This 1-hour Stanford lecture by Joel Peterson will teach you more about negotiation and getting what you want than most people learn in years. Bookmark it and give it an hour, no matter what.
Most of what we call "agentic AI" is a string of tiny decisions. Which tool gets called. Which document matters. Which request gets routed where. Whether a message trips a guardrail.
And we keep handing those decisions to giant models that write a paragraph of reasoning before picking option B.
That's slow, it's expensive, and at high volume it adds up fast. I've built enough integrations to know the bottleneck usually isn't intelligence. It's latency and cost on the hundreds of small calls nobody budgets for.
So I've been looking at Drex, a new small model from Nace.AI built for exactly this job. It doesn't generate text. It looks at the options you give it and returns calibrated probabilities over them in a single pass. No long reasoning trace, no token bill for thinking out loud.
The numbers they're reporting: #1 on the Decision Index 0.1, beating Jev 1.13.0. 136ms response time. Up to 32K context. Up to 1.5x faster than Jev.
Under the hood it's a diffusion architecture trained with reinforcement learning from adjusted feedback loops, which is a very different bet than scaling up another chat model.
Where I'd actually use it: agent routing and tool selection, reranking search results, document classification, compliance and policy checks, security guardrails, and risk flags. Basically any step in a pipeline where you need a fast, confident pick instead of an essay.
The fun part is they dropped it into Chess, DOOM, StarCraft, and Lemmings to show it making calls in real time. Worth a look if you want to see what 136ms decisions feel like.
They're giving 250M free tokens to the first 10,000 builders. If you run agents or routing layers, test it on your own workload and see where it beats what you've got.
nace.ai/drex
#ad #AI #AIagents #LLM
4
10
11
1,451
How To Speak Articulately (Full Course)
Most of what we call "agentic AI" is a string of tiny decisions. Which tool gets called. Which document matters. Which request gets routed where. Whether a message trips a guardrail.
And we keep handing those decisions to giant models that write a paragraph of reasoning before picking option B.
That's slow, it's expensive, and at high volume it adds up fast. I've built enough integrations to know the bottleneck usually isn't intelligence. It's latency and cost on the hundreds of small calls nobody budgets for.
So I've been looking at Drex, a new small model from Nace.AI built for exactly this job. It doesn't generate text. It looks at the options you give it and returns calibrated probabilities over them in a single pass. No long reasoning trace, no token bill for thinking out loud.
The numbers they're reporting: #1 on the Decision Index 0.1, beating Jev 1.13.0. 136ms response time. Up to 32K context. Up to 1.5x faster than Jev.
Under the hood it's a diffusion architecture trained with reinforcement learning from adjusted feedback loops, which is a very different bet than scaling up another chat model.
Where I'd actually use it: agent routing and tool selection, reranking search results, document classification, compliance and policy checks, security guardrails, and risk flags. Basically any step in a pipeline where you need a fast, confident pick instead of an essay.
The fun part is they dropped it into Chess, DOOM, StarCraft, and Lemmings to show it making calls in real time. Worth a look if you want to see what 136ms decisions feel like.
They're giving 250M free tokens to the first 10,000 builders. If you run agents or routing layers, test it on your own workload and see where it beats what you've got.
nace.ai/drex
#ad #AI #AIagents #LLM
2
3
4
874
A billionaire literally explained how the entire economy works in 42 minutes better than most $200,000 MBA programs ever could.
Most of what we call "agentic AI" is a string of tiny decisions. Which tool gets called. Which document matters. Which request gets routed where. Whether a message trips a guardrail.
And we keep handing those decisions to giant models that write a paragraph of reasoning before picking option B.
That's slow, it's expensive, and at high volume it adds up fast. I've built enough integrations to know the bottleneck usually isn't intelligence. It's latency and cost on the hundreds of small calls nobody budgets for.
So I've been looking at Drex, a new small model from Nace.AI built for exactly this job. It doesn't generate text. It looks at the options you give it and returns calibrated probabilities over them in a single pass. No long reasoning trace, no token bill for thinking out loud.
