Evan Fleischman

founder, golfer-turned-dad, sandwich connoisseur

[Evan Fleischman @EMFleisch]

People forget the OG of the hype train - Long Blockchain Iced Tea.

Best to ever do it might be getting outclassed by a sneaker

https://t.co/0jEaq6KoUr

[Evan Fleischman @EMFleisch]

Dario's LinkedIn touting that he wrote 5k lines of Matlab and Fortran back in the day.

Possibly even more impressive now than it was then

[Evan Fleischman @EMFleisch]

Still using Opus 4.5 as the daily coding driver

Vibes point to the "Sonnet 5 -> Opus 4.6" rebranding being true, so 4.5 is actually SOTA

[Evan Fleischman @EMFleisch]

Who's written the claude code skill for updating claude code when a new version is released every 20 minutes

[Evan Fleischman @EMFleisch]

@zeeg How big was https://t.co/a17vrq9D6m before the move to skills?

[dax @thdxr]

if you don't stay in the loop when coding with AI you produce a lot more tech debt

the immediate response to that is "we can just make AI fix it" and everyone's been rolling with that

but stop and think about it - this is an absurdly high bar to clear

[Evan Fleischman @EMFleisch]

😳

[Bun @bunjavascript]

Bun is joining Anthropic!

https://t.co/06NlXDZhNE

[Evan Fleischman @EMFleisch]

Sonnet to plan. Composer to execute

[Evan Fleischman @EMFleisch]

hold up... haiku-4.5 kind of rips...

(and is _so_ fast . doubles down the unusability of gpt-5 atm)

[Evan Fleischman @EMFleisch]

How does anyone actually use GPT5 (reg or codex) for code?
Its so slow, which promotes terrible dev behavior (ask a question, leave to do something else for 10 minutes while 'blocked' on this task, come back to answer)

[Evan Fleischman @EMFleisch]

This is now getting insane - i downgraded my Claude account YESTERDAY and they go and drop 4.5

Now 3 for 3 with my account cancelation leading to a next-day major anthropic product release

[Claude @claudeai]

Introducing Claude Sonnet 4.5—the best coding model in the world.

It's the strongest model for building complex agents. It's the best model at using computers. And it shows substantial gains on tests of reasoning and math.

[Evan Fleischman @EMFleisch]

I have my new litmus test for AI capabilites -

Upgrade a single usage of Google Places Service Place Autocomplete to Google Places (new) Autocomplete.

Will take AGI to complete

[Evan Fleischman @EMFleisch]

@lennysan is on a run for the ages right now

[Evan Fleischman @EMFleisch]

If this is an attack, targeting Rocket league is taking it TOO FAR

[Evan Fleischman @EMFleisch]

something is going on - I got multiple Microsoft 2FA codes this morning, X had outage, OpenAI and Google having service disruptions...

[Evan Fleischman @EMFleisch]

ngl Gemini 2.5 Pro cooks

[Evan Fleischman @EMFleisch]

@zeeg they're really good at making a shadcn marketing website.

and once you know what that looks like, you see it _everywhere_

[Evan Fleischman @EMFleisch]

Happened again! just as im getting frustrated that Claude can't do web search it drops.

They refuse to let go of my $20

[Evan Fleischman @EMFleisch]

If youre only using Claude 3.7 in Cursor, highly suggest doing some dev directly with claude dot com.

Back to using the UI as one of my main workflows. 3.7 thinking has been stellar (reverted back to 3.5 in cursor)

[Evan Fleischman @EMFleisch]

Claude - coding
ChatGPT - writing
Gemini - ecosystem (and free tier capabilities)
DeepSeek - chaos