@stack72

Infra Coder with a militant interest in DevOps and Systems Thinking. Speaks with a funny accent. **Evangelist of Life**. DAF

 Derry / Londonderry
stack72.dev
Joined August 2009
  • Tweets11K
  • Following501
  • Followers5.3K
1.3KMedia
Hey @AnthropicAI we still waiting on you to fix that broken GitHub Action build so people can use Opus 5.5 in CI github.com/anthropics/claude…

Release with 2.1.280 failed in CI · Issue #1849 · anthropics/claude-code-action

https://github.com/anthropics/claude-code-action/actions/runs/35755580506 This CI run on main failed, which prevents release of the tag that includes 2.1.280, which prevents us from switching to cl...

github.com
1
2
267
IT JUST GOT INTERESTING!!! As a person who didn't find a model I liked past Opus 4.6 (except GPT Terra...) Now I have my choice :D
2
2
321
As if by magic, the normal message of "this new model is great and it has been my daily driver for the last few weeks" - that means the day they shipped Fable 5.2 they changed away from using that model and used this instead.... curious
1
1
329
Been trying to upload a GitHub Release for about an hour now with no error but it's doing nothing - @githubstatus says there's no issue but their API says otherwise...
1
328
Had a really fun time talking at @kc_dc today - had so many fantastic questions and enjoyed hearing people’s views on building with AI I will post the slides on this page this evening: swamp-club.com/kcdc

The humans architect the system, the AI writes the code.

Paul Stack at KCDC 2026 — the humans architect the system, the AI writes the code. What it looks like to run an engineering team where every line is generated by an LLM.

swamp-club.com
1
5
204
I think I am being dumb - but there's really no built in equivalent of `claude -w` for @opencode ?
1
204
So the majority of coding agents have standardised on AGENT.md and the same skills style - but the way to configure your own settings in them differs wildly :/
1
1
276
Went to sign up for a Google AI Account and it's so fucking confusing as to what I am paying for. Had I a personal google account and not made my email a workspace account, it would be way more straightforward
1
1
218
Paul Stack retweeted
Guillermo Rauch
@rauchg
Sep 5
Anyone that works in AI is working the hardest they've ever worked in their lives. On the surface it's somewhat ironic (AI should give us back time!), but the reality is that it's the most fun, fascinating and empowering epoch in human history. The intelligence revolution.
349
605
179
6,721
678,556
Paul Stack retweeted
dax
@thdxr
Sep 2
however you feel about the reading code debate as an exercise you should try talking to the LLM and see if you can get to confidence about the code without directly reading the code
102
35
9
1,144
70,398
I wish I could quantify it and present the evidence but I am pretty disappointed by the trend in the frontier models. They peaked for me at Opus 4.6 - each model after introduces token bloat, incredible cost increases, more verbose output and less good outcomes for me 1/
5
1
9
1,065
more replies
I have tried gpt models and they feel incredible slow - people tell me they are blazing fast... my the "feelz" I get doesn't suggest that. As a person who is continually trying to increase my productivity and the efficiency of my software flow, I am definitely concerned. 3/
1
1
186
I saw some positivity with Google Antigravity (AGY) yesterday and Gemini 3.1-pro but meh I know the people training and creating these models are incredibly smart but, as a user, I feel there's something missing in them right now. Token efficiency is only part of that. YMMV 4/4
124
Added support to Swamp today for @AmpCode @pidotdev and Google AntiGravity. The list of native agent integrations is growing :)
1
9
698
I need to do a comparison of Opus 4.6 price vs Fable 5.1 price - I have not moved because the token bloat of Opus 4.7, 4.7 and 5 were bad for me and not giving good results I did a bunch of work with Fable 5 earlier this week and almost blew my weekly limit...
2
4
626
TL;DR: If the workload can be handled well by Opus 4.6, it has a substantial cost advantage. Fable 5.1 needs to produce meaningfully better results, require fewer retries/agent steps, or make heavy use of cached context to justify its 100% premium on input and output tokens.
1
125
New model comes out: "This is the best model yet, I use it for everything" Of course you have, you have unlimited credits! I would tell Claude to do every single thing just because I could if I had unlimited tokens!
4
5
674
All I want to christmas is for @claudeai to follow an AGENTS.md convention to allow my repos to work for all tooling!
1
5
269
Completely related to my just posted blog post is a conference talk titled: Trust Chains for Agent-Verified Code Coming soon to a conference near you :)
5
213
Load more