The numbers they're reporting: #1 on the Decision Index 0.1, beating Jev 1.13.0. 136ms response time. Up to 32K context. Up to 1.5x faster than Jev.
Under the hood it's a diffusion architecture trained with reinforcement learning from adjusted feedback loops, which is a very different bet than scaling up another chat model.
Where I'd actually use it: agent routing and tool selection, reranking search results, document classification, compliance and policy checks, security guardrails, and risk flags. Basically any step in a pipeline where you need a fast, confident pick instead of an essay.
The fun part is they dropped it into Chess, DOOM, StarCraft, and Lemmings to show it making calls in real time. Worth a look if you want to see what 136ms decisions feel like.
They're giving 250M free tokens to the first 10,000 builders. If you run agents or routing layers, test it on your own workload and see where it beats what you've got.
nace.ai/drex
#ad #AI #AIagents #LLM
3
4
577
Radha Tripathi retweeted
Most of what we call "agentic AI" is a string of tiny decisions. Which tool gets called. Which document matters. Which request gets routed where. Whether a message trips a guardrail.
And we keep handing those decisions to giant models that write a paragraph of reasoning before picking option B.
That's slow, it's expensive, and at high volume it adds up fast. I've built enough integrations to know the bottleneck usually isn't intelligence. It's latency and cost on the hundreds of small calls nobody budgets for.
So I've been looking at Drex, a new small model from Nace.AI built for exactly this job. It doesn't generate text. It looks at the options you give it and returns calibrated probabilities over them in a single pass. No long reasoning trace, no token bill for thinking out loud.
The numbers they're reporting: #1 on the Decision Index 0.1, beating Jev 1.13.0. 136ms response time. Up to 32K context. Up to 1.5x faster than Jev.
Under the hood it's a diffusion architecture trained with reinforcement learning from adjusted feedback loops, which is a very different bet than scaling up another chat model.
Where I'd actually use it: agent routing and tool selection, reranking search results, document classification, compliance and policy checks, security guardrails, and risk flags. Basically any step in a pipeline where you need a fast, confident pick instead of an essay.
The fun part is they dropped it into Chess, DOOM, StarCraft, and Lemmings to show it making calls in real time. Worth a look if you want to see what 136ms decisions feel like.
They're giving 250M free tokens to the first 10,000 builders. If you run agents or routing layers, test it on your own workload and see where it beats what you've got.
nace.ai/drex
#ad #AI #AIagents #LLM
12
5
11
12,096
She literally explained How to trigger dopamine in conversation and master small talk:
The part i care about isnt that it runs on a schedule..
its that i dont have to come back and finish the task myself.
the email gets answered.
the follow-up gets sent.
the meeting gets booked.
thats the difference between a reminder and actually getting the work done.
1
6
7
803
INSTEAD OF WATCHING NETFLIX TONIGHT. Spend 2 hours with this. Claude AI FULL COURSE that teaches you how to BUILD and AUTOMATE anything. The people who watch this tonight will wake up tomorrow with a new skill. Watch it and bookmark it now.
What happens when you clean up voice before sending it to speech-to-text?
Krisp just released an open benchmark testing this across:
• 265 real recordings
• 11 STT configurations
• Call centers, offices & moving cars
The result?
📉 ~73% lower word error rate across the benchmark.
And the full dataset is now public on Hugging Face.
This is especially interesting for voice agents dealing with real-world noise:
🎙️ Background conversations
⌨️ Keyboards
🚗 Traffic
❄️ AC noise
🗣️ People speaking nearby
Because in real-world environments, speech-to-text doesn't always get clean audio.
See the benchmark + results:
🔗 partner.krisp.ai/shruticodes…
1
5
9
1,367
Dr. Pratosh at IISc Bengaluru tells his students that rapid advances in AI are commoditizing intellectual labour.
He raises an unsettling question: if companies stop recruiting on campus in the next 5–10 years—even at India’s top universities—what will the purpose of a university education be ?
A must watch video for everyone in tech
I’ve been catching up on posts from Apsara Conference 2026, and one thing really stood out to me:
AI is starting to feel less like something that lives on a screen.
