Gustavo Ayala @GusCaos]

@ui_jedi The problem are not the mistakes, the problem are the kind and moment of mistakes.

[Bilgin Ibryam @bibryam]

Why AI hasn’t replaced software engineers, and won’t
https://t.co/e9T33TTntG

[Gustavo Ayala @GusCaos]

@paulabartabajo_ The thing most people don't get is that better structure/redability for humans is better redablity for LLMs. A well organized, abtstracted, layered, modular, with good cohesion and decoupling where makes sense and good names, is easy to read for LLMs.

[Gustavo Ayala @GusCaos]

@thdxr Have you seen any “best software of your life” lately? I’ve seen a lot of slop delivered from big companies full of financial, intellectual, and infrastructure resources. Something is very odd in this whole narrative

[Gustavo Ayala @GusCaos]

@GaryMarcus I had an obsidian vault, and used opencode with deepseek. It interpreted wrongly a symlink did a bad copy and when trying to revert erased the whole one year knowledge vault. Luckily it was synced with gdrive, and could restore it.

[dax @thdxr]

guys stop trying to argue with the "code doesn't matter" posts and just ask them if they're so ahead of the game why aren't they richer

[Lisan al Gaib @scaling01]

an intelligence company is the competitor of every company on earth

companies need to realize this and act accordingly

they are trojan horses, gobbling up your data and hollowing you out the same way globalization destroyed the industrial base of many countries

[Gustavo Ayala @GusCaos]

@KentBeck But LLMs also benefits from understanable code. Good abstraction, encapsulation, cohesion, appropriate coupling, modularisation, etc. makes easy to an LLM understand something with low tokens, and avoid overload context, and then give rise to tipical LLMs errors and hallucination

[Pau Labarta Bajo @paulabartabajo_]

My 2 cents on AI coding models in mid-2026:

> RL optimism is justified where goals are clear. Unit tests are the perfect reward signal. This part I believe.

> What I do not believe is the "autocomplete your codebase" pitch and the "spec2code" mantra. I still spend more time correcting GPT5.5 than coding.

> And the training data problem is real. The web is now full of AI slop. Models are quietly post-training on their own worst output.

> We may have already passed peak coding model quality on everyday tasks. More compute is not going to fix a poisoned data distribution.

[Gustavo Ayala @GusCaos]

@allenholub Was reading XP, code is not just for features, it is a precise way to encode behavior or the conception of a problem, trying to code forces a clear conception, this at the same time feeds back to the original problema formulation, and bot evolves mutually. AI wipes all that

[Lisan al Gaib @scaling01]

Fable 5 refused 200 out of 200 ProgramBench tasks lmao

[Gustavo Ayala @GusCaos]

@ID_AA_Carmack Yes, that is tradicional coding practices. What makes something more understandable for a human makes it for a LLM also. Abstraction, encapsulation, managing coupling and cohesion, using meaningful API for abstractions, meaningful structure of the files, etc.

[Gustavo Ayala @GusCaos]

@ValerioCapraro @GaryMarcus This a big point in coding. Code is not just a way to get a feature. Is a way to formulate and comunicate a problem. An with this, is a way to understand it. The way of understanding it gives way to other ways to see the problem, i.e. creativity

[Gustavo Ayala @GusCaos]

@ThePrimeagen @bcherny Lets say that you have to code a program that finds the square root of any number. How would you be sure the program is correct?

[Edgar Allen Poe @allenakinkunle]

📌

[Florian Roth ⚡️ @cyb3rops]

I think AI coding hype follows roughly four stages:

1. Amazement

You try it and can’t believe how much code it generates from a few prompts.

2. Expansion

You start more and more projects because shipping suddenly feels cheap and fast.

This is also the phase where people start convincing everyone around them:

- coworkers
- management
- friends in other companies

because nobody wants to “fall behind” in 6–12 months.

That creates a massive snowball/FOMO effect.

3. The grind phase

You realize the generated code has architectural issues, sloppy mistakes, weird abstractions, duplicated logic, broken edge cases, etc.

So you start:

- re-prompting
- switching models
- increasing reasoning effort
- reviewing fixes
- generating fixes for previous fixes

And suddenly you spend your days reviewing AI-generated pull requests instead of building software.

4. Realization

You realize AI coding increases output much faster than it increases certainty.

The code still needs:

- review
- testing
- ownership
- architectural understanding
- long-term maintenance

Usually by expensive senior engineers.

And the interesting thing is:
this whole cycle can take many months or even more than a year because people become socially and professionally invested in the narrative themselves.

Once teams, managers, and entire companies have been convinced that this is the future, it becomes psychologically and politically very hard to later say:

“Actually, the ROI is much lower than we expected.”

[Addy Osmani @addyosmani]

https://t.co/jKCIAEzai7

[kirbycrypto @kirbyongeo]

It’s been 6 hours since.
How is it still down?

Aren’t there like 4600 (86%) employees left?

[Adam Cochran (adamscochran.eth) @adamscochran]

The day after the CEO lays off a ton of staff and says:

“Non-technical teams are now pushing code to production with AI”

@coinbase has a major outage on their trading engine, and even their status page doesn’t work.

😂