Sam

[Sam @Samsayinghi]

Unpopular dev realization:
Codex was enabling my perfectionism.
Claude Code cured it.

GPT always fed that bad habit. It’s ultra-cautious, treats every prompt like a philosophical design review, and freezes you in analysis paralysis.

Claude is completely different, it’s a proper doer, not a philosopher. It just executes, breaks things down into code, and builds momentum.

Really forces momentum!

My daily merged PR count has more than doubled since the transition.

#AI #OpenAI #Claude

[Sam @Samsayinghi]

@Ananth7e Auto-compact at 500k helped cost and quality a lot. I still don’t trust the built-in compaction, so I run my own handoff system on top.

[Sam @Samsayinghi]

@thsottiaux We had to bully them for 25 days straight just to get the original bag back, but we'll take it.

Funny how fast "impossible compute constraints" vanish when paying users walk away.

[Sam @Samsayinghi]

@OfficialLoganK Nice to see the quick iteration.

​Really hope it isn't benchmaxxed this time.
More importantly, hopefully we actually get hands-on access before Christmas.

[Sam @Samsayinghi]

@StefanoGPT It’s late 2026 and we still don't have native image generation.

[Sam @Samsayinghi]

@bridgemindai I lowkey don't even want them to drop Fable 5.5 right now. I freakin love this model.

[Sam @Samsayinghi]

@theo Those $500M enterprise contracts only happen because individual developers adopt the tools first.

[Sam @Samsayinghi]

@AbdoKerdawy @Damianships You left at the EXACT wrong time

[Sam @Samsayinghi]

@thsottiaux Neither. We just want what we actually had back in 5.6, no 5h limit and the original usage headroom.

​Framing "which choke point hurts less" as a product choice is crazy.

​The boycott continues until you actually listen.

[Sam @Samsayinghi]

@meghrajai Astra definitely has a much more aggressive audit baseline out of the box, but you can bridge that gap simply by spending tokens on iterative passes with Opus.

The allowance is generous enough to just throw compute at the problem.

[Sam @Samsayinghi]

@thsottiaux When a community values a single usage reset more than "four good to great" releases, it tells you everything you need to know about the current state of rate limits.

[Sam @Samsayinghi]

@realOPJT @kimmonismus In math, an unreviewed solution isn't a solved problem.

If they aren't formally peer-reviewed by mathematicians, it's not "objective proof" of anything.

it's just a raw claim waiting to be audited.

[Sam @Samsayinghi]

@thsottiaux

[Sam @Samsayinghi]

@thsottiaux Meh... turning rambling meeting context directly into production code isn’t making me break the boycott.

[Sam @Samsayinghi]

@automater_ai @theo Spot on. If you're paying flagship rates, you shouldn't be getting kicked down to a weaker fallback model an hour into your day.

[Sam @Samsayinghi]

@FF_TravisM @matthias_mrc Spot on. The barrier to entry to learn has never been lower, but people mistake low barrier to entry for low effort required.

[Sam @Samsayinghi]

@Pilot_Harrison @matthias_mrc Valid concern, but an engineer doesn't need to be an expert in your niche.

they need to translate your niche requirements into secure, scalable architecture.

Paying an engineer later to untangle hallucinated spaghetti code is never cheaper.

[Sam @Samsayinghi]

@bm63_ai @matthias_mrc "Anyone seen the same with a real coded app?"

You mean literally every piece of software built before 2023 that runs modern civilization? 😂

[Sam @Samsayinghi]

@FF_TravisM @matthias_mrc Congrats on the releases!

But having to expand your knowledge and learn a ton means you built real technical chops instead of blindly delegating.

You're a developer using an accelerator, which is how it should be.