World's first harness Workshop by Dr. Raj, MT PhD | Starting Aug 3rd

[00:00:00]Hello everyone. I'm recording this video to announce the launch of the world's first harness engineering workshop. The workshop will start on August 3 and last for 5 days, until August 7. First came prompt engineering, then contextual engineering, and now we have what we call harness engineering. So, this is point A, from which we will start our workshop.

[00:00:34]And by the time you finish, you will reach point B. And what is point B? At point B, you will be able to create harnesses like Claw Code from scratch. You will understand how agentive " bindings" like Claw Code are built from the ground up. You will consider such "straps" as Pi.

[00:00:57]You'll also look at a very popular wrapper that has about 200,000 stars on GitHub called AutoGPT, and you'll be able to build your own agent-based wrapper completely from scratch. Each day of the workshop will be focused on achieving this goal, and we will also create a final project in which you will apply your own "strap ". Harness engineering is an incredible field that combines several domains. This is truly a superset. Therefore, prompt engineering and context engineering are subsets of harness engineering.

[00:01:36]By mastering harness engineering, you will understand how a coding agent works from the ground up, which is a very useful skill in engineering today.

[00:01:47]That is why we decided to hold this workshop. The workshop timings are from 7:00 AM to 9:00 AM Indian Standard Time (IST) and will run from August 3 to 7. We will build a "pyramid" during the workshop, and it essentially consists of starting with so-called loops.

[00:02:15]There is a new term for this—“loop engineering.” So, we will look at the origin of this term. Then we'll look at how to add tools to these cycles. Next, we will look at context management. Then we will move on to subagents. And after that, we will also consider security issues. The agentive " binding" essentially contains all of these elements. And within that, we'll delve into even more detail.

[00:02:43]For example, here we will look at prompts.

[00:02:47]Here we will look at skills. In terms of security, we will also look at something called sandboxing. In the topic of subagents, we will look at agentive orchestration. This will be an incredible and intense workshop where we will constantly be creating something. So, I'm going to do this workshop where we're going to cover the basics, of course, but every day we're going to build something, and at the end of it, you're going to be one of the few people in the world who has your own agent framework built from scratch.

[00:03:22]Many of us only know about Claude Code, but there are many more open source agent frameworks, such as Pi, whose entire repository on GitHub is fully accessible.

[00:03:35]Before this workshop, I studied this repository, and we will be replicating our own agent framework, very similar to how Pi was built. As I mentioned, Hermes is another such framework that has about 200,000 stars on GitHub right now. This is incredible. The GitHub repositories for both of these projects provide incredible ideas on how you can build your own framework without having to rely on Claude Code every time, which is a bit expensive. We must learn to think independently about creating frameworks from scratch.

[00:04:08]Remember: if the LLM is the engine of a car, then the frame is the entire car. And as a machine learning engineer or LLM engineer, you should have knowledge of agent frameworks.

[00:04:20]You have to understand exactly how this car was built. I look forward to meeting you during the workshop. Thank you.