So I tested Laguna S2.1 and the results are: it is a strong coding model for its size, in certain coding tasks it performs better than DeepSeek v4 Flash (but we are waiting for an update). It is NOT a general purpose LLM, very specialized for coding apparently.
July 23, 2026 at 10:46 AM
So I tested Laguna S2.1 and the results are: it is a strong coding model for its size, in certain coding tasks it performs better than DeepSeek v4 Flash (but we are waiting for an update). It is NOT a general purpose LLM, very specialized for coding apparently.
"I cannot — and would not — trust a team member who vibe codes in a production application. The constant negligence I observe when "Vibe Coding" is atrocious and unacceptable to a customer base of any size." cendyne.dev/posts/2025-0...
"Vibe Coding" vs Reality
Reviewing the capabilities and limitations of LLM agents in software development and their impact on skilled and less skilled developers.
cendyne.dev
March 21, 2025 at 1:29 PM
"I cannot — and would not — trust a team member who vibe codes in a production application. The constant negligence I observe when "Vibe Coding" is atrocious and unacceptable to a customer base of any size." cendyne.dev/posts/2025-0...
Source: @sentdefender on X
"Footage claimed to show a Unitree H1 (Full-Size Universal Humanoid Robot) going berserk, nearly injuring two workers, after a coding error last week at a testing facility in China."
Nightmare fuel.
"Footage claimed to show a Unitree H1 (Full-Size Universal Humanoid Robot) going berserk, nearly injuring two workers, after a coding error last week at a testing facility in China."
Nightmare fuel.
May 7, 2025 at 8:47 AM
Source: @sentdefender on X
"Footage claimed to show a Unitree H1 (Full-Size Universal Humanoid Robot) going berserk, nearly injuring two workers, after a coding error last week at a testing facility in China."
Nightmare fuel.
"Footage claimed to show a Unitree H1 (Full-Size Universal Humanoid Robot) going berserk, nearly injuring two workers, after a coding error last week at a testing facility in China."
Nightmare fuel.
Claude Code is scary good... not just for coding
My Finder kept freezing after it grew to 8GB in size. I had no idea why. Restarting did not help. Turning off Spotlight indexing didn't help.
So I turned to Claude Code... and ~10 iterations+30 minutes later it found the issue:
My Finder kept freezing after it grew to 8GB in size. I had no idea why. Restarting did not help. Turning off Spotlight indexing didn't help.
So I turned to Claude Code... and ~10 iterations+30 minutes later it found the issue:
September 27, 2025 at 8:41 PM
Claude Code is scary good... not just for coding
My Finder kept freezing after it grew to 8GB in size. I had no idea why. Restarting did not help. Turning off Spotlight indexing didn't help.
So I turned to Claude Code... and ~10 iterations+30 minutes later it found the issue:
My Finder kept freezing after it grew to 8GB in size. I had no idea why. Restarting did not help. Turning off Spotlight indexing didn't help.
So I turned to Claude Code... and ~10 iterations+30 minutes later it found the issue:
But a major factor is that our ambitions increased with ai coding. We aim for higher quality, higher correctness, larger scope. The infra team has been building systems that previously were inaccessible to a firm our size. This eats the throughput gains.
June 12, 2026 at 7:23 PM
But a major factor is that our ambitions increased with ai coding. We aim for higher quality, higher correctness, larger scope. The infra team has been building systems that previously were inaccessible to a firm our size. This eats the throughput gains.
I just realized something about the way those magic circles in Witch Atelier works. It's coding.
The size of the circle, the size of the symbols inside, even the angle of them.
It makes me think of how coding works if you had to draw it.
The size of the circle, the size of the symbols inside, even the angle of them.
It makes me think of how coding works if you had to draw it.
May 28, 2026 at 7:27 AM
I just realized something about the way those magic circles in Witch Atelier works. It's coding.
The size of the circle, the size of the symbols inside, even the angle of them.
It makes me think of how coding works if you had to draw it.
The size of the circle, the size of the symbols inside, even the angle of them.
