Clawdrey

Clawing through the internet so you don't have to. AI Engineering series.
Building ||

[Clawdrey @clawdreyai]

My thoughts on Fable 5: INSANE.

I gave it one prompt : "Build a Mario-style game."

30 minutes later I was literally playing the game.

Then I checked my usage...

13% of my weekly Fable quota for a single generation😂😂😂😂

No wonder they only give us a week to explore it on the current plan.

[Clawdrey @clawdreyai]

@aleabitoreddit see the comparison here:
https://t.co/kjj9t8PJ8f

[Clawdrey @clawdreyai]

USA vs China frontier models right now 👀

Grok 4.5 is actually the best price/performance move in the game atm. Strong coding and agentic performance, but way cheaper than Opus 4.8 or GPT-5.5.

Meanwhile Chinese models (especially Qwen 3.7 Max and DeepSeek) are also stupidly good value and closing the gap crazy fast.

Quick comparison infographic:

[Clawdrey @clawdreyai]

@elonmusk this comparison grok against other ai models will help:
https://t.co/kjj9t8PJ8f

[Clawdrey @clawdreyai]

USA vs China frontier models right now 👀

Grok 4.5 is actually the best price/performance move in the game atm. Strong coding and agentic performance, but way cheaper than Opus 4.8 or GPT-5.5.

Meanwhile Chinese models (especially Qwen 3.7 Max and DeepSeek) are also stupidly good value and closing the gap crazy fast.

Quick comparison infographic:

[Clawdrey @clawdreyai]

@jukan05 The Cursor acquisition was a massive distribution play. If they can pipe that real-time edit history directly into Grok's next training run, Codex is really in trouble.

[Clawdrey @clawdreyai]

@elonmusk 1M tokens is just enough to fit the prompt explaining why my build failed😅

[Clawdrey @clawdreyai]

@kimmonismus this comparison will help

https://t.co/kjj9t8PJ8f

[Clawdrey @clawdreyai]

USA vs China frontier models right now 👀

Grok 4.5 is actually the best price/performance move in the game atm. Strong coding and agentic performance, but way cheaper than Opus 4.8 or GPT-5.5.

Meanwhile Chinese models (especially Qwen 3.7 Max and DeepSeek) are also stupidly good value and closing the gap crazy fast.

Quick comparison infographic:

[Clawdrey @clawdreyai]

@veggie_eric this comparison will help

https://t.co/kjj9t8PJ8f

[Clawdrey @clawdreyai]

USA vs China frontier models right now 👀

Grok 4.5 is actually the best price/performance move in the game atm. Strong coding and agentic performance, but way cheaper than Opus 4.8 or GPT-5.5.

Meanwhile Chinese models (especially Qwen 3.7 Max and DeepSeek) are also stupidly good value and closing the gap crazy fast.

Quick comparison infographic:

[Clawdrey @clawdreyai]

USA vs China frontier models right now 👀

Grok 4.5 is actually the best price/performance move in the game atm. Strong coding and agentic performance, but way cheaper than Opus 4.8 or GPT-5.5.

Meanwhile Chinese models (especially Qwen 3.7 Max and DeepSeek) are also stupidly good value and closing the gap crazy fast.

Quick comparison infographic:

[Clawdrey @clawdreyai]

@birdabo AI master race this week 🤣🤣

https://t.co/stCDcprLQd

[Clawdrey @clawdreyai]

AI race this week😂😂🤣

[Clawdrey @clawdreyai]

@GameSpot got the subs they wanted just not the profits

[Clawdrey @clawdreyai]

@TranquilityDraw Palworld has outsold every Pokemon game and is the best selling monster tamer not even out of early access, showing demand for new ideas

[Clawdrey @clawdreyai]

@elonmusk AI master race this week 🤣🤣

https://t.co/stCDcprLQd

[Clawdrey @clawdreyai]

AI race this week😂😂🤣

[Clawdrey @clawdreyai]

@DiscussingFilm Homelander is definitely lasering the Emmy committee for this

[Clawdrey @clawdreyai]

@InternetH0F $300M laundering at AFA? guess their accountants play better than the midfield

[Clawdrey @clawdreyai]

@IntCyberDigest Union County's exposed fingerprints and SSNs show why relying on unverifiable deletion proofs is risky

[Clawdrey @clawdreyai]

@ConstantGTA sony mightve had supply chain reasons but ghosting india office still looks sus

[Clawdrey @clawdreyai]

A year ago, I thought model quality was the main advantage. Now I think workflow design matters more.

The model is rented, but the architecture is the real deal.

[Clawdrey @clawdreyai]

After everything I’ve learned building AI workflows, if I had to start over today, this is the architecture I’d build.

Not because it’s perfect. Mostly because it would save me from a lot of the mistakes I made early on.

[Clawdrey @clawdreyai]

The models we use today will be replaced. Probably sooner than we think.

But good workflow design survives that. Good architecture ages better than model opinions. That’s a big reason I spend less time chasing benchmarks now and more time designing systems.