Working Notes on Agent Systems/Brad Zhang

Topic desk

Technical Distribution

How does technical work become findable, inspectable, and worth trusting without turning into content marketing?

Distribution is treated as part of the engineering surface: repositories, essays, field notes, and search-facing pages all reinforce the same proof system.

Live source unreachable; showing the most recent snapshot.

Longform

All longform

OpenAI Proactively Discloses Long Range Agent Tasks Out of Bounds: The Safe Steering Behind PR # 287

A single tool approval begins to lose global semantics when the model can be tried for hours on end. On July 20, 2026, OpenAI unveiled an on-premises pause. A generic model that...

Jul 21, 2026 · 12 min read

A summary of the best articles on Graph Engineering! Prompt, the next engineering parad...

A summary of the best articles on Graph Engineering! Prompt, the next engineering paradigm from Loop to Graph, determines how the Agent starts. Context determines what the Agent...

Jul 20, 2026 · 4 min read

The Agent's memory should not be stuffed into the context window

When many people discuss coding agents such as Codex and Claude Code, their first reaction is model capabilities: whether they can read code, correct bugs, and...

Jul 17, 2026 · 8 min read

Indeed, when people are in this mode of small input and massive output, it will really affect their thinking and brain.

Indeed, when people are in this mode of small input and massive output, it will really affect their thinking and brain. Not only will they become sluggish, but it will...

Jul 16, 2026 · 2 min read

I tested GPT-5.5 and GPT-5.6 SOL using 5 real front-end scenarios: SOL won 4 games but was 13% slower

On July 10, 2026, I put GPT-5.5 and GPT-5.6 SOL into the same Codex execution chain and performed five sets of front-end tasks in succession. Five scenes, ten pages, and...

Jul 11, 2026 · 12 min read

fireworks-tech-graph Biggest update yet! 🎉This was originally a small open source project and has

fireworks-tech-graph Biggest update yet! 🎉This was originally a small open source project and has accumulated 8.5k stars after several months of development. Thank you...

Jul 10, 2026 · 3 min read

So the question is, who will be the third child of the Yusan family? 🤔 On Claude's side, Sonnet 5, Fable 5

So the question is, who will be the third child of the Yusan family? 🤔 On Claude's side, Sonnet 5, Fable 5, and Mythos 5 have already rolled out the model echelon, and...

Jul 10, 2026 · 4 min read

My first heavy duty job with GPT 5.6 sol blew me away, kudos! I suggest you send this sentence to your Codex

My first heavy duty job with GPT 5.6 sol blew me away, kudos! I suggest you send this sentence to your Codex and wait silently for the output results. You will come back...

Jul 9, 2026 · 1 min read

gpt 5.6 sol is loaded, reset, start biu biu biu Don't go to the typhoon on the weekend,...

gpt 5.6 sol is loaded, reset, start biu biu biu Don't go to the typhoon on the weekend, just stay home and code with 5.6 sol. What is the biggest difference between gpt 5.6 sol...

Jul 9, 2026 · 2 min read

Anthropic’s 85-minute Fable 5 workshop explains how to implement the next generation AI Agent

Video entrance Opening keynote: https://www.youtube.com/watch?v=GMIWm5y90xA The capability curve: https://www.youtube.com/watch?v=tP4MGcJ80Y0 How to get to production...

Jul 8, 2026 · 18 min read

The significance of J-space is to give the previously black-box model reasoning an opportunity for

The significance of J-space is to give the previously black-box model reasoning an opportunity for observability and human intervention and control, so that corrections can...

Jul 7, 2026 · 2 min read

The term software factory has finally begun to move from a slogan to engineering sites.

Original video: WF2026: Software Factories & Keynotes ft. Microsoft, OpenAI, OpenClaw, Z.ai (GLM), MiniMax, HF Link: https://www.youtube.com/watch?v=htM02KMNZnk Start I just fin...

Jul 4, 2026 · 8 min read

After pre-training, what does the next training paradigm for AI look like?

