broke boy

I find the AI lectures everyone will quote next month.

Agents, models, inference & AI startups.

[broke boy @0xfuckpoverty]

https://t.co/TtMbG4JSwO

[broke boy @0xfuckpoverty]

Most AI agents never leave the demo.

OpenAI’s 46-minute Build Hour shows the production stack that separates a chatbot from a system that can actually work.

• 4:50 - Agent Builder: design multi-step workflows instead of one-shot prompts

• 21:27 - ChatKit: turn an agent workflow into a real product experience

• 24:53 - Evals: grade traces, find failures, and improve agents over time

• 35:17 - How companies like Ramp, HubSpot, and Rippling use agentic workflows in production

The real agent stack is:

Build → Deploy → Evaluate → Improve

Most builders stop after “Build.”

That is why most AI agents look impressive for five minutes - then fail in production.

Production agents need workflows, tools, guardrails, state, and evals.

Bookmark this before shipping another chatbot disguised as an agent.

[broke boy @0xfuckpoverty]

Most AI agents never leave the demo.

OpenAI’s 46-minute Build Hour shows the production stack that separates a chatbot from a system that can actually work.

• 4:50 - Agent Builder: design multi-step workflows instead of one-shot prompts

• 21:27 - ChatKit: turn an agent workflow into a real product experience

• 24:53 - Evals: grade traces, find failures, and improve agents over time

• 35:17 - How companies like Ramp, HubSpot, and Rippling use agentic workflows in production

The real agent stack is:

Build → Deploy → Evaluate → Improve

Most builders stop after “Build.”

That is why most AI agents look impressive for five minutes - then fail in production.

Production agents need workflows, tools, guardrails, state, and evals.

Bookmark this before shipping another chatbot disguised as an agent.

[broke boy @0xfuckpoverty]

https://t.co/TtMbG4JSwO

[broke boy @0xfuckpoverty]

Waymo Co-CEO Dmitri Dolgov:

“The demo is only 1% of the work.”

Waymo now runs 500,000 autonomous trips every week — with 17× fewer serious-injury crashes than human drivers.

Most AI founders optimize the demo.

The companies that win optimize the other 99%.

• 6:52 - Why turning an AI demo into a real product took Waymo 15 years

• 36:41 - Every physical AI company needs a simulator before it needs scale

• 43:22 - Evals are not a safety feature. They are the competitive advantage.

• 46:09 - How Waymo became 17× safer than human drivers

The next AI winners will not be the teams with the best demo.

They will be the teams with the most reliable system behind it.

[broke boy @0xfuckpoverty]

@silentguyy66 bookmarked for tonight

[broke boy @0xfuckpoverty]

Waymo Co-CEO Dmitri Dolgov:

“The demo is only 1% of the work.”

Waymo now runs 500,000 autonomous trips every week — with 17× fewer serious-injury crashes than human drivers.

Most AI founders optimize the demo.

The companies that win optimize the other 99%.

• 6:52 - Why turning an AI demo into a real product took Waymo 15 years

• 36:41 - Every physical AI company needs a simulator before it needs scale

• 43:22 - Evals are not a safety feature. They are the competitive advantage.

• 46:09 - How Waymo became 17× safer than human drivers

The next AI winners will not be the teams with the best demo.

They will be the teams with the most reliable system behind it.

[unicode @unicodef1wn]

https://t.co/HTmdoJcobB

[broke boy @0xfuckpoverty]

@beamnxw oi boy bookmarked

[broke boy @0xfuckpoverty]

@silentguyy66 Persistent context is the foundation of agent memory

[broke boy @0xfuckpoverty]

Sam Altman just gave a 39-minute talk every AI founder should watch.

His core message:

AI is raising the ceiling on what a tiny team can build.

• 4:17 - Why startups matter more now than at any point in the last decade

• 6:57 - The coming golden age of ambitious startups

• 9:30 - Why small teams win during major technology shifts

• 11:46 - Building OpenAI when almost nobody believed AGI was possible

• 30:41 - How fast AI models will improve - and why founders should build for that future

• 36:06 - What the best version of an AI future should actually look like

Most founders use AI to ship faster.

The real opportunity is to take on problems that used to require an entire company.

AI is not just a productivity tool.

It is a capability unlock.

Bookmark this for the day you decide what is worth building.

[broke boy @0xfuckpoverty]

@spectnfa always welcome spect🤝

[broke boy @0xfuckpoverty]

Sam Altman just gave a 39-minute talk every AI founder should watch.

His core message:

AI is raising the ceiling on what a tiny team can build.

• 4:17 - Why startups matter more now than at any point in the last decade

• 6:57 - The coming golden age of ambitious startups

• 9:30 - Why small teams win during major technology shifts

• 11:46 - Building OpenAI when almost nobody believed AGI was possible

• 30:41 - How fast AI models will improve - and why founders should build for that future

• 36:06 - What the best version of an AI future should actually look like

Most founders use AI to ship faster.

The real opportunity is to take on problems that used to require an entire company.

AI is not just a productivity tool.

It is a capability unlock.

Bookmark this for the day you decide what is worth building.

[broke boy @0xfuckpoverty]

https://t.co/TtMbG4JSwO

[broke boy @0xfuckpoverty]

@silentguyy66 glad to know it

[broke boy @0xfuckpoverty]

@beamnxw already grabbed

[broke boy @0xfuckpoverty]

@unicodef1wn thanks god for unicode in my recommendations

[broke boy @0xfuckpoverty]

@spectnfa 11 steps to become profi

[broke boy @0xfuckpoverty]

@distortgeekin prompting course should be useful for many peeps

[broke boy @0xfuckpoverty]

@hitu_monke probably everyday lol

[broke boy @0xfuckpoverty]

Alexandr Wang, leading Meta Superintelligence Labs, calls agentic looping an “astronomical opportunity.”

Most AI startups are still building products that answer once - then forget everything.

That gap will get brutal.

• 11:27 - Personal superintelligence will be built around each individual, not one generic assistant.

• 16:36 - AI only changes the world when intelligence becomes cheap enough for everyone.

• 26:51 - The biggest untapped opportunity: systems that let models keep working toward an outcome.

The next AI winners will not have the best prompt.

They will have the best loop around the model.

One-shot AI is already becoming a commodity.

[broke boy @0xfuckpoverty]

https://t.co/TtMbG4JSwO

[broke boy @0xfuckpoverty]

Alexandr Wang, leading Meta Superintelligence Labs, calls agentic looping an “astronomical opportunity.”

Most AI startups are still building products that answer once - then forget everything.

That gap will get brutal.

• 11:27 - Personal superintelligence will be built around each individual, not one generic assistant.

• 16:36 - AI only changes the world when intelligence becomes cheap enough for everyone.

• 26:51 - The biggest untapped opportunity: systems that let models keep working toward an outcome.

The next AI winners will not have the best prompt.

They will have the best loop around the model.

One-shot AI is already becoming a commodity.

[broke boy @0xfuckpoverty]

@mentalium_ 1 hour of pleasure

[broke boy @0xfuckpoverty]

@beamnxw lots of gems on arxiv