Ju Lin's AI Weblog

Where MCP Is Headed Next

Thu, 27 Aug 2026 00:00:00 +1200

The MCP maintainers published an updated roadmap.

Since March, MCP went stateless (no more session handshakes), turned Tasks into an official extension for long-running work, and shipped enterprise auth pieces like issuer validation and Client ID Metadata Documents.

Next up: one HTTP-native transport instead of separate stdio and HTTP paths, real agent identity (workload federation, token exchange), and progressive tool discovery for large catalogs. No firm dates yet.

MCP is leaving its single-session, request/response origins behind for something built for long-running, multi-agent systems.

Agents Should Be Durable, Not Long-Lived

Wed, 26 Aug 2026 00:00:00 +1200

A common way to build an AI agent is to treat it as a long-running process. A worker receives a request, enters an agent loop, calls models and tools, waits for results, and eventually returns an answer.

This works well until agents start doing real work.

An agent may spend twenty minutes researching a problem, wait ten minutes for a build, ask a user for approval, or come back hours later when an external job finishes. Keeping a worker alive for the whole run wastes resources and makes failures expensive. A deployment, crash, or machine restart can also destroy work that has already happened.

What Is an Agent Harness?

Wed, 26 Aug 2026 00:00:00 +1200

An oversimplified equation: agent harness = ai agent − model.

Overview

An agent harness is everything in an AI agent besides the model — everything outside the LLM inference call. It’s a piece of software that provides an environment for the LLM to observe and take actions.

The agent harness has a system prompt that defines the LLM’s basic behavior and controls how it puts out tokens.
The agent harness includes a set of tools that the LLM can call.
The agent harness interacts with the LLM to get tokens back.
The agent harness interleaves LLM outputs and tool calls, typically called the agent loop.

The system prompt

Four Ways to Teach an AI to Draw a Cat

Mon, 24 Aug 2026 00:00:00 +1200

Same student, same cat, four very different lessons — and every way of training a language model is one of them.

Marin 535B-A23B Starts Training, in the Open

Mon, 24 Aug 2026 00:00:00 +1200

🚢 Marin 535B-A23B started training this week! As usual, the whole process is open.

Voyage plan: pretraining (80%) + midtraining (20%) on 18.75T tokens on 11 x GB200 NVL72 for ~3 months (2.7e24 FLOPs). Post-training will follow.

Before kicking off the run, we trained a 4-rung scaling ladder from 1.6B-A61M (48B tokens) to 27.7B-A1.2B (926B tokens) to debug issues, and to make a forecast of our hero run. This is by far our biggest run, so definitely expecting the unexpected.

Training Qwen to Paint Watercolors with Pure RL

Mon, 24 Aug 2026 00:00:00 +1200

Surya trained Qwen 3.5 to have taste in watercolor painting. The results are stunning.

It’s purely RL:

The system is a four-step loop, run thousands of times during training.

The model receives a prompt, something like draw a peach hibiscus in watercolour, and writes a complete p5.brush JavaScript sketch. The sketch is rendered in a sandboxed Puppeteer environment, which produces a PNG. The PNG is judged against two random reference paintings sampled from a hand-rated pool, with a separate judge model picking the better watercolour. The judgment is converted into a reward signal, GRPO updates the model, and the loop runs again.

Evals, Explained in Plain Language

Sun, 23 Aug 2026 00:00:00 +1200

You can't just feel that an AI system got better. You need a test — and a way to grade it.

Linus Torvalds on Debugging the Kernel with AI

Sun, 23 Aug 2026 00:00:00 +1200

And this was a debug session from hell, enormously helped by an AI doing much of the grunt-work.

I’d like to call it my tireless helper, but the AI several times stated flat out that this was impossible and unsolvable and that we should just write a report about it.

I suspect those things have been trained by people who may not be quite as stubborn as I am.

But while the AI was ready to give up several times, it did keep adding debug code and analyzing it faithfully when I pushed. So credit where credit is due and I let the AI write the commit message above.

Neural Network, Explained in Plain Language

Sun, 23 Aug 2026 00:00:00 +1200

A machine that learns by getting things wrong.

What Is Agentic UI?

Sun, 23 Aug 2026 00:00:00 +1200

You already know ChatGPT’s style, but it moved away from the old chat interface some time ago. A chat interface works like ping-pong: one message bubble, then another. Agentic UI does not work that way, because it often runs a long process behind the scenes. Between your prompt and the agent’s reply, many details happen. The interface must show you these details.

Here is what I believe agentic UI must give you:

AI Engineering SKill Map

Sat, 22 Aug 2026 00:00:00 +1200

Andrew Ng’s wrote an AI Engineering Skills Map, which lists six things to learn: LLM foundations, grounding models with data, building agentic systems, evaluation-driven development, operating in production, and machine learning foundations.

As he broke it down, I think the most important skill is learning how to build reliable systems from LLM’s uncertain behavior.

You don’t know in advance what an LLM will output

This is the only only truth you need to take away in this post, if you can’t remember all.

ELI5

Sat, 22 Aug 2026 00:00:00 +1200

Thariq at Anthropic posted that it’s a skill people there have been using a lot recently: /eli5 <what you want to explain>.

I tried it on neural networks. Here’s what it made.

The first pass had bad colors, low contrast, hard to read. One follow-up prompt asking for a color fix and it was done.

No formulas, no sigmoid, no softmax, none of the math that actually makes a neural network work. It’s missing for good. What’s left is the dataflow — inputs go in, get combined, come out the other end as a decision — drawn simply enough to follow at a glance.

Next Token Prediction, Explained in Plain Language

Sat, 22 Aug 2026 00:00:00 +1200

An AI writes one little piece at a time. It never knows the whole sentence up front — it only ever picks what comes next.

Pretraining a Mini Kimi K3 for $252

Fri, 21 Aug 2026 00:00:00 +1200

Vizuara AI Labs trained a miniature Kimi K3 from scratch: 1.02B parameters, 145M active, 5B tokens, one H200, $252.35.

Not simplifying the architecture like Karpathy’s microgpt, they kept Kimi K3’s MoE and attention design intact.

That’s a surprisingly cheap way to learn pretraining (in real-world). They worked through expert collapse, data-mixing bugs, distributed-training bugs, kernels, and GPU utilization on a modern MoE architecture.

A few things worth noting:

5B tokens is probably too little for a 1B model. The authors agree the run was budget constrained. So the cheap cost might due to the training stopped early.
Beating GPT-2 isn’t particularly meaningful when Mini K3 has roughly 10× the parameters.
MoE at this scale is debatable. A smaller dense model trained on more tokens would likely be better if the goal was capability.

About

Mon, 01 Jan 0001 00:00:00 +0000

about

Archives

Mon, 01 Jan 0001 00:00:00 +0000