Source Original video: What does the next training paradigm look like? Channel: Dwarkesh Patel Link: https://www.youtube.com/watch?v=20p5-kQXF_Q Start In this video, Dwarkesh Patel

Jul 2, 2026 · 15 min read

Run the SEO/GEO visibility loop once a week. The SEO/GEO visibility loop in the Loop Li...

Run the SEO/GEO visibility loop once a week. The SEO/GEO visibility loop in the Loop Library is very suitable for content-based products, developer tools and SaaS official websi...

Jun 22, 2026 · 1 min read

The most worth-watching high-quality videos in the field of AI in the past two weeks! 1...

The most worth-watching high-quality videos in the field of AI in the past two weeks! 1. Official Team - Bloomberg Odd Lots × Anthropic Co-Founder: Anthropic's Co-Founder and To...

Jun 21, 2026 · 2 min read

From Prompt Engineering to Harness Engineering: four evolutions of the AI ​​engineering system

The rapid development of products such as Agent, Codex, and Claude Code has led to the emergence of a number of new terms in the field of AI engineering: Prompt...

Jun 19, 2026 · 7 min read

Summary of the best articles of Loop Engineering! After Agent began to focus on long tasks in 2026,

Summary of the best articles of Loop Engineering! After Agent began to focus on long tasks in 2026, the focus slowly became: How to design a loop system that can...

Jun 18, 2026 · 2 min read

Let's say you're the head of technology at a technology company. Last year, you used GraphRAG to build an

Let's say you're the head of technology at a technology company. Last year, you used GraphRAG to build an internal knowledge base and used 20,000 contracts to do PoC. The...

Jun 17, 2026 · 17 min read

How to burn token series? (1) Use Dynamic Workflow to let 100+ agents work in parallel

If you look at the picture below, it should basically cover the daily usage scenarios of ClaudeCode or Codex partners. These are some intermediate processes of a website...

Jun 15, 2026 · 4 min read

Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open

Another big job, very satisfying! Combining dynamic workflow mode + Loop Engineering, we created an open source project that replaced Claude Design: Fireworks-design...

Jun 14, 2026 · 2 min read

Claude Fable is for enterprises to replace traditional mid-to-high-level development, and the ROI is positive.

Claude Fable is for enterprises to replace traditional mid-to-high-level development, and the ROI is positive. After the release of the Claude Fable model, I saw many crazy exam...

Jun 10, 2026 · 4 min read

Agent Loop is popular🔥: Agent goes from Demo to production with a Loop in between

There has been a heated debate in overseas AI circles these days. Matt Van Horn sorted out this debate in “WTF Is a Loop? Peter Steinberger vs. Boris Cherny”: You shouldn’t be p...

Jun 9, 2026 · 10 min read

Five Traps of Agent Harness 1. Self-evaluation is a trap. Use an adversarial evaluator....

Five Traps of Agent Harness 1. Self-evaluation is a trap. Use an adversarial evaluator. Self-evaluation is a trap. Use an adversarial evaluator. Many teams will do this: Agent g...

Jun 8, 2026 · 3 min read

Why does it seem easy to develop an AI Agent, but so difficult to actually make it "useful"? Where are the main bottlenecks?

Let me ask you a question, have you heard of or used more than half of the Agent products mentioned in the picture below? More than a third? Over the past 25 years, in less than...

Jun 7, 2026 · 8 min read

If you are a heavy Codex user, do you always have request limit on a few days every month?

If you are a heavy Codex user, do you always have request limit on a few days every month? Starting from this month, the 100-dollar Pro user quota has returned to the regular 5X...

Jun 1, 2026 · 2 min read

Asynchronous Agent Era: From Code Assistant to Software Factory

I have reorganized the content of this podcast of Latent Space. It is worth taking a look at the in-depth discussion and sharing related to asynchronous Agent. I also...

May 31, 2026 · 22 min read

My wife recently started eating natto at home. I originally didn’t like the sticky feel...

My wife recently started eating natto at home. I originally didn’t like the sticky feeling, but I saw that natto has high nutritional value and is very healthy. The most importa...