When people say “embodied AI,” I usually think of humanoid robots from companies like Unitree or Tesla.
But Apsara showed a much wider picture.
AI is showing up in everyday products and real-world applications, alongside the chips, compute, cloud infrastructure and AI agents making all of it possible.
My friend who attended the conference was telling me how much was happening across the floor, and honestly, Shenzhen and Yiwu seem to have sprouted wings.
What I found most interesting wasn’t just what AI can do today.
It’s seeing where it’s starting to show up next.
1
3
5
966
She literally explained How to trigger dopamine in conversation and master small talk:
I like the idea of being able to start an agent, close my laptop, and come back to the same session later.
No rebuilding the environment.
No trying to remember where the work stopped.
DigitalOcean Managed Agents keeps the runtime and session around while you bring the agent you already use Claude Code, Codex, LangGraph, or your own container.
It’s still in Public Preview, but this is the kind of infrastructure I’d actually want when working with agents for longer tasks.
4
4
883
An IIT Madras professor teaches Quantum Mechanics for free through NPTEL. 🎓
This lecture dives into the two-level atom, a foundation for quantum computing and laser physics.
No tuition. No paywall. Just 40+ lectures, a professor, and a blackboard.
Instead of watching Netflix, spend an hour learning something that can change how you think.
🔖 Save this before you lose it.
Saved 9 AI papers this week. Read 0.
Rene called me out, summarized the top 3 in one line each and put 30 minutes on my calendar for the best one.
1
3
6
1,309
This 1-hour Stanford lecture by Joel Peterson will teach you more about negotiation and getting what you want than most people learn in years. Bookmark it and give it an hour, no matter what.
I've spent 10 years wiring GitHub into everything. Never once into a video editor.
Poolday fixes that. Connect your repo and every shipped feature gets an on-brand launch video cut from your real UI and Figma components.
Your changelog just got a marketing department.
#ad
5
14
2,178
Anthropic senior engineer just released a 1-hour course on building a team of agents with loops & graphs:
• 00:27 - introduction to CLAUDE.md & Plan mode
• 11:24 - building "skills" & "hooks" for Claude agents
• 37:02 - building agents & subagents with Claude
• 52:47 - self-improving loops & graphs for Claude agents
this 1-hour watch will replace a $500 agentic engineering course
watch today, then read how to build a team of self-improving agents that work together
AI agents are only as powerful as the models they can access.
Now imagine having 150+ multimodal models available through one MCP layer.
Swap models. Test ideas. Build faster.
No painful integrations every time you want to try something new.
This is how the AI agent stack gets seriously interesting. 🔥
2
5
784
She literally explained How to trigger dopamine in conversation and master small talk:
I’ve always found CRM updates to be one of those things sales teams say they’ll keep clean… and then slowly stop doing.
Important stuff ends up in Slack, emails, calls, DMs etc.
Then someone has to piece it all together before a meeting.
That’s why Lightfield caught my attention.
Instead of asking people to keep updating the CRM, it builds the customer context from all those interactions and lets agents work from it.
4
79
466
19,820
IIT, chalkboard, zero slides
A 22-year-old at IMC gets paid $500,000 in year one for exactly one skill. Not picking winners. Seeing every trade as a full spread of outcomes, each one with its own odds attached.
Her name is Prabha Sharma. She teaches this exact skill on a bare chalkboard, for free, in a lecture almost nobody outside that room has ever watched.
Most people see a trade as a coin flip. Win or lose. Nothing else.
You already think this way about your own trades, don't you.
That's the mistake. A trade is never just win or lose. It's a spread of outcomes, each with its own odds and its own payoff attached before a single dollar moves. The formula for this is called the binomial distribution, and it's far simpler than it sounds.
Here it is:
P(X = k) = C(n,k) × p^k × (1-p)^(n-k)
n = how many bets you take
p = your real win probability on each one
k = how many of them you actually want to win
Try it yourself right now. Say you have a genuine 55% edge (p = 0.55) and you take 10 trades (n = 10). What's the actual chance you win at least 6 of them?
