Skip to content
Josh Pazmino
ProjectsLinksAboutColophon

Links

codegen

336 blocks13 of 336 blocks13 items

How to be more ambitious: A tutorial for humans with models

How to be more ambitious: A tutorial for humans with models

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,

Oct 13, 202651359713⁠
Panda@VivaLaPandaThis block isn’t public yet.
The programming models often struggle to write code and design systems well, and one common thread to me is an inability to separate "the instantiation of a system" from "the abstract design of the system".

Like, you can have a queue and events and stuff, without ever specifying SQS vs RabbitMQ. The choice between those two matters, but its often very important to be able to reason through a design in an implementation-agnostic way, so that you can leave yourself room to play around with different potential implementations.

From the opposite side, when you are totally in abstract land, you sometimes miss ways in which any particular implementation might break the abstraction. Often if you find yourself struggling with the implementation over and over, its because the abstract structure you've created doesn't actually describe the problem you want to solve correctly.

I think being a good engineer often involves holding both in your head, and moving between the two fluidly. Being able to say "what is the thing this system is even trying to do" -> "is it, as constructed, doing that thing effectively" -> "if not, is it because I implemented the abstraction poorly, or because I picked the wrong abstraction"

One way I recently ran into this was when Fable was arguing with me about merge logic. I have N sources of data, and I want to view it in a log in once place. To Fable: the "thing viewing the log", the "API(s) which populate the log" and "the datasources from which those APIs pull" are all distinct things.
However, if we treat it all as one abstraction, we know that *no matter what we choose*, we're writing ~similar merge logic somewhere (either on the frontend, on the data access layer, in the SQL queries, or in the way the DB is populated). The question is just where the best place to put that logic is.

But Fable just will never naturally think to turn the problem of "how should the frontend render these various sources of data" into "what does the merge look like in each of the places it could potential occur, and which is best". Its too focused on the particular implementation details to zoom out and see the system as a whole.
Nov 11, 2026⁠1 replies2 reposts32 likes52645274⁠This block isn’t public yet.
shellac!@she_llac
software engineering with LLMs is sort of like diffusion
refining the program you want out of a noisy version
May 20, 2026⁠5 replies3 reposts51 likes46363990⁠
kat@transkatgirl
ngl, doing software development in the age of (good) LLMs feels far more like caring for a bonsai tree than doing engineering
kat@transkatgirl
like, the codebase grows and changes in response to requirements, and your role is mostly just trying to keep it pretty and small
Aug 28, 2026⁠44 replies130 reposts3,172 likes49916777⁠
Thorne 🌸 needs a job!@ens0.meThis block isn’t public yet.
Before AI, there would be times where I would be stumped by a bug so much I would just have to step away and look at it again with fresh eyes hours or days later. Basically all SWEs had moments like this.

But it's just not something I experience anymore at all
Oct 10, 2026⁠6 replies6 reposts105 likes51251727⁠This block isn’t public yet.
zach lieberman@zachliebermanThis block isn’t public yet.
I feel like the opposite now in some way I can’t fully articulate yet !
coding was never the point for me but it was always about the kinds of connections computation and software allow you to make. I feel the urgency of that now more than ever
Kyle McDonald@kcimc2026-09-26
i tried opus 5.5. for 15 years my bio has said "artist working with code". today i changed it to "artist".
Sep 26, 2026⁠9 replies3 reposts169 likes50813799⁠This block isn’t public yet.
chia@nonperformantThis block isn’t public yet.
vibe coding could be like "i deeply explored this domain, its existing artifacts & efforts, consulted its experts, and made something generative that was technically expensive before" but it's often like "i made imdb but slightly different and a lot lot worse"
Aug 14, 2026⁠0 replies0 reposts114 likes52566017⁠This block isn’t public yet.
XH@xhflozThis block isn’t public yet.
I keep seeing new projects from people I admired that feel « uninhabited » —— as in, I would've gotten 99% of the same experience had they given their original prompt

A big reason for this is, I think, related to « no tears in the reader, no tears in the writer », or — « no surprise in the maker, no surprise in the audience »

Because AI programming is so back-and-forth, rarely do I get into the zone, so I also rarely get immersed and totally absorbed enough to enter into a kind of flow state, where you're so singularly focused that you stop rationalizing + self-critic'ing, and just try things

At best, when I'm in that state, I tend to arrive at things that surprise me. Said stronger: I only surprise myself when in this self-absorbed flow.

And the surprising things often end up being the truest/most self-revelatory. This happens when I am designing, writing, having a conversation — it's not just in "creative practice," but a kind of inhabited enthrall that can happen at high and low planes

x.com/xhfloz/status/…⁠
XH@xhfloz2026-08-28
I spent the week barely using chatbots — just reading, thinking, and writing — and feel much more clarity than I have in months

Reminds me of this Oliver Sacks quote about clarity after his 'false productivity' of using amphetamines
XH@xhfloz
I have increasingly few qualms with AI programming, but this remains a big one: that it makes tools + products + experiences too outcome-oriented.

For many industrial things, that's okay, if not desirable. It's what technology's generally trended toward
XH@xhfloz
But a personality only comes through in the self-obliviated inhabiting during the making — flow, if you can shake the productivity guru connotations

Recently, after a months long rut, I've come to see flow — "autotelic" — as one of the most perennial sources of fulfillment
Sep 25, 2026⁠1 replies1 reposts32 likes50769677⁠This block isn’t public yet.
XH@xhfloz
Had a long rut recently, much due to AI, and wrote up some thoughts x.com/i/article/2103…⁠
XH@xhfloz
And if there's anything to take away from this, it's this:

"I've come to realize only recently that moments of flow are my primary sources of fulfillment and being in life, and without them, it's not just I'm "not productive," but rather I'm entirely what Maslow called "an experiential zombie." A lot of my work today is thinking about + designing experiences for flow."
Sep 25, 2026⁠16 replies9 reposts66 likes50780160⁠
XH@xhfloz
IMO Tools for Thought was a warning shot to show the folly of creating systems that were only ever used to improve said system — or worse: unused & only ever theoretical

Building with/on AI has a much much more addictive game loop & deceptive plausibility trap. Godspeed
max@maxbittker2022-10-23
bro are you seriously developing tools without insights drawn from a serious context of use right now...
Oct 21, 2026⁠1 replies2 reposts41 likes51659512⁠
lumpenspace and 99+ others@lumpenspaceThis block isn’t public yet.
it's open source in that you can show this post to astra.
Sep 7, 2026⁠0 replies0 reposts14 likes51133039⁠This block isn’t public yet.
Liminal Warmth ❤️‍🔥@liminal_warmthThis block isn’t public yet.
People really haven’t updated yet on what it means that everyone can have whatever they want (in software/games/media) just by describing it soon

The world is going to look so weird in just a few years
Sep 29, 2026⁠11 replies0 reposts114 likes50876020⁠This block isn’t public yet.
norvid_studies@norvid_studiesThis block isn’t public yet.
programming by hand
croissanthology in SF@croissanthology2026-06-14
no I'm just noting the passage of time. I've never programmed by hand basically ever so the evolution of what custom software I could make has always been directly linear with the quality of models and harnesses and it's crazy and cute to think I can qt a past self who was doing things with sonnet 3.7. Most of my website is sonnet 3.7! imagine if fable 5 instead
Sep 20, 2026⁠3 replies0 reposts17 likes50606062⁠This block isn’t public yet.
previousAI analogiesnextsoftware development
Are.na⁠Bluesky⁠Twitter⁠GitHub⁠LinkedIn⁠Email⁠