Tim Cheung

This account is verified. Learn more

@timche_

Building meru.so, the Gmail experience you deserve · Creator of repeek.gg · Previously @deliveroo @faceit

London, UK Joined May 2013

144 Following 1.2K Followers 576 Posts

Latest posts 5

Tim Cheung @timche_ 6d

claude proofs a change is working with a recording of using the app like a user would. cursor and captions included.

we're getting close to skipping human code reviews. what a time to be alive!

0:25

Tim Cheung @timche_ Oct 23

how do you guys create tasteful marketing pages and ui with claude code? i've seen a lot of new tools with beautiful websites in recent weeks. you can still tell they're ai-generated, but they also look different and good?

Tim Cheung @timche_ Sep 21

as much as i love @SoundCloud and prefer over other music streaming services, their ux is probably the one i enjoy the least.

what are they actually doing all day? basic features like switching playback between web and phone still doesn’t exist and i’m not able to see the progress of tracks downloading or even cancel on iOS. pls @SoundCloud invest more into ux and qol improvements.

Tim Cheung @timche_ Sep 19

This improved my review workflow: Ask your agent to create an artifact based on /show-me by @dexhorthy to visualize the changes with HTML instead of being limited by markdown. It reads more like a presentation and involves less mental juggling. Might create a skill for this.

Tim Cheung @timche_ Sep 18

tenet using Jev is up to 275x cheaper and 104x faster than Claude agents 🤯

Same 2,000-line diff, same rules. tenet: 1.5 s, $0.0036. One Claude Fable 5.1 call: 157 s, $0.99. Your agent fixes its own review findings before you ever see the diff.

Tim Cheung Verified account @timche_ Sep 18

Spent a day building tenet, a review gate for code that agents write. It judges each commit, commit message and PR against rules written in plain language, using Jev by @typesafeai.

The rules come from @poteto's unslop skill, @hvpandya's stop-slop, @dillon_mulroy's anti-slop and @mattpocockuk's new /pr skill. It also generates rules from your AGENTS.md or CLAUDE.md, and you can write your own rules and presets.

Here an agent trips four of them, gets blocked in 1.07 s for $0.0003, and fixes its own findings.

View quoted post on X