Luis Sanchez

I really want to be a software engineer, trying to.

[Mario Zechner @badlogicgames]

imma go use this for the big pi refactor in anger. it looks really helpful.

[Mario Zechner @badlogicgames]

new pipi bot brain just dropped

[Luis Sanchez @MrSanders]

@thorstenball Love it! Am curious about your ways of working. I feel stupidly asking this, but do you use management tools? Do you move by tickets and estimations somehow? Roadmaps?

[Luis Sanchez @MrSanders]

This is great but at the same time I see companies still working the usual way, focusing on estimating, tickets, and methodologies.

[Thorsten Ball @thorstenball]

An internal Slack post I shared with the Amp team a few weeks ago:

Building software is learning.

https://t.co/gasUel9aAv

[Matan Grinberg @matanSF]

If your daughter needs tutoring in algebra, you can probably find someone cheaper than Albert Einstein.

Giving every task to GPT5.5 or Opus 4.8 is overkill. Often times you can get the task done just as well, but 10x cheaper and 10x faster with a smaller model.

[kate @whoiskatrin]

🎈 we've just shipped agents sdk v0.14.0

you can now build agents with skills, messengers, schedules, and durable workflows on cloudflare

out of the box support for recurring tasks, think workflows, chat recovery, mcp transport improvements, and better client-tool continuations

go make cool stuff with it!

[Luis Sanchez @MrSanders]

@whoiskatrin You are delivering so many useful things. Might I ask you how are your ways of working ? Are you still using management tools to track stuff? estimates? Roadmaps? Am really curious 🙏

[Luis Sanchez @MrSanders]

Am really curious about how @CloudflareDev ways of working are? Do you still estimate ? What methodologies do you use? Do you use ticketing systems (jira, linear etc)? How are you measuring value?

From what I see is crazy we are still moving at the pace of poker, tshirt estimations, and tickets. But I really want to learn if it’s like this still.

[Luis Sanchez @MrSanders]

there are real gaps in the industry: there are teams delivering real value vs there are teams buried in bureacrecy and old metodologies.

[Luis Sanchez @MrSanders]

@dillon_mulroy am really curious about companies like cloudflare how do you work? i see in my current job that we are still talking about scrum, estimations, and i cannot stand it, i feel is kind of outdated ways of working... I dont see people talking about it. am i wrong to hate estimations? specially in the ai age? sorry to tag you or maybe @badlogicgames ? help me go out of my frustration

[Luis Sanchez @MrSanders]

Crazy and I would like to. But, Is kind of an excuse or not, but not everybody has the money to go by their own.

[Mario Zechner @badlogicgames]

that pewds becomes a local inference fighter was not on my bingo card. but here we are. wildest timeline. the codebase isn't even super terrible.

https://t.co/jJLkYP3rXf

[Luis Sanchez @MrSanders]

@antonioleivag La carga cognitiva que mencionas es totalmente real yo pensaba que era que me había vuelto perezoso para leer

[Carlos Santana @DotCSV]

El paradigma de ingesta del segundo cerebro de Karpathy, donde tienes una carpeta donde echar tus datos en crudo y que luego un agente procesa y estructura para agregarla a una wiki o segundo cerebro, es un patrón escalable a otras tantas aplicaciones.

Yo en mi caso por ejemplo ya tenía creada una aplicación financiera que usaba un sistema similar: mis extractos bancarios, facturas, datos traídos de APIs, modelos de impuestos en pdf... todo en crudo en una carpeta. Con la idea de luego llamar a un agente que trabaje en dar orden y forma a esos datos (una única vez) para procesarlos adecuadamente de cara a que luego lo consuma una aplicación (en este caso en vez de Obsidian, un front-end).

Se me antoja como un nuevo tipo de aplicación con un patrón arquitectónico que funciona por poner en su diseño a un agente que cada cierto tiempo sale a pasear para dar orden al caos de la carpeta de datos. No es un script determinista que sepas que va a funcionar siempre igual, con lógicas encorsetadas a formatos concretos, sino que tiene la flexibilidad de comerse y procesar cualquier dato crudo que pongas en la carpeta.

