kurisina

AI researcher | Content creator

[kurisina @kurisinaa]

Google CEO Sundar Pichai calls 2027 an inflection point for agents

Devs on the frontier are already "deploying agents, orchestrating agents"

If you don't learn how to build a harness that runs them now
You'll spend 2027 catching up with those who started today

Researcher → Writer → Tester → Reviewer → Fixer

In this interview, Sundar explains why top engineers are moving from writing code to operating agent harnesses

The harness manages the agents

The human becomes the operator, not the bottleneck

Agents → Harness → Feedback → Verification → Autonomous Systems

This interview is worth more than most $3,500 agent engineering courses

Bookmark it and watch it today

Then read the full harness engineering guide below ↓

[kurisina @kurisinaa]

Jev founder Diogo Almeida (ex-OpenAI):

"Claude Code and Codex are not the next era. They still belong to the assistant era, with a human in the loop. JEV is what comes after that for LLMs

x200 faster, x400 cheaper, 0 hallucination, no human in the loop. That's JEV, and that's what LLMs are going to look like"

in a 36-minute tech talk he breaks down why RLHF is already done and how the next generation of LLMs gets built

you'll get more out of these 36 minutes than out of a Stanford ML degree

watch it today, whatever else is on your list, then read the article below on how to become a Jev Engineer

[Asteri @Asteri_eth]

https://t.co/hJUKtndNOL

[kurisina @kurisinaa]

Jane Street engineer:

"at Jane Street, 85% of our quants are running dozens or hundreds of AI agents

This helps us develop our own LLM, which we use for trading and market research to make $20.5B/year"

In 15 min he breaks down how Jane Street uses AI inside their fund - and how they trained their own model that prints money for them

Bookmark & watch

[kurisina @kurisinaa]

Andrej Karpathy:

"Prompting is going away. Delete everything, keep Graph."

In 69 minutes he shows how to build Graphs, and why it's the only thing left standing at the end

Prompts → Agents → Loops → Graphs

a loop keeps one agent working until the job is done

a graph decides which agents exist and what each one hands to the next

anyone can build an agent, almost no one builds the graph that runs them

watch it today

[kurisina @kurisinaa]

SpaceXAI engineer (ex-Cursor):

"250 applications, two years, zero offers. Then he stopped applying as one engineer and showed up as one running a squad of agents. We offered him $850k.

99% of candidates still sit in a single chat window. The ones we hire run a Chief of Staff on top of CloseBot, ProdBot, StalkBot and ProtoBot, wired into graphs and loops that ship overnight while they sleep.

One agent saves you an hour. A squad replaces the staff function you can't afford to hire.

Not a skill gap. A stack gap. One evening closes it."

GrokBot → Chief of Staff → CloseBot / ProdBot / StalkBot / ProtoBot → Graphs & Loops → Autonomous Workflows

A SpaceXAI engineer just dropped a 1-hour workshop on building a GrokBot squad from scratch.

Worth more than most $1000 AI courses.

Watch it today.

[kurisina @kurisinaa]

Andrew Ng:

"100% of my tasks now run through AI agents

The hype has actually blown past my expectations
The harness around the model is the next step"

"In 3-6 months, everyone will be building agent harnesses

Prompting alone won't cut it"

In this 30-minute talk, Andrew Ng breaks down how to build AI agents that work reliably beyond a single prompt

Model → Agents → Harness → Feedback → Self-Improving Systems

Worth more than most $1500 AI engineering courses

Bookmark it and watch today

[kurisina @kurisinaa]

@0xCodez truly intelligent words

[kurisina @kurisinaa]

SpaceXAI engineer Lauren Tan:

"I'm running 20+ GrokBot agents with /loop and /goal inside pstack

My bots ship code while I sleep

At SpaceXAI, 85% of engineers already run 10+ GrokBot agents in parallel

One Chief of Staff agent knows every other bot and runs the whole stack"

GrokBot → /loop → /goal → Chief of Staff → Agent Teams

In a 1-hour workshop, Lauren walks through building a team of GrokBot agents from scratch

Research → Code → Review → Ship

Beats most $1500 agent engineering courses

Bookmark it and watch today

[kurisina @kurisinaa]

Sam Altman, CEO of OpenAI:

"You don't need better prompts. You need to master agent engineering with GPT-6 Astra."

99% of people are still fiddling with wording in one chat window. The 1% split the work across agents and let the system do the running. That's not a prompting gap, it's an agent engineering gap, and one evening is enough to close it.

Prompt → Agents → Delegation → Orchestration → Autonomous Systems

In 21 minutes he breaks down what OpenAI engineers actually do differently, and how they split and orchestrate work between agents.

Build → Split → Orchestrate → Ship

These 21 minutes beat every paid agent course I've seen.

