Rzac

AI + Agents || Science · Economics · Data · Trust · Safety · Human Agency · Builder || @uber · @google · @cisco · @stanford · @ohiostate || Views are my own

[Rzac @renilzac]

Wrapped up #NvidiaGTC feeling inspired and grateful! 🚀 Congrats to the organizers. Amazing seeing #NVIDIA Blackwell and DGX SuperPOD capabilities. From groundbreaking exhibits to enlightening conversations, excited for the future of AI.
#GTC2024 #AIWoodstock

[Rzac @renilzac]

Project Value is Moving Upstream:
our real value now lies in the inputs- the deep research, sharp positioning, and original thinking that shape the system and not in the outputs.

[Zara Zhang @zarazhangrui]

This is so good

Increasingly the output of an agency looks like a folder of files for agents, instead of one-off assets

"Get paid for your mind, not your hands"

[Rzac @renilzac]

Down 29. Won by 1. The greatest comeback in Finals history just happened at the Garden. Sports is the great unifier.

[ESPN @espn]

A PLAY NEW YORKERS WILL REMEMBER FOR THE REST OF THEIR LIVES ‼️

[Rzac @renilzac]

“The reusable unit inside the loop is a skill, not a prompt.
Loops that call sharp named skills compound; loops that re-derive everything just burn money”

[Matt Van Horn @mvanhorn]

https://t.co/DM0CAuyprS

[Rzac @renilzac]

Claude writing its own harness to get more determinism

[Thariq @trq212]

https://t.co/R6exTuF7P8

[Rzac @renilzac]

And then we got Sites.
AI competition is bringing feature velocity to consumers like never before.

[Rohan Varma @TheRohanVarma]

We just launched Sites into Codex!

Software creation was always about more than writing code. Sites in Codex fundamentally gives the power of end-to-end software creation to every user, no matter their technical fluency.

These Sites are fully deployed to a URL, private to workspaces, come with authentication, can have static files, and can store dynamic data in databases.

It is in preview for business and enterprise teams and will be rolling out to all workspaces over the next day. Give it a try by typing @ Sites into Codex and ask it to build anything!

This project took a massive amount of effort across hundreds of people at OpenAI - proud that we were able to get this out and excited to see what you all build with it!

[Rzac @renilzac]

Evolution of software seems to be moving through five stages-

• Level 0: Humans do the work.
• Level 1: Humans use AI tools.
• Level 2: Humans delegate work to AI agents.
• Level 3: AI runs the workflow; humans supervise.
• Level 4: AI runs the factory; humans focus on what to build next.

Many companies believe they’re at Level 3. Most are still at Level 1 or 2.

[Alex Lieberman @businessbarista]

https://t.co/v7Uphcl9Nw

[Rzac @renilzac]

Building relevant context continuously for right agents at the right place, at the right moment in a trusted way. That’s the game we’re playing.

[Tom Blomfield @t_blom]

Imagine replacing 90% of your employees with a team of geniuses who have no idea how your company operates.

Total chaos. Nothing works.

That’s what AI feels like today.

The missing piece is extracting all the domain knowledge from people’s heads and providing that as structured context to the models.

[Rzac @renilzac]

I genuinely feel like I can think, learn, and build faster than at any other point in my career.
Ideas don't just get scribbled into a notes app and forgotten anymore. They get explored, challenged, refined, and quickly evolve into something tangible, a project, a prototype, a folder, a repo, a living thing with momentum of its own.

It's hard not to be excited by that. We're witnessing one of the biggest shifts in how ideas become reality.

The unit cost of intelligence is rapidly approaching zero, and for the first time, the unit cost of turning ideas into action is falling right alongside it.

[Jean-Michel Lemieux @jmwind]

My best ideas happen during workouts or Sunday morning long bike rides. I used to have to think hard to remember, then jot down a quick reminder and email myself a "don't forget note" then pick it up on Monday morning.

Now I talk to Siri via air pods to email @stillaai with the framework of the idea. The email references a few skills I've built: brainstorm, challenge, narrow, analyze, present options.

While I'm still riding, agents are arguing with the idea, finding flaws, exploring alternatives, building examples, mocking things up, and drafting press releases. I always ask for a polished HTML report to review.

The weird part isn't that AI can write code. That was cute. The weird part is that random thoughts now have a CI/CD pipeline.

Living in the future is still the best feeling of my entire career.

[Rzac @renilzac]

One example of why everyday computing will need a fundamental full stack redesign in how we interact with it.

[Farza 🇵🇰🇺🇸 @FarzaTV]

Watch me control my computer with just my voice. This is the future of operating systems.

No hands.

GPT-Realtime 2.0 is very, very underrated.