Plug in k = 6: P(X=6) = C(10,6) × 0.55^6 × 0.45^4 ≈ 23.8%
Add that up across every value of k from 6 to 10, and your real odds of winning at least 6 out of 10 land around 47%. Just under a coin flip, with a genuine edge in your favor.
Sit with that for a second. A 55% edge does not feel like 55% once you actually run the math over a small number of trades.
This is the exact formula that decides how much to risk on any single bet, once you actually know your edge. Sharma builds the whole thing from zero, in notation simple enough for a complete beginner to follow every step.
Wall Street pays $500,000 a year for this one skill. An IIT professor already gave the entire thing away, for free, on a blackboard nobody bothered to watch.IIT, chalkboard, zero slides
A 22-year-old at IMC gets paid $500,000 in year one for exactly one skill. Not picking winners. Seeing every trade as a full spread of outcomes, each one with its own odds attached.
Her name is Prabha Sharma. She teaches this exact skill on a bare chalkboard, for free, in a lecture almost nobody outside that room has ever watched.
Most people see a trade as a coin flip. Win or lose. Nothing else.
You already think this way about your own trades, don't you.
That's the mistake. A trade is never just win or lose. It's a spread of outcomes, each with its own odds and its own payoff attached before a single dollar moves. The formula for this is called the binomial distribution, and it's far simpler than it sounds.
Here it is:
P(X = k) = C(n,k) × p^k × (1-p)^(n-k)
n = how many bets you take
p = your real win probability on each one
k = how many of them you actually want to win
Try it yourself right now. Say you have a genuine 55% edge (p = 0.55) and you take 10 trades (n = 10). What's the actual chance you win at least 6 of them?
Plug in k = 6: P(X=6) = C(10,6) × 0.55^6 × 0.45^4 ≈ 23.8%
Add that up across every value of k from 6 to 10, and your real odds of winning at least 6 out of 10 land around 47%. Just under a coin flip, with a genuine edge in your favor.
Sit with that for a second. A 55% edge does not feel like 55% once you actually run the math over a small number of trades.
This is the exact formula that decides how much to risk on any single bet, once you actually know your edge. Sharma builds the whole thing from zero, in notation simple enough for a complete beginner to follow every step.
Wall Street pays $500,000 a year for this one skill. An IIT professor already gave the entire thing away, for free, on a blackboard nobody bothered to watch.
CRM has always been weird.
Humans do the work → then manually update the CRM about the work.
Lightfield is flipping that model.
Instead of asking reps to keep the CRM updated, it builds the customer context from calls, emails, Slack, LinkedIn, etc.
Then agents can actually work from that context.
This feels much closer to what an AI-native CRM should be.
$47M Series A led by a16z is a pretty strong signal too.
2
2
1,048
“To be successful at anything, you don’t have to be special. You just have to be what most people aren’t: consistent, determined and willing to work for it. No shortcuts.”
— Tom Brady
I’ve always found CRM updates to be one of those things sales teams say they’ll keep clean… and then slowly stop doing.
Important stuff ends up in Slack, emails, calls, DMs etc.
Then someone has to piece it all together before a meeting.
That’s why Lightfield caught my attention.
Instead of asking people to keep updating the CRM, it builds the customer context from all those interactions and lets agents work from it.
2
10
33
2,582
Radha Tripathi retweeted
This 1-hour Stanford lecture by Joel Peterson will teach you more about negotiation and getting what you want than most people learn in years. Bookmark it and give it an hour, no matter what.
7
30
107
6,814
This guy literally explained why some people become successful while other stay average.
The reason is uncomfortable. Game Theory
Watch this:
I build agentic systems for a living. So I am the last person who should be impressed by one.
Three weeks ago I added an AI employee to a Slack channel and gave him one job: run the eval suite against the new model and tell me where it actually loses.
Ran all 340 cases, diffed against last month's baseline, wrote up the six it fails and why. It was in my drive the same morning.
Then on day 19 he opened the channel on his own.
"A new checkpoint shipped six hours ago. I reran the suite on it so you are not comparing against a stale baseline."
I had not seen the release yet.
