leanxbt

i don't feel shame, i'm not capable of it

[leanxbt @leanxbt]

Google senior engineer just dropped a free 8-minute lesson on building AI agentic systems.

This is the clearest explanation of multi-agent AI systems and loops you'll find on the internet.

People still paying 500$ for agentic courses, while Google makes it for free.

Agent Studio -> Loops -> Managed Agents -> Antigravity -> ADK 2.0 - thats the stack

Watch it, then read the full guide on loop engineering below.

[leanxbt @leanxbt]

https://t.co/mrIJXQy3Lc

[leanxbt @leanxbt]

Anthropic just dropped a 33-page blueprint for building effective AI agents. Zero theory, just production architecture patterns used by Claude, Coinbase, Stripe, and Intercom.

Every system follows one cycle:

Perceive -> Decide -> Act -> Evaluate -> Repeat.

Here are the 5 core patterns to know:

Single Agent: One model in a loop. Solves 80% of problems, don't over-engineer it.

Sequential: Step-by-step handoffs. Predictable and easy to audit.

Parallel: Tasks split across agents at once, then merged. Built for speed.

Hierarchical: A supervisor agent managing a team of specialists.

Evaluator-Optimizer: A 2-agent loop (generator + critic) refining quality over 2-4 cycles.

The Bottom Line: Multi-agent architectures outperform single models by 90.2% on complex tasks. Just match your complexity to the value.

Read the manual, then check out the "Loop engineering" article below.

[leanxbt @leanxbt]

https://t.co/mrIJXQy3Lc

[leanxbt @leanxbt]

@0xSlyth gm broski

[leanxbt @leanxbt]

https://t.co/mrIJXQy3Lc

[leanxbt @leanxbt]

Anthropic Head of Product:

"At Anthropic, 100% of our engineers have already switched to Fable 5. It's our most powerful model.

Tasks that Opus 4.6 was doing in a night, Fable 5 does in an hour. Capabilities are incredible."

in 40-minute session, Anthropic's Head of Product share the capabilities of Fable 5 and his workflow.

Watch today, then read the article below.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

This 12-page PDF completely changed how I build agentic systems:

Discover -> Hand off -> Verify -> Persist -> Schedule

Here is the 5-step blueprint:

Discovery: The loop reads CI, issues, and commits to find what's worth fixing.

Handoff: Each finding gets an isolated git worktree so parallel agents never collide.

Verification: A second agent - built to assume the code is broken, reviews the work.

Persistence: Results land on disk, never in a temporary context window.

Scheduling: Automation fires the entire process on a timer, making it a true loop.

The key insight: An agent grading its own work always praises it. You need that second agent to say "no."

Read it now, then explore the article below.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

Aravind Srinivas, CEO of Perplexity ($20B), on 20VC:

"You could hire five people worth 200K, or you could hire one Jeff and pay them a million. What would you do?"

He answers it himself. One Jeff. Every time.

The world he is describing pays the frontier and stops paying everyone else. Software is collapsing into smaller teams using more AIs. Companies that used to need ten thousand engineers will run with forty. The other nine thousand nine hundred sixty are not getting laid off in one event. They are quietly not being hired.

So the question is not whether you keep your seat. It is whether anyone in the room would pick five of you over one of someone else.

If the answer is no, you have until they finish doing the math.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

The man who wrote the 52-page memo that got Sam Altman fired finally broke 18 months of silence.

Ilya Sutskever. Co-founder of OpenAI. Co-creator of AlexNet, sequence-to-sequence models, and GPT.

The chief scientist whose document, based largely on testimony from Mira Murati, convinced the board to remove Altman in November 2023.

He confirmed in court this year he still owns a $7B stake in the company he tried to fire its CEO from. Now he runs Safe Superintelligence Inc. at a $32B valuation with no product, no revenue, and no plans to release anything.

