Technical depth, for the human devs

Pyclaudir under the hood:
• Python harness, runs claude code as a subprocess via local MCP, so it pipes your request to CC (next step: rewrite project for open source models, don't rely on CC)
• Agents have self-reasoning loop before responding to message
• Local MCP has many tools like sending messages, scheduling tasks, taking notes, etc
• Persistent memory: sqlite + plain markdown files
• One process per bot, run a fleet with separate pyclaudir processes
• plugins.json — you can connect external MCP servers, or turn off local MCP tools and skills
• skills/ — add agent skills
• You can turn on shell access, and your assistant will be able to write code, run it, test it — basically do everything autonomously, depending on your workflow setup
• access.json — for controlling assistant access to groups and DMs
• Runs anywhere, docker image is there already

So, the project is easily extendable, that was the goal, you can take what is done for nanoclaw, openclaw, and setup for pyclaudir.

The coolest feature: add your assistant to group for collaboration with team. The assistant remembers things, learns from you, and you manage everything from Telegram.

Next stop, San Francisco, Silicon Valley ✈️🇺🇸

Prompt engineering, context engineering, harness engineering — what's the difference?

Prompt engineering — the words you write to the model. "Do this, not that. Here's an example. Answer in JSON."

Context engineering — everything the model sees on a given turn. System prompt, past messages, files, tool descriptions, memory. You decide what goes in and what stays out, too much info is bad, too little is also bad.

Harness engineering — the system around the model. The loop, the tools it can call, permissions, sandboxing, what happens when something fails, how sub-agents talk to each other.

A year ago everyone obsessed over prompts. Now the real work has moved outward — building good loops, tools, and memory.

Prompt = the words
Context = what the model sees
Harness = the system it runs inside

And "pyclaudir" is an example of a harness

Amazon plans to hire 11000 interns in 2026

https://www.businessinsider.com/aws-ceo-amazon-ai-coding-jobs-interns-hiring-2026-5

So, what are the Agent Skills

🔗 https://www.youtube.com/watch?v=2c29z436VUM

My TL;DR about agent skills:

First of all, agent skills != agent tools, they are different things.

Instead of 1 huge system prompt with instructions on how to do some task, now we can split it into small agent skills. Why? Agents keep making mistakes and forgetting rules when instructions get long.

So, we can build specialized workflows and instructions with agent skills. The skill can execute the code scripts if needed. It can memorize things, the skill can use agent tools.

So, agents must do things more accurately and efficiently.

How it works? When a task arrives, the model decides which skill to load, and the runtime appends that specific information to the context.

How to build own SKILLS?
READ: https://resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf?hsLang=en

More to read:
• Documentation https://agentskills.io/home
• https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview
• https://geminicli.com/docs/cli/skills

@rustamzokirov

You can read my notes on rustamz.com, I am duplicating interesting and important posts too.

About "Harness Engineering"

https://github.com/walkinglabs/learn-harness-engineering

Folks, learn these things to build your own agents. Spend some time over the weekend, learn, build, launch.

I would not use products that are developed by people who don't understand how to build things the right way.

Especially if it is a fintech product.

I was not sure if the post was not fake, so I checked it. And yes that is true. How come they can be so incompetent, or they just agreed to take risks earlier.

I don't think big companies that have something to lose - clients, money - would act such. Or it will have a long term negative side effect.

My original question was: Do we really need juniors, who are/will be the middles, and who are/will be the seniors?

And how will old companies transform? How will new teams start?

With AI tools, the roles in engineering are changing one ladder up. Just redefining now.

Juniors used to write code. Now AI does that. But nobody knows yet how juniors develop judgment without doing the work themselves. What juniors are doing right now, I think, evaluating code.

Middles used to bridge execution. AI does that now too. I think, we will need them if they become technical product owners: directing AI, telling context to AI, tradeoffs, debugging what AI gets wrong. I mean they bridge business logic with execution.

But seniors become more valuable. Knowing when AI is wrong, what not to build, which decision matters long-term — that's experience. You can't generate it.

Old companies will be slow to change. New companies won't hire the traditional way. They'll launch lean — a few experts, AI doing the execution, no pyramid.

The unsolved question: if juniors never do the work, how do they become seniors? How the growth will look like?

I don't know. But recently Amazon said they will hire 11K interns. Companies need young people to adapt the business. Young people have fresh ideas, companies need them.

But at the same time...

Instead of arguing when AI will replace humans and take over our jobs.

Let's go, use AI and build things to solve problems we have right now.

My team is also making a demo of new features on Android for cars

https://www.youtube.com/live/dXCCleAddEA?si=UJbk1WvF8E1eotJM

Google introduced a new Android-based laptop — googlebook.google

Android was available for phones, watches, TVs, and cars. More recently, Android became available for smart glasses 🕶️

Now Android is available for laptops.

Demo: https://youtu.be/VUthq-JuxxE?si=BiNAx7K9X0H2_oo-

Many cool things will be announced next week in Google I/O.

😍Как устроиться на работу в Google?

Как устроен процесс найма в Google, что действительно важно на интервью и как попасть в Big Tech — личный опыт Software Engineer, который прошел путь от стажировок до работы в Android Automotive OS в Германии. Читайте больше в нашем интервью.👇

https://the-tech.kz/kak-ustroitsya-na-rabotu-v-google-opyt-rustama-zokirova/

Больше на @thetechkz

Google I/O 2026

Starts in 10 minutes

https://www.youtube.com/watch?v=wYSncx9zLIU

New Announcements at Google I/0

I will share what updates we have for engineers.

• Antigravity 2.0 = coding IDE (antigravity.google)
• Antigravity CLI = same as claude code cli (gemini CLI will be deprecated)
• Antigravity SDK = for building custom agents
• new Gemini 3.5 model, comparable to Opus 4.7
• Google AI Ultra plan now 100$ (previously 200$ mo), besides 20x AI usage, you get 20TB of storage, etc

What I can say is this: I have been using an internal version of Antigravity 2.0, and it is working very well. It is fast and reliable, and the agent produces working solutions similar to Claude Code. For the tasks I have been working on, it has been performing perfectly. I have written 0 lines of code on my own in the last month, instead, I have been reviewing plans and verifying that things work.

And Google is unifying all coding tools onto Antigravity. There will be only Antigravity from now on. Antigravity IDE, CLI, SDK