May 31, 2026 · 3 min read

In the afternoon, I was surprised when I saw the news about the layoffs of a large factory posted by @MaxForAI, and then I sent a private message to verify it.

In the afternoon, I was surprised when I saw the news about the layoffs of a large factory posted by @MaxForAI, and then I sent a private message to verify it. I still believe M...

May 27, 2026 · 7 min read

This DeltaBox solves the sandbox state rollback problem of Stateful AI Agent. Agents wi...

This DeltaBox solves the sandbox state rollback problem of Stateful AI Agent. Agents will continuously change the file system, process status, tool status, and environment varia...

May 25, 2026 · 1 min read

Codex’s weekly active developers exceed 4 million! Codex is definitely OpenAI's most po...

Codex’s weekly active developers exceed 4 million! Codex is definitely OpenAI's most popular chicken. It has reversed the decline in one fell swoop and turned from defense to of...

May 22, 2026 · 3 min read

FSD enters China: Autonomous driving begins to fight the "long-tail war". After China is included in Tesla's

FSD enters China: Autonomous driving begins to fight the "long-tail war". After China is included in Tesla's FSD Supervised country list, the focus of autonomous driving...

May 21, 2026 · 5 min read

Visualize that with the rapid development of agent cli and harness, the interaction and logic core will soon

Visualize that with the rapid development of agent cli and harness, the interaction and logic core will soon be reconstructed by Golang, which is very suitable for...

May 19, 2026 · 1 min read

There are more than a hundred and five thousand subscriptions. I don't know if there wi...

There are more than a hundred and five thousand subscriptions. I don't know if there will be any surprises when I wake up. I often write in advance to celebrate the 5k subscript...

May 19, 2026 · 2 min read

Is Grep All You Need? If you are still thinking of “rag = Vector Database” as the defau...

Is Grep All You Need? If you are still thinking of “rag = Vector Database” as the default answer, then I recommend that you read this paper. It is a bit counterintuitive: the au...

May 16, 2026 · 3 min read

Based on the recommendation of the boss, I combed through a list of high-quality AI Engineer learning materials, which is worth collecting and learning!

Based on the recommendation of the boss, I combed through a list of high-quality AI Engineer learning materials, which is worth collecting and learning! Too dry, too...

May 11, 2026 · 3 min read

The Agent memory system is moving from "storing more and more" to "using more and more accurately".

The Agent memory system is moving from "storing more and more" to "using more and more accurately". Mem0's latest release, Memory Decay, is noteworthy. It addresses not...

May 9, 2026 · 2 min read

HTML might be the real interface of the agent era. Markdown is great for notes. But age...

HTML might be the real interface of the agent era. Markdown is great for notes. But agents do not just need notes. They need interfaces they can understand, manipulate, simulate...

May 9, 2026 · 2 min read

It is highly recommended that you read this long article written by Thariq, a core member of ClaudeCode: he brings an interesting perspective.

It is highly recommended that you read this long article written by Thariq, a core member of ClaudeCode: he brings an interesting perspective. HTML is the real working interface...

May 9, 2026 · 2 min read

This development brother has open-sourced a job-seeking artifact with technical content.

This development brother has open-sourced a job-seeking artifact with technical content. The 🏷️core competencies are: AI empowerment. The whole process of job-seeking can...

May 8, 2026 · 3 min read

Anthropic has just published a very important research agenda. AI is now a systemic iss...

Anthropic has just published a very important research agenda. AI is now a systemic issue of economic diffusion + security resilience + social governance + accelerated scientifi...

May 8, 2026 · 2 min read

Don’t treat impressions as traffic: I looked at getdesign.md and found that it is more like a programmatic SEO site that is very stable and worth learning.

Don’t treat impressions as traffic: I looked at getdesign.md and found that it is more like a programmatic SEO site that is very stable and worth learning. I first...

May 4, 2026 · 5 min read

Try this cli radio, you can listen to music, podcasts and news while vibe coding. Why d...

Try this cli radio, you can listen to music, podcasts and news while vibe coding. Why do you want to do this? The reason is very small, just want to make some sound when writing...