Demo:

[Rzac @renilzac]

If you’re deep into AI or big tech stories, Acquired’s recent Google AI episode is must-listen territory.
It’s like their deep dives on Apple or Nvidia, they cover how Google created the transformer that powers everything today yet still got surprised by OpenAI, plus all the internal drama and Gemini launch. Along with many origin stories packed in that make you think.
Thanks @AcquiredFM !

https://t.co/DRUzbhQcQb…

[Rzac @renilzac]

Everyone wants the outcome. No one wants the process. But the process is where success is made, and the process is pain.

What pain do you secretly enjoy that everyone else runs from?

[Nicolas Dessaigne @dessaigne]

My advice to founders in 2026: spend tokens, not headcount.

Record everything. Make your company queryable. Build self-improving loops.

Before long, AI won’t just help you operate your company. It will make it self improving.
Don't think AI adoption, think AI transformation.

This is the biggest shift in how startups get built since cloud computing.

[Rzac @renilzac]

If the model is doing more, the user should be doing less.

Not the other way around.

A complicated interface is not proof your model is smart. It is proof your design is not.

[Rzac @renilzac]

@brivael More important than ever that we evolve AI responsibly and thoughtfully, as it begins influencing the future good and bad ideas.

[Rzac @renilzac]

Lots of deep insights on the year ahead

[Shane Parrish @shaneparrish]

My conversation with @winstonweinberg, co-founder of @Harvey.

0:00 The List that Powers His life and Work
2:20 How to Say “No”
7:26 3 Principles for Decision-Making
8:18 How Harvey is Changing the Legal World
11:36 The Cold Email to Sam Altman
12:56 The Demo Strategy that Shocked
17:55 Advice Winston Didn't Take
19:34 The Deal that Almost Killed Harvey
21:56 How to Build Resilience to Failure
24:00 How Winston Hacks His Stress
29:36 Creating a Sense of Urgency on Your Team
31:29 Who Not to Hire
35:09 How to Screen for Resiliency in Interviews
45:28 Does AI Make a Better Lawyer?
48:54 The Future Law Firms
54:52 Why Legal Costs Aren't Going Down
00:56:48 3 Principles The Work
01:00:54 How Winston Defines Success

Listen now 👇

(Includes paid partnerships.)

[Rzac @renilzac]

The thing about building a second brain: it works best when you still use the first well.

[Corey Ganim @coreyganim]

https://t.co/ZJALhQuRsV

[Rzac @renilzac]

@garrytan This is the birth of Tribal knowledge.

Errors become judgment.
Judgment becomes principles.
Principles become culture.

Learning from ‘generating mistakes at maximum density’ is the compounding you want.

[Rzac @renilzac]

‘Jagged intelligence’ still rings true. The smarter AI gets, the more clearly I see where human intelligence is sharp and where the seams show.

[Andrej Karpathy @karpathy]

Jagged Intelligence

The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.

E.g. example from two days ago - which number is bigger, 9.11 or 9.9? Wrong.
https://t.co/dUrR6wm8GC

or failing to play tic-tac-toe: making non-sensical decisions:
https://t.co/XarwfUBtod

or another common example, failing to count, e.g. the number of times the letter "r" occurs in the word "barrier", ChatGPT-4o claims it's 2:
https://t.co/xpffK2r0pv

The same is true in other modalities. State of the art LLMs can reasonably identify thousands of species of dogs or flowers, but e.g. can't tell if two circles overlap:
https://t.co/HCXxBxosAu

Jagged Intelligence. Some things work extremely well (by human standards) while some things fail catastrophically (again by human standards), and it's not always obvious which is which, though you can develop a bit of intuition over time. Different from humans, where a lot of knowledge and problem solving capabilities are all highly correlated and improve linearly all together, from birth to adulthood.

Personally I think these are not fundamental issues. They demand more work across the stack, including not just scaling. The big one I think is the present lack of "cognitive self-knowledge", which requires more sophisticated approaches in model post-training instead of the naive "imitate human labelers and make it big" solutions that have mostly gotten us this far. For an example of what I'm talking about, see Llama 3.1 paper section on mitigating hallucinations:
https://t.co/pjuxoIOJCY

For now, this is something to be aware of, especially in production settings. Use LLMs for the tasks they are good at but be on a lookout for jagged edges, and keep a human in the loop.

[Rzac @renilzac]

The idea doesn’t have to be original.
It has to be authentic to matter.

[George from 🕹prodmgmt.world @nurijanian]

if you want to design with AI agents, these skills are amazing

- impeccable https://t.co/Wcykv4uHwT
- taste https://t.co/rThOMA1z76
- layers https://t.co/VKFrJiCoBN
- superdesign https://t.co/cibZkPsd3D

I also made a plugin based on Refactoring UI (use for polish): https://t.co/hV6iVpBgqf

find more here: https://t.co/5LJs9brXDv