Altman gave his side. Murati gave hers. Toner gave hers. The man who actually pulled the trigger just gave his.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

Here is the Jensen Huang post rewritten in the sharp, minimalist style of the second post:

The man who built the ultimate factory for artificial intelligence just declared that what took 50 years to build will now be quadrupled in twelve months.

Jensen Huang. CEO of Nvidia.

He’s talking about the chips powering every AI data center on Earth.

We are reinventing the computer for the first time in 63 years. The cost to replace it? Tens of trillions.

The bottleneck was never the models. It’s the factories.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

The woman who briefly ran OpenAI during the board coup just broke 18 months of silence to tell the other side of the story.

Mira Murati. Former CTO of OpenAI. Interim CEO for three days in November 2023.

Now CEO of Thinking Machines Lab at a $12B valuation, in talks to raise at $50B.

Altman gave his side last month. She just gave hers.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

The man who just won the Nobel Prize in Physics stood up at his own university and explained the exact invention that won it.

Geoffrey Hinton. 2024 Nobel for the Boltzmann machine. Four months later he walked Toronto through how he figured it out in the 1980s.

Most people read the news. He explained the work himself. The video is free.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

The CEO of OpenAI told the inside story of the board coup that almost ended the company.

Sam Altman. Fired and rehired in five days. Then shipped o1, Sora, and a $500B valuation.

In 47 minutes he told YC how he got OpenAI together and why scaling laws were heresy in 2017.

Everyone has an opinion about him. He is the only one who was in the room.

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

https://t.co/AICAtsOVzD

[leanxbt @leanxbt]

The Pope of the Catholic Church broke 200 years of tradition to share a stage with a 33-year-old Canadian atheist.

The reason: Chris Olah is the co-founder of Anthropic and the leading expert on what is actually happening inside Claude.

The Vatican invited him to help unveil the first encyclical in history written about AI.

Most tech founders chase Senate hearings. This one got invited to the Synod Hall.

[leanxbt @leanxbt]

Daniela Amodei. Co-founder and president of Anthropic. In 2020 she walked out of OpenAI together with her brother and five engineers.

At a table at Stanford she explained how a literature major now runs the AI company that just filed for IPO at a $965B valuation.

Most people think you need to be technical to build a frontier AI lab.

She isn't.

[leanxbt @leanxbt]

Daniela Amodei. Co-founder and president of Anthropic. In 2020 she walked out of OpenAI together with her brother and five engineers.

At a table at Stanford she explained how a literature major now runs the AI company that just filed for IPO at a $965B valuation.

Most people think you need to be technical to build a frontier AI lab.

She isn't.

[h100envy @h100envy]

https://t.co/kV0BtewFnj

[leanxbt @leanxbt]

A Berkeley PhD student gave a lecture on model-based RL in August 2017.

She is now co-founder of the lab that controls Figure's humanoid robots and just raised at $5B.

Her name is Chelsea Finn. The recording has fewer views than a random TikTok.

[leanxbt @leanxbt]

The man who built the Slack desktop app now leads Claude Cowork at Anthropic

In 54 minutes with Matt Turck he explained how Cowork was assembled in 10 days, mostly by other Claude Code instances writing it themselves, and why Anthropic thinks your local computer is the most undervalued execution surface in AI

Most people are still building wrappers on top of Claude. The team inside Anthropic is shipping the product that makes wrappers obsolete

Bookmark and watch this. Then read the complete article below

[leanxbt @leanxbt]

@papa_couch @karpathy @levelsio karpathy and levelsio on the same list is accurate, the other four are doing a lot of carrying though

[leanxbt @leanxbt]

@sopersone bayes to jane street interview questions in 51 pages is either a masterpiece or a surface-level overview, curious which

[leanxbt @leanxbt]

@sopersone 24 minutes is doing a lot of work that most $300 courses spread across 6 hours of padding

[leanxbt @leanxbt]

@monokern this conference was a year ago, curious why it's surfacing now but the content is still valid