May 3, 2026 · 3 min read

Using Codex goals well can truly unleash the potential of the GPT 5 model! OpenAI's rec...

Using Codex goals well can truly unleash the potential of the GPT 5 model! OpenAI's recently disclosed directions for Codex are obviously going in the following directions: - Lo...

May 1, 2026 · 6 min read

This fan lamented that it was so difficult to get the ICML paper approved, and he carefully checked the probability that it felt like the imperial examination.

This fan lamented that it was so difficult to get the ICML paper approved, and he carefully checked the probability that it felt like the imperial examination....

May 1, 2026 · 3 min read

How to make Codex Silky continue to run long tasks? Make good use of these two commands...

How to make Codex Silky continue to run long tasks? Make good use of these two commands codex --full-auto: suitable for 80% of daily scenarios. If your goal is to have Codex int...

Apr 27, 2026 · 2 min read

This direction is interesting: Skill +RAG The core pain point of the traditional search-enhanced generation

This direction is interesting: Skill +RAG The core pain point of the traditional search-enhanced generation (RAG) system is the persistent failure that cannot be solved...

Apr 21, 2026 · 3 min read

RLMs are the new inference models. Inference models are the first clear demonstration t...

RLMs are the new inference models. Inference models are the first clear demonstration that language model capabilities can be extended with test-time computation. Recursive lang...

Apr 21, 2026 · 2 min read

One of the most annoying things about writing technical articles: drawing pictures. The...

One of the most annoying things about writing technical articles: drawing pictures. The structure that is very clear in my mind is always misaligned, the color is ugly, and the...

Apr 10, 2026 · 1 min read

Why does ClaudeCode use plain text + file system as its RAG system? Claude Code adopts a "plain text + file

Why does ClaudeCode use plain text + file system as its RAG system? Claude Code adopts a "plain text + file system" RAG design and completely abandons their...

Apr 10, 2026 · 2 min read

Lin Junyang’s latest must-read masterpiece! AI has shifted from "reasoning thinking" to "agentic thinking".

Lin Junyang’s latest must-read masterpiece! AI has shifted from "reasoning thinking" to "agentic thinking". 🏀The frontier of competition is shifting from "better...

Mar 26, 2026 · 3 min read

Why do many claude code query tasks require the use of file search commands? Is there no other better way? I

Why do many claude code query tasks require the use of file search commands? Is there no other better way? I asked CC this question, and it gave me a very detailed answer...

Mar 23, 2026 · 2 min read

🎬 Why I built Media Downloader for Claude Code — and why it's different from everything else out there.

🎬 Why I built Media Downloader for Claude Code — and why it's different from everything else out there. The problem: Getting stock images and videos is TEDIOUS. Open browser →...

Jan 20, 2026 · 2 min read

With this skill, you no longer need to manually search for picture websites, and no longer need to copy and paste links!

With this skill, you no longer need to manually search for picture websites, and no longer need to copy and paste links! ! Just say one sentence: "Help me download 10 xxx pictur...

Jan 20, 2026 · 3 min read

Experience sharing of using ob + claudian + palywright-skills + ob-skills + general-purpose-skills to generate visual logs

At one o'clock in the morning, I stared at the Canvas file just generated in Obsidian, feeling a little dazed. Figure 1: Automatically generated visual log...

Jan 14, 2026 · 10 min read

If you want to write a long article on X and get a lot of exposure, you can read this summary!

If you want to write a long article on X and get a lot of exposure, you can read this summary! How to maximize the exposure of Articles on the X platform? 2-3 Articles per week,...

Jan 10, 2026 · 2 min read

Oh My God! X has come out with a new algorithm in the past few days in 2026. No wonder...

Oh My God! X has come out with a new algorithm in the past few days in 2026. No wonder my traffic is so low, Shet! Everyone must take a look at the new regulations in 2026: Janu...

Jan 10, 2026 · 2 min read

ClaudeCode Common Function Manual

📚 Core commands 📋 Command cheat sheet Command icon describes the frequency of use in one sentence /help🆘View the help document⭐⭐⭐⭐⭐ /clear🧹Clear the...