Two days later, again without me asking. Two of the tools in the stack broke on the new schema. I patched both, wrote tests and left the PR open for you.
What changed in three weeks is who starts the conversation.
The work that eats a builder's week is not building. It is rerunning the thing you already built every time something upstream moves.
Judgment stayed with me the whole time. Everything came back as a proposal, and every one of them was mine to keep or drop. What I gave up was having to remember to start.
Builders: what is the job you rerun by hand that you have never automated because it felt too small?
In partnership with Viktor
viktor.com
4
5
2,028
Google just released the best 1-hour course on Graph Engineering: from single agent to a full 24/7 system
00:00 - What Graphs are
09:16 - Build an agent
21:15 - Graph engineering explained
41:03 - Graph engineering practice
52:21 - Self improving Graphs
Free, the best thing on Graph engineering I've come across
Watch it!
4
7
1,155
IIT, chalkboard, zero slides
A 22-year-old at IMC gets paid $500,000 in year one for exactly one skill. Not picking winners. Seeing every trade as a full spread of outcomes, each one with its own odds attached.
Her name is Prabha Sharma. She teaches this exact skill on a bare chalkboard, for free, in a lecture almost nobody outside that room has ever watched.
Most people see a trade as a coin flip. Win or lose. Nothing else.
You already think this way about your own trades, don't you.
That's the mistake. A trade is never just win or lose. It's a spread of outcomes, each with its own odds and its own payoff attached before a single dollar moves. The formula for this is called the binomial distribution, and it's far simpler than it sounds.
Here it is:
P(X = k) = C(n,k) × p^k × (1-p)^(n-k)
n = how many bets you take
p = your real win probability on each one
k = how many of them you actually want to win
Try it yourself right now. Say you have a genuine 55% edge (p = 0.55) and you take 10 trades (n = 10). What's the actual chance you win at least 6 of them?
Plug in k = 6: P(X=6) = C(10,6) × 0.55^6 × 0.45^4 ≈ 23.8%
Add that up across every value of k from 6 to 10, and your real odds of winning at least 6 out of 10 land around 47%. Just under a coin flip, with a genuine edge in your favor.
Sit with that for a second. A 55% edge does not feel like 55% once you actually run the math over a small number of trades.
This is the exact formula that decides how much to risk on any single bet, once you actually know your edge. Sharma builds the whole thing from zero, in notation simple enough for a complete beginner to follow every step.
Wall Street pays $500,000 a year for this one skill. An IIT professor already gave the entire thing away, for free, on a blackboard nobody bothered to watch.
4
66
278
36,717
This 2 hour Harvard interview with Lee Kuan Yew, the man who turned Singapore from a tiny island into one of the richest nations on Earth, will teach you more about leadership, discipline, and nation-building than most business books ever will.
1
5
7
753
Hiring Tony Robbins today costs 1 million dollars for just one day.
This is a 21-minute tape recorded in his own home more than 30 years ago.
There, he explains exactly how to get anyone to say yes.
The same material for which they now charge a fortune... completely free.
It's a rare, unfiltered recording from the time when he still didn't charge millionaires just for being in the same room.
When someone says no, they give two excuses:
“I don’t have time” or “I don’t have money”
Neither is true.
The real reason is that they still don’t believe it’s worth it.
It’s not a money problem. It’s a state problem.
Tony teaches something he calls “attack and confess”
Instead of arguing the objection, you confess your own:
“I had the chance to go six months ago and I didn’t until two months ago.
I can’t even imagine the time I wasted”
The room goes silent. No one argues.
Then he gets the person on the “yes train”
Each little yes adds to the next.
Until saying no at the end feels harder than saying yes.
When it’s time to sign, the person has already said yes five times without realizing it.
Just 21 minutes.
There you learn the two only moves that people pay 1 million a day for:
how to read anyone’s state… and how to shift it.
Most people spend years guessing in sales.
He wrote it on a flipchart in his living room in less than half an hour.
A seat in that room cost $125 dollars.
Today it costs $1 million dollars a day to sit in front of him.
The tape is free right now.
And the answer is in this video.
2
6
15
1,918