[kurisina @kurisinaa]

Stanford just dropped a free course on building ai agents from scratch

"Anyone can build an agent in 60 minutes"

the flow: prompt → agent → automation → revenue

Timestamps that matter:

00:00 - your first agent, built live
48:17 - no-code path for non-devs
54:39 - the $100k/month agent playbook

While you're scrolling, someone else is quietly learning a skill anthropic pays $750k for

This free course beats most $500 paid ones i've reviewed

[kurisina @kurisinaa]

SpaceXAI engineer Peng Zheng:

"250 applications, two years, zero offers. Then he stopped applying as one engineer and showed up as one engineer running 20 agents. We offered him $850,000.

I stopped using GrokBot like Google. I built the system once and now it runs 95% of my life and work on its own. Only 1% of people set up a single Chief of Staff agent that manages the other ~20 and knows everything about each one.

That's not a skill gap. That's a stack gap, and it takes one evening to close."

Chief of Staff → 20 Agents → Auto-Delegation → 24/7 System

40-minute workshop from SpaceXAI engineers on how to stop babysitting agents one by one, and how they actually run GrokBot at home.

Research → Build → Launch → Improve

Saves you 20 hours of useless agent tutorials.

[kurisina @kurisinaa]

Google Brain founder Andrew Ng:

"Prompting will be dead in 6 months.
Agent harnesses built with loops and graphs will replace it."

Agent → Harness → Feedback → Loops → Graphs → Self-Improving Systems

1-hour Stanford lecture. He shows what top engineers are building instead and where to start today.

Prompt → Run → Verify → Improve

The first 20 minutes deliver what most $1,500 courses charge for.

For free.

Bookmark and watch it today.

[kurisina @kurisinaa]

SpaceXAI engineer, Lingxi Li:

"Two years of shipping like everyone else, one ticket at a time. Then he stopped writing code and started building the team that writes it. That profile is what we pay $850,000 for.

My team of 10+ GrokBot agents ships code at 3 am. A Chief of Engineers agent sits above the PM and 10+ coding agents. Conflicts are rare because the graph is right, not because I'm lucky. My job is picking who reports to whom.

99% of engineers still run a single chat window. That's not a skill gap. That's a stack gap, and it takes one evening to close."

Chief of Engineers → PM → Coding Agents → Graph → Overnight Shipping

30-minute workshop where Lingxi walks the exact GrokBot setup running his team.

Design the graph → Build the agents → Ship overnight → Kill what breaks

Replaces the $1000 agentic engineering course you almost bought.

Watch it today.

[kurisina @kurisinaa]

She is 18, built an AI agent with live memory, and Anthropic reportedly acquired it for $2.2M.

Then at Stanford, she broke down the entire process from zero:

00:40 - building a live-memory agent with Opus 5 in under 60 minutes

18:44 - how the memory layer handles work that previously required huge engineering teams

39:42 - turning the first working agent into a $2.2M Anthropic deal

I copied the workflow, gave Opus 5 a single prompt, and had a working memory agent running about an hour later.

This is one of those talks worth saving. Watch it first.

Then use the article below to rebuild the same live-memory architecture from scratch.

[kurisina @kurisinaa]

SpaceXAI engineer:

"99% of people use GrokBot like Google.

Only 1% are running swarms of self-learning GrokBot agents.

I'm running 100+ agents in a loop. I have a Chief agent and PM agents. They manage the whole team."

GrokBot → Chief Agent → PM Agents → Agent Swarms → Self-Learning Systems

In a 30-minute workshop, a SpaceXAI engineer breaks down how to get more out of GrokBot while keeping costs low.

The interesting part is how you organize the work: who assigns tasks, who checks the results, and what the team learns before the next loop.

Assign → Execute → Review → Improve

Worth more than another $1,000 vibe-coding course.

Watch it today. Then pick one task you repeat every week and build a small agent team around it.

Get that working before you scale to 100.

[kurisina @kurisinaa]

@silentguyy66 designing for the weakest agent makes sense. curious how they catch code that passes CI but solves the wrong problem

[kurisina @kurisinaa]

@0xCarnagee the handoff between agents is what I’m curious about. who catches it when one bot gives the next one bad output?

[kurisina @kurisinaa]

@sopersone You’re the first person who wrote to me

thank you so much

[kurisina @kurisinaa]

Anthropic engineer, Boris Cherny (creator of Claude Code):

"We deleted 80% of Claude Code's system prompt and the model got smarter."

Most of those instructions were correcting behaviours Opus 5 already handles. Remove the extra guidance, and the model has more room to work.

Yet engineers keep feeding it 400-line CLAUDE.md files filled with rules written for earlier models. Every update brings new capabilities, but the old instructions never leave.

One evening of testing could show you how many still belong there.

Delete the prompt → Run it raw → Watch where it fails → Add one line back

In a 15-minute talk at YC, with no slides, Boris Cherny explains how Anthropic strips the harness down and rebuilds only where the model needs help.

Remove → Run → Observe → Rebuild

The reported result: one prompt, one dynamic workflow, 11 days. Bun's entire runtime rewritten from Zig to Rust. Over 100,000 lines, reportedly in production.

Before spending $1,000 on another AI engineering course, spend 15 minutes questioning the instructions you already use.

Watch the talk today.

Tomorrow, test your workflow without CLAUDE.md and let the failures decide what goes back.