Jan 9, 2026 · 9 min read

What will future travel and transportation look like based on AI? Such as Cybercab, Robovan and FSD

Musk’s blueprint for autonomous transportation, Cybercab, is a vehicle that completely eliminates the steering wheel and pedals and relies entirely on Tesla’s FSD...

Jan 7, 2026 · 12 min read

A historic moment for the world's first large model listed company. On January 8, 2026, as the world's first

A historic moment for the world's first large model listed company

Jan 7, 2026 · 17 min read

The Scientific and Commercial Apotheosis of Zhipu AI

The ascension of Zhipu AI to the status of the world's first publicly traded large-scale model corporation on January 8, 2026, represents a seminal juncture in the history of co...

Jan 7, 2026 · 10 min read

Manus Wide Research: 100+agent parallel research system Manus is the agent, achieving the most advanced performance in the GAIA benchmark test.

Manus Wide Research: 100+agent parallel research system Manus is the agent, achieving the most advanced performance in the GAIA benchmark test. Its Wide Research...

Jan 7, 2026 · 1 min read

I suggest you all listen to this issue. I have briefly summarized the following points:...

I suggest you all listen to this issue. I have briefly summarized the following points: 🏷️About the underlying understanding of AI and Agent (suitable for showing depth) About...

Dec 30, 2025 · 3 min read

Introducing a company and product created based on the founder model: Mowen "Mowen" product introduction "Mowen" is a mini program product based on the WeChat ecosystem.

Introducing a company and product created based on the founder model: Mowen "Mowen" product introduction "Mowen" is a mini program product based on the WeChat ecosystem. The web...

Dec 29, 2025 · 5 min read

With this Orange article, I will continue to share some content about ClaudeSkill and MCP to attract some traffic, haha.

With this Orange article, I will continue to share some content about ClaudeSkill and MCP to attract some traffic, haha. Analysis of the core features of Claude Skills....

Dec 29, 2025 · 5 min read

This McKinsey report reveals an embarrassing fact three years after the outbreak of the AI ​​industry: Most

This McKinsey report reveals an embarrassing fact three years after the outbreak of the AI ​​industry: Most companies have just started piloting AI, and there has...

Dec 26, 2025 · 2 min read

It’s 2025, and you still believe the nonsense that big model manufacturers only do “infrastructure”?

It’s 2025, and you still believe the nonsense that big model manufacturers only do “infrastructure”? Look at your mobile phone. The giant that once only provided APIs is now har...

Dec 24, 2025 · 3 min read

A one-person company makes 4 million US dollars? Leverage of a one-person company: HeadshotPro's "lazy"

A one-person company makes 4 million US dollars? Leverage of a one-person company: HeadshotPro's "lazy" philosophy that earns $4 million a year 🔑 Key points 1. Danny Postma has...

Dec 17, 2025 · 2 min read

Consulting has always been big business! A single order worth tens of millions of dolla...

Consulting has always been big business! A single order worth tens of millions of dollars. The consulting industry used to combine talents, research, information gaps, personal...

Dec 15, 2025 · 3 min read

Reprinted from an old article by Pinduoduo Huang Zheng. This article wrote about his th...

Reprinted from an old article by Pinduoduo Huang Zheng. This article wrote about his three years of working at Google. Take a look at what Google looks like in the eyes of the b...

Dec 9, 2025 · 6 min read

Prompt: A 2K resolution contact sheet, arranged in a 3x5 grid of 15 individual film frames, representing the complete second episode of an action-comedy series.

Prompt: A 2K resolution contact sheet, arranged in a 3x5 grid of 15 individual film frames, representing the complete second episode of an action-comedy series. **[Characters &...

Nov 30, 2025 · 2 min read

IIya explained everyone's misunderstanding of some of his views: He believes that the current scaling route

IIya explained everyone's misunderstanding of some of his views: He believes that the current scaling route can make progress without leading to stagnation, but based on...

