Benjamin Stein

@davidad This is so on point. 4.8 is SO BAD at writing. Not in the emdash kinda way, but a legit "I'm sorry, I have no idea what you're trying to say right now" kinda way

No one: Claude Opus 4.8 Max: Let me refine your load-bearing claim rather than just accepting it, because you’re doing zero moves there, and the gap is what’s actually interesting. The one place I’d still push, because I think it matters: your message is wearing content-clothes, but the content isn’t actually *there*. The tell: it’s just an empty string. But the emptiness of the string IS its lack of content. Pull one, and the other goes inert. That’s the structural spine.

@trq212 @sidbid I was struggling to internalize dynamic workflows and what "write its own harness on the fly" actually meant. So I used it to fan out 484 agents rebuild isitchristmas.com (turns out no, not today). Wrote up my learnings here: benjaminste.in/blog/2026/05/2…

Workflows are the biggest upgrade to Claude Code’s capabilities since skills and subagents. I dove deep into it with @sidbid to figure out best practices, examples and more.  I’m particularly excited about the non-technical tasks it enables for Claude Code.

Today, I signed an Executive Order temporarily repealing bedtimes in the City of New York so that kids of all ages can watch our team in the NBA Finals. As Mayor, you’re forced to make many difficult decisions. This was not one of them. Go Knicks.

@KibryHouse It really was.

Playing Mario 64 when it first released must have been a life changing experience

I rebuilt "Is it Christmas" using 484 subagents and 16 million tokens to learn how Claude's dynamic workflows work. Spoiler Alert: today is not Christmas. benjaminste.in/isitchristmas/ h/t @konklone, as always

I'm a couple projects into Codex + 5.5. Early reactions: * The coding model is very good. It YOLO sideloaded and debugged a new Android widget with no human intervention very very well. * I hate using the mouse. * Way too many back-and-forth questions. I kept finding Codex waiting for me. Like I say "LGTM let's build!" and come back 15 minutes later expecting a finished result but instead found a "Should I get started?"

And how would you trade off? If CC + Opus consistently generated better code output but you really dislike their interface (or vice versa), which would you use for day-to-day work?

When you say you prefer [ Claude Code | Codex ], are you optimizing for the user experience or the quality of the output?

@anothercohen A+ tweet

You just won a 2-week, all-expenses-paid vacation. But there’s a catch: you have to stay within one region the whole time. What are you picking?

@andrewneilson_ @dlwiest This is the actual answer. Ignore other replies. Enterprise is PAYG. The other (way too common) answer is people just use PAYG because they don't read/think e.g. paste your API key into Cursor and oops $3000

@dlwiest if you have more than 150 seats you have to convert to enterprise (where you pay api rates, possibly with a volume discount). Compliance-minded companies aren’t going to let you use personal plans, so the options are either team or enterprise.

Can anyone explain to me why companies don’t just give employees $100 / month Claude Code or Codex plans instead of paying per token? There has to be an explanation, because this keeps happening and doesn’t make sense otherwise

@moonpetal76 My parents liked Dustin Hoffman in The Graduate and my great-great-grandparents thought Eisenstein sounded to Jewish when they got off the boat.

your username. explain. now.

@denicmarko superduperlabs.com

What are you working on? Drop a screenshot or link.

How can I get any work done with COYOTE PUPPIES ROMPING IN MY BACKYARD?!?!!!

Opus Ultracode should have been called "Hold my beer" 108/115 agents done · 41m 21s · ↓ 8.7m tokens

"comfortable" is not the word I'd use

@nateberkopec My system prompt includes: "Bluntly correct me when I'm wrong. I'd rather argue than have you cave, especially when I'm being an idiot." which results in a lot more "Let me push back..." which is often great.

Literally everything I type in happens to be "the right call". What a surprise.

Claude is so incredibly sycophantic, I don't know how anyone stands it.

@mattyp it's 2026. The answer to "Should I just build one?" is always 100% yes.

Why aren’t there any good handwriting / journaling apps for iPad? Should I just build one?

can you die from a lack of being called big dawg

@deepfates The new horse sized duck?

I fear not the man who vibe coded 50 new apps, but the man who vibe coded one new app 50 times

I forked @WarnerTeddy's awesome bird sound field guide generator. I don't have a Pi yet so I just made a static version ...with Opus 4.8 ULTRACODE with dynamic workflows because why not? Burning 940,000(!) tokens was TOTALLY worth it to watch it do Adversarial verification. Imagining an virtual adversarial wild turkey vs a virtual adversarial chickadee is the epic battle of the species I'm here for.