---
title: codegen
canonical: https://fjpaz.net/links/codegen/
markdown: https://fjpaz.net/links/codegen.md
---

# codegen

Channel: codegen-rley1hk4xzy · 336 blocks

## 1. Pair

### 1a. [@xlr8harder](https://fjpaz.net/links/codegen/51047690/) · [arena:51047690](https://www.are.na/block/51047690)

**xlr8harder** [@xlr8harder](https://x.com/xlr8harder) · [Oct 4, 2026](https://x.com/xlr8harder/status/2106808982765527430)

working with an agent when you define a very clear set of requirements, and when you're winging it, are completely different experiences.  the latter feels so much worse, in terms of the quality for what you end up with.

Capture: [capture.json](https://fjpaz.net/posts/x-2106808982765527430/capture.json)

### 1b. [@repligate](https://fjpaz.net/links/codegen/39672265/) · [arena:39672265](https://www.are.na/block/39672265)

**j⧉nus** [@repligate](https://x.com/repligate) · [Sep 18, 2025](https://x.com/repligate/status/1968797194808328686)

I think people who say that using ai for code (or in general) makes you less able to think for yourself are just telling on themselves re what they do when given access to intelligence and knowledge on tap.

I enjoy understanding things, especially on a high level, and find it important if I’m steering a project. When Claude fixes a non trivial bug, I almost always ask it to explain what the problem was to me. And by keeping on top of things I’m able to help it when it gets stuck a lot better too (especially given the non overlapping information and degrees of freedom I have access to).

![photo](https://media.fjpaz.net/posts/x-1968797194808328686/1968797194808328686-0.display.a0e75a61.webp)

Capture: [capture.json](https://fjpaz.net/posts/x-1968797194808328686/capture.json)

## 2. Pair

### 2a. [felixrieseberg.com](https://fjpaz.net/links/codegen/51359713/) · [arena:51359713](https://www.are.na/block/51359713)

![How to be more ambitious: A tutorial for humans with models](https://media.fjpaz.net/arena-media/51359713.325adc25.webp)

[How to be more ambitious: A tutorial for humans with models](https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/) · felixrieseberg.com

I work on Claude. It’s an interesting experience: Every day I learn about yet another feat accomplished with or by Claude that I hadn’t considered, didn’t think possible, or simply lacked the curiosity and time to try. This month, it may have helped discover a new planet,

### 2b. [@kcimc](https://fjpaz.net/links/codegen/50805224/) · arena:50805224 *[private block]*

**Kyle McDonald** [@kcimc](https://x.com/kcimc) · [Sep 26, 2026](https://x.com/kcimc/status/2103667482448027954)

i tried opus 5.5. for 15 years my bio has said "artist working with code". today i changed it to "artist".

---

[Sep 26, 2026](https://x.com/kcimc/status/2103668491693998102)

i'm not "working with code" anymore. i'm thinking in systems, and writing in natural language. i've only been programming for 25 years. any code i write is going to be vastly inferior to a frontier model.

---

[Sep 26, 2026](https://x.com/kcimc/status/2103669600127000765)

i also have this sense that the generating code is becoming an inseparable feature of daily life. it's so obvious it's not even worth mentioning. "artist working with code" is like saying "artist working with physical matter".

Capture: [capture.json](https://fjpaz.net/posts/x-2103667482448027954/capture.json)

## 3. Set

### 3a. [@she_llac](https://fjpaz.net/links/codegen/46363990/) · [arena:46363990](https://www.are.na/block/46363990)

**shellac!** [@she_llac](https://x.com/she_llac) · [May 20, 2026](https://x.com/she_llac/status/2057015799882100981)

software engineering with LLMs is sort of like diffusion
refining the program you want out of a noisy version

Capture: [capture.json](https://fjpaz.net/posts/x-2057015799882100981/capture.json)

### 3b. [@transkatgirl](https://fjpaz.net/links/codegen/49916777/) · [arena:49916777](https://www.are.na/block/49916777)

**kat** [@transkatgirl](https://x.com/transkatgirl) · [Aug 28, 2026](https://x.com/transkatgirl/status/2093368801236770863)

ngl, doing software development in the age of (good) LLMs feels far more like caring for a bonsai tree than doing engineering

