Charles Cheung

AI Founder, full-stack. I build the AI/MarTech systems behind $10M+ GMV in DTC/cross-border e-commerce. Shenzhen / Hong Kong / Tokyo.

[Charles Cheung @Kee_Kun]

If you want to grow a business in China, Xiaohongshu (RED) is the channel that matters most. And the only real way to grow an account there is boring: make genuinely good content. AI helps a lot — but people are still the decisive factor.

[Charles Cheung @Kee_Kun]

Agent memory isn't storage. It's a recommendation system — pulling a slice of your history back into the context window each call, same pipeline as any feed: retrieval, ranking, filtering. Except there's no reward signal. No clicks, no dwell time, nothing tells you when it retrieved the wrong thing.

[Charles Cheung @Kee_Kun]

@aarai666 这是跟BYD的合资公司BTET在深圳坪山开发的，除了牛头标基本都是BYD。

[Charles Cheung @Kee_Kun]

Nice result, but not cheap — and hard to say the PR is really what moved the needle on AEO.

[Dan ⚡️ @d4m1n]

wtf press releases seem to work for AEO, but also SEO?!

looks like a pay to win atm, not sure for how long this is gonna work.

I bought a press release package and literally waited for 2 months.

now getting decent AI mentions (it was nothing before) and a nice bump in impressions.

I mean it's not a trillion mentions but it's a super nice kickstart for a new domain IMO

[Charles Cheung @Kee_Kun]

Vercel just shipped a course teaching you to build a coding agent harness from scratch — tool loop, sandbox isolation, context pruning, safety gates. It's the right skeleton, and genuinely useful for understanding how these systems work.

But if you're actually shipping this for real work, don't rebuild it from zero. Teams like Pi have already burned real production traffic on the parts a course can't teach — crash recovery, durable tool execution, concurrent sessions. Build on top of that layer instead of re-deriving it the hard way.

https://t.co/ivXjyxUQWL

[Charles Cheung @Kee_Kun]

$10k into real SEO compounds for months. This buys a few hours at #1.

[Jonathan Wilke @jonathan_wilke]

this is unreal

[Charles Cheung @Kee_Kun]

OpenAI's Codex as a platform post gets the direction right: stop bolting chat onto everything, put the agent loop inside the product you already have. Harness tuning alone took Sol from 13.3% to 38.3% on ARC-AGI-3 with 6x fewer tokens. Worth doing the same for your own product.
https://t.co/FBDtRwtUPK

[Charles Cheung @Kee_Kun]

Opus 5 seems super solid at Three.js.

[Dilum Sanjaya @DilumSanjaya]

Prototyped a cozy Wild West railroad tycoon / logistics game
This is the result after ~2 hours with Opus 5

[Charles Cheung @Kee_Kun]

Fantastic idea!

[Meng To @MengTo]

I’m blown away by the quality of the landing pages that Opus 5 can create in Three.js.

Prompting it with "orbit using mouse" is enough to create a 3D parallax effect. The same goes for "add particles to pointer." It’s like creating a living, breathing world.

More prompts I used:

- Add a butterfly that flies to the branch and flies away when I mouse over it.
- During the intro, first load a wireframe of the branches, similar to the field-scanning effect in Death Stranding.
- Add a scan effect to the card images while they load.

Then I asked it to use the shader buttons and nav bar I made in previous projects.

As explained in my last video, I started with an inspiration and the prompt, “Recreate this in Three.js in a single HTML file. Self-verify until perfect.” Opus 5 ran for two hours. All the code is less than 1 MB.

Inspiration:
https://t.co/BGMHijEF8B

Let me know if I should open-source this.

[Charles Cheung @Kee_Kun]

Start small, iterate fast.

[Deedy @deedydas]

Crazy to think that this was the first version of OpenRouter.

[Charles Cheung @Kee_Kun]

Thought base model parity meant role distinctions were obsolete.

Turns out both economics and multi-agent systems dictate otherwise: roles aren't about capability ceilings, they’re about coordination architecture and breaking low-variance failure modes.

Specialization > brute-force homogeneity.

https://t.co/DJyKetUBor

[Charles Cheung @Kee_Kun]

Always letting self-driving cars pass when driving in China. Gotta secure my spot on Skynet’s mercy list.

[Marc Lou @marclou]

I want to be sure they remember me when AGI takes over

[Charles Cheung @Kee_Kun]

As someone who used to daily-drive Gentoo and Arch back in the day, I never expected to see a brand new, highly-opinionated end-user Linux distro pop up in 2026 like Omarchy.

Wrapping Arch + Hyprland into an out-of-the-box experience? Young me would have either cried of joy or called it heresy lol.

[DHH @dhh]

Omarchy Quattro is out!! This is one of the greatest software releases in my professional career. I hope you enjoy using it just as much as I did building it ✌️ https://t.co/81wLStgH7P

[Charles Cheung @Kee_Kun]

Just scrolling and realizing a girl I went on a single date with has already speedrun an entire controversial marriage and messy divorce. @ManusAI

[Charles Cheung @Kee_Kun]

Agree

[Elon Musk @elonmusk]

@ARKInvest China is awesome. I strongly encourage people to visit.

[Charles Cheung @Kee_Kun]

Too many models

[WeChat @Weixin_WeChat]

Meet WeLM — a family of large language models built by the Weixin team with resource efficiency at its core.

[Charles Cheung @Kee_Kun]

Doubt any AI proactively reaches for this method yet.

[IP @ipwanciu]

That's a small reminder! 👇

As of June 2026, HTTP has a new method - a GET with a `body` 😎

Meet QUERY (RFC 10008).

It lets you send complex queries in the body (like POST) while staying safe, idempotent, and cacheable (like GET). Already works on servers (Node.js, Go, Laravel…).

Browsers are still adding full support, usable now with care, with widespread adoption coming later in 2026+.

[Charles Cheung @Kee_Kun]

Codex agent got locked out needing a YubiKey touch — so it made the speaker loop-shout "please touch the yubikey" until a human showed up.

[paperpaper @paperpaper886]

codex总在一些奇怪的地方展示AGI

[Charles Cheung @Kee_Kun]

Want BYOM on Cloudflare OS instead of the built-in model catalog? Skip the wizard deploy. Use the https://t.co/bRFiy7Tc3g repo, keep AI Gateway mode off, and just plug in your own Kimi/GLM/whatever API key. The wizard-deployed version won't let you do this at all.

[Cloudflare Developers @CloudflareDev]

In four months, our internal AI agents flagged nearly 250,000 potential problems and blocked 16,000 bad merges.

Here's how we gave every Cloudflare employee AI they can actually use, including what worked and where we fumbled.

https://t.co/u3AlNtuml1

[Charles Cheung @Kee_Kun]

Two things Grok Bot gets right:
1. A persistent, stateful, networked PC out of the box — most alternatives (https://t.co/hJ1LI0e92v, Cloudflare Workers) are ephemeral sandboxes that reset every run.
2. A wide set of built-in connectors — though you can get similar coverage with @composio too.