Y donde además cualquier dato alimenta al sistema y lo mejora para hacerlo crecer.

Además, obviamente los agentes no sólo actúan como procesadores de esa información sino que luego se nutren de todo el sistema para poder hacerle consultas mucho más completas o hacer crecer tu aplicación con cada nuevo dato crudo que se agrega.

Estamos empezando a diseñar software alrededor de datos caóticos, confiando en las capacidades de una nueva capa agéntica. El usuario no se adapta al software sino que el software se adapta al caos del usuario.

So good

[Luis Sanchez @MrSanders]

@kitlangton @thdxr 😂😂😂

[Antonio Leiva @antonioleivag]

Las novedades de Codex de ayer vuelven a estar desperdigadas, así que te hago un hilo para que no te pierdas nada.

El primero y más importante (aunque quizá en Europa no mucho) es que Computer Use ya está en Windows!

Y también funciona la App de ChatGPT con Codex en Windows para usarlo en remoto.

[Luis Sanchez @MrSanders]

Watch the entire video. I feel the attention problem with many terminals and windows... i tried to move to an ide many times and i cant i hate the learing curve even if is small, but i will try again with zed. another one is de visuals, asking the llm to put in flow or diagrams to build this mental model of whats going is great.

Crazy being able to work in so many projects at a time. great video!

[Luis Sanchez @MrSanders]

@zeeg They cannot do you can’t prompt

[Luis Sanchez @MrSanders]

@miilad2025 @dillon_mulroy No he have share his pi session

[Luis Sanchez @MrSanders]

Thanks for sharing of the session @dillon_mulroy i learnt a lot. This is the analisis done with an agent for the prompts used. Curious to me when you drop everything and ask to rebuild the thing entirely.

**P1 — Trigger an architecture review (anchored)**
I'm not happy with the patterns/design of @<file/module>. Review its integration/composition with @<collaborator>. Study the patterns in <reference codebases/libs> and tell me where the design's locality/cohesion breaks down.

**P2 — Plan as iterable text (no report/file)**
Don't generate a report or a file: give me the analysis as text in the message so we can iterate on it line by line right here.

**P3 �� Structural artifact (the core) ⭐**
For each of the <N> options, sketch in <language> pseudocode, concretely and concisely (these are sketches, not final code):
1. Type API / public interface of the seam
2. Call stack / call graph from entrypoint down to the leaf
3. Seams (where behavior is injected)
4. Adapters (production implementation vs test/in-memory)

**P3‑bis — Prod‑vs‑test call graph (the viral artifact)**
Show me the final call graph in two versions —Production and Tests — so the shape is identical and the only thing that changes is the injected layers/adapters (real vs in-memory).

**P4 — Targeted review loop (revise, don't rewrite)**
On "<snippet/line X>": <concrete change>. Apply only this and re-emit the affected section — don't rewrite everything.

**P5 — Consolidate into a spec**
Consolidate everything we agreed on into a single, complete tech spec.

**P6 — Refine the ubiquitous language**
What would be a better name for <concept>? → Use <NewName> and re-emit the full spec with that rename applied throughout.

**P7 — Encode a cross-cutting convention (and persist it)**
Rule: don't use <antipattern> in <context>; instead <correct pattern>.
Encode this in AGENTS.md if it isn't already. Then re-emit the spec with that change.

**P8 — Hand off to implementation**
Implement the spec at @<spec-path> using red-green-refactor TDD.
Start by bootstrapping <foundation>. Hard constraints: no <X>, no <Y>, no <Z>.
When you're done, tell me what's left in ≤5 todos.

[Dillon Mulroy @dillon_mulroy]

my "plans" largely look like pseudo code composed of mostly types/interfaces, how they compose, and their boundaries

ive recently started including call stacks - been very helpful for both me and agents when implementing