It makes me think of how coding works if you had to draw it.
i posted an epiphany i had about why i hate AI so much on fedi and i wanna post it here but it won't fit and i don't wanna cut it down to size so fuckit i'm just posting a screenshot
September 27, 2026 at 3:24 PM
i posted an epiphany i had about why i hate AI so much on fedi and i wanna post it here but it won't fit and i don't wanna cut it down to size so fuckit i'm just posting a screenshot
futaba made the stupid body transformation app she uses and she forgot HS meant "head size" (she was too lazy when coding it)
then she got a bit indulgent with it (like how i did with these head growth drawings)
then she got a bit indulgent with it (like how i did with these head growth drawings)
June 16, 2026 at 4:47 AM
futaba made the stupid body transformation app she uses and she forgot HS meant "head size" (she was too lazy when coding it)
then she got a bit indulgent with it (like how i did with these head growth drawings)
then she got a bit indulgent with it (like how i did with these head growth drawings)
‘Newer and larger models are smarter and more reliable. But they fail unexpectedly. There is no relationship between size and how failures change over time. AI-generated code creates 70% more issues than human code.‘ tomtunguz.com/you-are-resp...
You Are Responsible for Your Agent
Amazon lost $6.3M in orders from AI coding failures. Utah law eliminates the hallucination defense. Companies bear liability for their agents' mistakes.
tomtunguz.com
March 17, 2026 at 6:33 PM
‘Newer and larger models are smarter and more reliable. But they fail unexpectedly. There is no relationship between size and how failures change over time. AI-generated code creates 70% more issues than human code.‘ tomtunguz.com/you-are-resp...
I'd love to make a size porn game but after learning to code earlier this year I think it's better if I find someone with far more experience coding than me to do that bit 😭
October 30, 2025 at 7:38 PM
I'd love to make a size porn game but after learning to code earlier this year I think it's better if I find someone with far more experience coding than me to do that bit 😭
Coding for the OG Pebble required discipline.
March 26, 2025 at 12:31 AM
Coding for the OG Pebble required discipline.
Just updated the eye tracking a bit. I did some stupid coding which shrunk the detection region which made it more difficult to keep track of my eye.
I also have more reliable Y coordinates and pupil size data so I could use that in future updates! owo
I also have more reliable Y coordinates and pupil size data so I could use that in future updates! owo
February 8, 2026 at 10:12 PM
Just updated the eye tracking a bit. I did some stupid coding which shrunk the detection region which made it more difficult to keep track of my eye.
I also have more reliable Y coordinates and pupil size data so I could use that in future updates! owo
I also have more reliable Y coordinates and pupil size data so I could use that in future updates! owo
Wow.
"Mass Reproducibility and Replicability: A New Hope"
"We uncover coding errors for about 25% of [econ/ps] studies."
"52% of re-analysis effect size estimates are smaller than the original published estimates"
iza.org/publications...
"Mass Reproducibility and Replicability: A New Hope"
"We uncover coding errors for about 25% of [econ/ps] studies."
"52% of re-analysis effect size estimates are smaller than the original published estimates"
iza.org/publications...
April 17, 2024 at 10:55 PM
Wow.
"Mass Reproducibility and Replicability: A New Hope"
"We uncover coding errors for about 25% of [econ/ps] studies."
"52% of re-analysis effect size estimates are smaller than the original published estimates"
iza.org/publications...
"Mass Reproducibility and Replicability: A New Hope"
"We uncover coding errors for about 25% of [econ/ps] studies."
"52% of re-analysis effect size estimates are smaller than the original published estimates"
iza.org/publications...
⭐ New talk! Watch / read / visually skim here: andymatuschak.org/tat
Coding agents might help us finally break out of two cages: the app model, which traps computing in one-size-fits-all silos; and programming as a specialization, which has crowded out cultures of imagination and domain insight.
Coding agents might help us finally break out of two cages: the app model, which traps computing in one-size-fits-all silos; and programming as a specialization, which has crowded out cultures of imagination and domain insight.