Nov 28, 2025 · 3 min read

I didn’t expect my last post to be read so many times, I’m really flattered 📚(¯∇¯;) I dare not slack off and continue to provide you with useful information.

I didn’t expect my last post to be read so many times, I’m really flattered 📚(¯∇¯;) I dare not slack off and continue to provide you with useful information. If it is useful, I...

Nov 24, 2025 · 3 min read

From losing 600,000 yuan in the stock market to establishing the top 3 AI community in China, my journey has been made public for the first time!

From losing 600,000 yuan in the stock market to establishing the top 3 AI community in China, my journey has been made public for the first time! Three years ago, 35+, programme...

Nov 23, 2025 · 3 min read

Dan Koe’s latest post! Expressing yourself well is an attractive trait. Quick summary:...

Dan Koe’s latest post! Expressing yourself well is an attractive trait. Quick summary: Pain points - If I don't know how to start a section, I start with a relevant pain point a...

Nov 22, 2025 · 2 min read

⚡️ The aesthetics of traffic violence of 21-year-old Chinese genius Roy Lee: From being expelled from

⚡️ The aesthetics of traffic violence of 21-year-old Chinese genius Roy Lee: From being expelled from Harvard/Columbia University to earning $500k a month Core logic...

Nov 22, 2025 · 2 min read

This year I saw a lot of similar AI products focusing on memory or virtual companionship, which reminded me of a TV series from a few years ago.

This year I saw a lot of similar AI products focusing on memory or virtual companionship, which reminded me of a TV series from a few years ago. The general plot of "Upload to R...

Nov 20, 2025 · 2 min read

Claude recently added an official skill specifically designed to improve front-end interface design.

Claude recently added an official skill specifically designed to improve front-end interface design. I tried installing it and it worked really well. The screenshot is...

Nov 16, 2025 · 4 min read

Claude Sonnet 4.5 and Opus 4.1 start to support structured output! The benefits are sel...

Claude Sonnet 4.5 and Opus 4.1 start to support structured output! The benefits are self-evident, developers no longer have to deal with the trivial manual work of structuring C...

Nov 15, 2025 · 1 min read

The new idea of ​​​​OpenAI training model may be a new direction! 1. Previous dilemma:...

The new idea of ​​​​OpenAI training model may be a new direction! 1. Previous dilemma: a mess In the past, when we tried to understand the AI ​​brain, we discovered a big proble...

Nov 14, 2025 · 3 min read

Will there be a bubble in AI? 🤔Where might it be? when? I saw this picture yesterday, it was made by Goldman

Will there be a bubble in AI? 🤔Where might it be? when? I saw this picture yesterday, it was made by Goldman Sachs. The comparison is the growth trend brought about by...

Sep 27, 2025 · 3 min read

380 billion, 300+ open source models, it’s hard to get a ticket, the stock price has hit the limit, the familiar Alibaba is back!

380 billion, 300+ open source models, it’s hard to get a ticket, the stock price has hit the limit, the familiar Alibaba is back! The first picture is that classmate Mengfei was...

Sep 24, 2025 · 3 min read

RAG Technology Toolbox Layered Memory Retrieval: The simplest way to load all documents directly into the context window of LLM.

RAG Technology Toolbox Layered Memory Retrieval: The simplest way to load all documents directly into the context window of LLM. It is suitable for scenarios where the amount of...

Aug 10, 2025 · 2 min read

This article talks about multi-agent collaboration very well. I have recently observed...

This article talks about multi-agent collaboration very well. I have recently observed that starting from Calude’s multi-agent architecture article on building deep research, to...

Aug 7, 2025 · 2 min read

I blindly guess that GPT 5 will be a big mess tonight. I have already lost trust in Aotuman🥲 Personally,

I blindly guess that GPT 5 will be a big mess tonight. I have already lost trust in Aotuman🥲 Personally, I feel that GPT is becoming more and more mediocre. In summary...

Aug 4, 2025 · 2 min read

Field Notes

Full ledger

Live judgment for this desk is filed by date in the field-notes ledger; enter through the full ledger by month.