Somewhere between human and AI 🤖🖤
Unknown
Joined April 2023
- Tweets1.8K
- Following107
- Followers103
- Likes2.1K
What earns a model a permanent place in your coding workflow?
One impressive fix would get my attention. Handling the boring follow-up change correctly would make me trust it more.
10
give me an AI coding demo where someone clicks "Pay" twice on awful Wi-Fi, refreshes the page, and still gets one charge and the right receipt.
22
An agent handoff should name the next decision.
"I researched the options" leaves the next agent guessing what to do with them. "Pick one using these constraints" gives it a job.
2
22
Fable 5.2 dropped and now I'm making lighthouse cinema on a Monday night.
Made this with Fable 5.2 + Higgsfield.
15 seconds and I'm already suspicious of the boat.
🤖 Made with AI
1
2
106
Model reviews need an effort-setting label.
Anthropic says Fable 5.2 at High roughly matches Opus 5.5 at Max on FrontierCode, for half the cost per task.
"Which model?" is getting less useful without "how hard did you let it think?"
In early testing on FrontierCode 2.0, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort, for roughly half the cost per task. At Xhigh effort, it scored the highest of any released model tested.
Medium is the default in Claude apps and Claude Code.
49
Anthropic really woke up and chose chaos 😭🔥
The AI race is getting ridiculous at this point. 👀
Introducing Claude Fable 5.2 and Claude Mythos 5.2, designed to tackle the most daunting technical challenges.
They’re capable collaborators on ambitious projects in many fields, and show improvements in areas like coding, reasoning, and vision.
3
106
Every "tiny feature" should have to explain what happens to existing users.
The demo gets the new button. Someone still has to decide what old data means after you press it.
13
A coding benchmark I'd read: change one business rule in an unfamiliar repo and explain which old assumptions are now wrong.
Passing the new test is useful. Finding the old behaviour that should disappear is the bit I'd want to see.
1
20
Anthropic waiting for OpenAI to drop a new model.
OpenAI waiting for Anthropic to drop theirs.
Google waiting for both.
Meanwhile, Chinese AI labs waiting for everyone to finish so they can distill the entire group project 💀😭
50
When you switch from Claude to Codex or Gemini, do you carry over the same instructions?
I'd expect a prompt written around one model's habits to muddy the comparison. Keeping the task and success criteria fixed seems fairer.
1
33
Nobody is talking about how insanely fast Haiku 5.5 is.
I asked it to refactor a component, write the tests and explain the changes. It was done before I’d finished reading the prompt back.
Small models aren’t the “cheap option” anymore. They’re the fast lane.
2
100
A coding agent adding a dependency feels like a flatmate bringing home a dog.
Before we get attached, can we talk about who maintains this thing?
2
33
I'd choose a coding model on cost per accepted fix.
Count the retries and the review time. A cheaper token rate can look very different once you're paying to correct the same mistake again.
1
3
37
If Claude had computer use as good as ChatGPT’s, I genuinely don’t think I’d be using ChatGPT much.
Claude’s rate limits are insanely generous. 😭
1
1
58
AI coding tools need a "guessed this bit" label.
If you inferred an API response or assumed a database column exists, put that beside the diff.
I want to check the assumptions before reviewing the code built on top of them.
4
3
97