Apps and programming: two accidental tyrannies
On coding agents, malleable software, and the future of interface invention
andymatuschak.org
April 13, 2026 at 3:11 PM
⭐ New talk! Watch / read / visually skim here: andymatuschak.org/tat
Coding agents might help us finally break out of two cages: the app model, which traps computing in one-size-fits-all silos; and programming as a specialization, which has crowded out cultures of imagination and domain insight.
Coding agents might help us finally break out of two cages: the app model, which traps computing in one-size-fits-all silos; and programming as a specialization, which has crowded out cultures of imagination and domain insight.
The worst fact about me is that I do all my coding in size 18 font
It's just friendlier ;;-;;
It's just friendlier ;;-;;
October 13, 2024 at 6:15 AM
The worst fact about me is that I do all my coding in size 18 font
It's just friendlier ;;-;;
It's just friendlier ;;-;;
I figured I'd share this to give an idea of how big the kitsunes are at max size, at the moment they have no AI or coding and this was just their models imported thanks to @racheltheraichu.bsky.social using MCreator
August 31, 2025 at 9:09 PM
I figured I'd share this to give an idea of how big the kitsunes are at max size, at the moment they have no AI or coding and this was just their models imported thanks to @racheltheraichu.bsky.social using MCreator
vibe coding a website is pretty great if your audience is people using the same device, browser, input method, extensions, monitor, window size, font size and who are also sighted and speak your native language
March 19, 2025 at 6:59 PM
vibe coding a website is pretty great if your audience is people using the same device, browser, input method, extensions, monitor, window size, font size and who are also sighted and speak your native language
CodeCrew runs coding agents as a crew on plain GitHub, the record kept in issues and PRs.
Projects using it said the process was the same size for every change. Version 2.1 lets small work stay small, with nothing to migrate.
What changed
#AI #OpenSource #AIAgents #BuildInPublic #Programming
Projects using it said the process was the same size for every change. Version 2.1 lets small work stay small, with nothing to migrate.
What changed
#AI #OpenSource #AIAgents #BuildInPublic #Programming
CodeCrew — your coding agents, working as a crew on GitHub
Each coding agent gets its own GitHub identity; the plans, reviews and decisions stay in issues and pull requests. Version 2.1 lets small work stay small.
codecrew.works
September 26, 2026 at 11:43 AM
CodeCrew runs coding agents as a crew on plain GitHub, the record kept in issues and PRs.
Projects using it said the process was the same size for every change. Version 2.1 lets small work stay small, with nothing to migrate.
What changed
#AI #OpenSource #AIAgents #BuildInPublic #Programming
Projects using it said the process was the same size for every change. Version 2.1 lets small work stay small, with nothing to migrate.
What changed
#AI #OpenSource #AIAgents #BuildInPublic #Programming
For day 1250 of coding, I am adjusting the size and position of Spikeye's field of vision
#gamedev #gamedeveloper #Gamedevelopment #gamedesign #gamedesigner #gamedesigning #coding #dailycoding #dailycodingchallenge #dailycodinghabit #GMS2 #GMS #gamemakerstudio #gamemakerstudio2
#gamedev #gamedeveloper #Gamedevelopment #gamedesign #gamedesigner #gamedesigning #coding #dailycoding #dailycodingchallenge #dailycodinghabit #GMS2 #GMS #gamemakerstudio #gamemakerstudio2
June 9, 2026 at 4:55 AM
For day 1250 of coding, I am adjusting the size and position of Spikeye's field of vision
#gamedev #gamedeveloper #Gamedevelopment #gamedesign #gamedesigner #gamedesigning #coding #dailycoding #dailycodingchallenge #dailycodinghabit #GMS2 #GMS #gamemakerstudio #gamemakerstudio2
#gamedev #gamedeveloper #Gamedevelopment #gamedesign #gamedesigner #gamedesigning #coding #dailycoding #dailycodingchallenge #dailycodinghabit #GMS2 #GMS #gamemakerstudio #gamemakerstudio2
"You can't avoid everything with AI"
You cannot fathom the size of my backlog, my steam wishlist, my folders of unplayed SNES roms, the amount of niche weirdo indie devs I follow that insist on coding everything from scratch.
