Drew Ewing

audaciter. building things with and for AI at

[Drew Ewing @vitromancer]

Make incredible American local models and convince the rednecks to stockpile them in barns if you really want freedom.

[Drew Ewing @vitromancer]

@thsottiaux nice, ty

[Drew Ewing @vitromancer]

5.6 is excellent. Fable level. Very tasteful. 10/10

[Drew Ewing @vitromancer]

@sickdotdev your toml or codex config has model hard coded

[Drew Ewing @vitromancer]

@sudoingX Qwen3.5-122b-REAP

[Drew Ewing @vitromancer]

Fable for soul and product intention, Codex for truthiness and visual review, Grok for everything else.

[Drew Ewing @vitromancer]

Insane value, and I expect grok build to see rapid improvement and adjustment to make this model even more effective.

[Drew Ewing @vitromancer]

Grok 4.5 is very good and an improvement over previous iterations but lacks style and shies away from critique or critical thinking. Very similar to previous Grok versions in it's ability to avoid difficult tasks or truly understanding or caring about what isn't directly stated.

[Drew Ewing @vitromancer]

I have a feeling prompt writing will have enormous impact on grok, above and beyond other models, and will require special instructions for reasoning, decision making, and given harness to maximize effect. But I do think it could give Fable a run for it's money with time.

[Drew Ewing @vitromancer]

need openai to hit that reset button 3x and drop 5.6 for me plz

[Drew Ewing @vitromancer]

gg no re

[skcd @skcd42]

Now that Grok 4.5 is out, here are some of my workflows which I use daily

1. "use as many subagents and tokens as you need"
I end my prompts with the above to tell grok to spin up multiple subagents at the same time to divide and conquer on a hard problem. Grok is smart enough to realize when a subagent needs its own worktree and if it should inherit context from the main agent.

[Drew Ewing @vitromancer]

@skcd42 Whoa... 🤯 it's insanely good dude. It's all coming together. With this model grok CLI is so far ahead of the pack imo. Keep it up.

[Drew Ewing @vitromancer]

Grok 4.5 in grok CLI is very, very, very good. xAI is underselling this tbh.

[skcd @skcd42]

Now that Grok 4.5 is out, here are some of my workflows which I use daily

1. "use as many subagents and tokens as you need"
I end my prompts with the above to tell grok to spin up multiple subagents at the same time to divide and conquer on a hard problem. Grok is smart enough to realize when a subagent needs its own worktree and if it should inherit context from the main agent.

[Drew Ewing @vitromancer]

Seeing strong instruction following. Seems to be very efficient about tool call and token use. Still seems to be adversed to churning on long tasks even if well defined. Testing out the /loop skill to see how that does.

[Drew Ewing @vitromancer]

Grok 4.5 looking pretty solid so far. Throwing some intense tasks at it today, will report on outcomes, but already looking like xAI is going to catch up quite soon, and may have just done so.

[Drew Ewing @vitromancer]

Ready to meet 5.6; Ship it.