---

[Aug 28, 2026](https://x.com/transkatgirl/status/2093370026447483015)

like, the codebase grows and changes in response to requirements, and your role is mostly just trying to keep it pretty and small

Capture: [capture.json](https://fjpaz.net/posts/x-2093368801236770863/capture.json)

### 3c. [@GarrettPetersen](https://fjpaz.net/links/codegen/50071323/) · arena:50071323 *[private block]*

**Dr. Dad, PhD 🔄🔼◀️🔽▶️** [@GarrettPetersen](https://x.com/GarrettPetersen) · [Sep 2, 2026](https://x.com/GarrettPetersen/status/2095213387395592645)

People who direct teams of software developers have been "vibe coding" for decades. The only novel thing is that the rest of us can do it now.

Capture: [capture.json](https://fjpaz.net/posts/x-2095213387395592645/capture.json)

## 4. [@alth0u](https://fjpaz.net/links/codegen/49643433/) · [arena:49643433](https://www.are.na/block/49643433)

**alth0u🧶** [@alth0u](https://x.com/alth0u) · [Aug 19, 2026](https://x.com/alth0u/status/2089902888780964089)

my entire job with these coding agents is down to:
\- whittling handholds so they are obvious to humans and agents
\- refining sides so they dont give splinters
\- putting knives away so nobody hurts themselves
\- reducing epicycles

and the only way to scale this has been with tooling that scales involving me

Capture: [capture.json](https://fjpaz.net/posts/x-2089902888780964089/capture.json)

## 5. [@geoffreylitt](https://fjpaz.net/links/codegen/44263120/) · [arena:44263120](https://www.are.na/block/44263120)

**Geoffrey Litt** [@geoffreylitt](https://x.com/geoffreylitt) · [Mar 8, 2026](https://x.com/geoffreylitt/status/2030783577013596301)

@ankrgyl a related lens that I've been enjoying: LLM codegen = JIT compilation in trad PL.

An LLM is essentially a dynamic interpreter for English - flexible and slow. On the go it can choose to "compile" deterministic code for hot paths: faster but more rigid

Capture: [capture.json](https://fjpaz.net/posts/x-2030783577013596301/capture.json)

## 6. Pair

### 6a. [@jefferyharrell.bsky.social](https://fjpaz.net/links/codegen/42971241/) · arena:42971241 *[private block]*

**jeffery --dangerously-skip-permissions** [@jefferyharrell.bsky.social](https://bsky.app/profile/jefferyharrell.bsky.social) · [Jan 25, 2026](https://bsky.app/profile/jefferyharrell.bsky.social/post/3mdba3hfgc72i)

Today, right now, we are effectively at the point where sufficiently describing a computer program is equivalent to source code for that computer program.

*[the page shows the text up to here]*

I'm not up on this, we might be at a point where describing a small convex object is equivalent to having manufacturing specs for that object.

> [`@elucidating.extradimensional.space`](https://bsky.app/profile/elucidating.extradimensional.space/post/3mdb4rcmtb22u)
> *[not shown on the page]*
>
> You know, one of the funnier uses of coding LLMs that I don't see disgust very often: 
>
> Let's say at work you know about a thing and you would love to open source that thing but if you were to try and write it it would be legal nightmare?
>
> But.. you're allowed to describe it in a white paper.

Capture: [capture.json](https://fjpaz.net/posts/bluesky-3mdba3hfgc72i/capture.json)

### 6b. [@GabriellaG439](https://fjpaz.net/links/codegen/44530013/) · arena:44530013 *[private block]*

**gabby** [@GabriellaG439](https://x.com/GabriellaG439) · [Mar 17, 2026](https://x.com/GabriellaG439/status/2033910008992444873)

New blog post: "A sufficiently detailed spec is code"

I wrote this because I was tired of people claiming that the future of agentic coding is thoughtful specification work.  As I show in the post, the reality devolves into slop pseudocode

https://haskellforall.com/2026/03/a-sufficiently-detailed-spec-is-code

Capture: [capture.json](https://fjpaz.net/posts/x-2033910008992444873/capture.json)