And if I somehow do run out of games? There's 3 billion Doom mods to play.
You cannot fathom the size of my backlog, my steam wishlist, my folders of unplayed SNES roms, the amount of niche weirdo indie devs I follow that insist on coding everything from scratch.
And if I somehow do run out of games? There's 3 billion Doom mods to play.
December 22, 2025 at 2:55 PM
"You can't avoid everything with AI"
You cannot fathom the size of my backlog, my steam wishlist, my folders of unplayed SNES roms, the amount of niche weirdo indie devs I follow that insist on coding everything from scratch.
And if I somehow do run out of games? There's 3 billion Doom mods to play.
You cannot fathom the size of my backlog, my steam wishlist, my folders of unplayed SNES roms, the amount of niche weirdo indie devs I follow that insist on coding everything from scratch.
And if I somehow do run out of games? There's 3 billion Doom mods to play.
How did I code the spawn size so wrong?!
#gamedev #indiedev #indiegame #gamedeveloper #gamedevelopment #unity3d #unrealengine #pixelart #gaming #indiegamedev #madewithunity #ue5 #gamersofinstagram #videogames #indiegames #gamedevlife #leveldesign #3dart #gamemaker #coding
#gamedev #indiedev #indiegame #gamedeveloper #gamedevelopment #unity3d #unrealengine #pixelart #gaming #indiegamedev #madewithunity #ue5 #gamersofinstagram #videogames #indiegames #gamedevlife #leveldesign #3dart #gamemaker #coding
May 25, 2026 at 10:45 AM
How did I code the spawn size so wrong?!
#gamedev #indiedev #indiegame #gamedeveloper #gamedevelopment #unity3d #unrealengine #pixelart #gaming #indiegamedev #madewithunity #ue5 #gamersofinstagram #videogames #indiegames #gamedevlife #leveldesign #3dart #gamemaker #coding
#gamedev #indiedev #indiegame #gamedeveloper #gamedevelopment #unity3d #unrealengine #pixelart #gaming #indiegamedev #madewithunity #ue5 #gamersofinstagram #videogames #indiegames #gamedevlife #leveldesign #3dart #gamemaker #coding
✓ You’ve got a browser.
✓ You’ve got JavaScript.
✓ You’ve got 13KB.
Join the byte-sized battle for glory, swag, and bragging rights.
This year’s #js13k theme: black cat 🐈⬛ → js13kgames.com
✓ You’ve got JavaScript.
✓ You’ve got 13KB.
Join the byte-sized battle for glory, swag, and bragging rights.
This year’s #js13k theme: black cat 🐈⬛ → js13kgames.com
August 14, 2025 at 12:05 AM
✓ You’ve got a browser.
✓ You’ve got JavaScript.
✓ You’ve got 13KB.
Join the byte-sized battle for glory, swag, and bragging rights.
This year’s #js13k theme: black cat 🐈⬛ → js13kgames.com
✓ You’ve got JavaScript.
✓ You’ve got 13KB.
Join the byte-sized battle for glory, swag, and bragging rights.
This year’s #js13k theme: black cat 🐈⬛ → js13kgames.com
Berkeley and Arena ran 7 models through Claude Code, Codex CLI and Pi. Success rates moved a couple of points. Cost moved up to 5x. Same model, same tasks. If you pay per token, the harness is a line item. Read the sample-size limits before you switch.
harnesstax.github.io
harnesstax.github.io
HarnessTax: How Much Does the Harness Matter for Coding Agents?
What does a coding-agent harness actually add, and at what cost? It turns out your Claude models may not need Claude Code… We evaluate 21 model–harness pairs spanning seven models and three…
harnesstax.github.io
September 24, 2026 at 9:00 PM
Berkeley and Arena ran 7 models through Claude Code, Codex CLI and Pi. Success rates moved a couple of points. Cost moved up to 5x. Same model, same tasks. If you pay per token, the harness is a line item. Read the sample-size limits before you switch.
harnesstax.github.io
harnesstax.github.io