“We are seeing that the main thing that determines whether an Agents succeeds or fails is the quality of the context you give it. Most agent failures are not model failures anyemore, they are context failures.”
www.philschmid.de/context-engi...
“People interacting with an application UI has always been the weakest link in getting work done. A manual and necessary evil. But with agents that can act on data to drive workflow, the idea that work can only be done by people via an application UI is blown to shreds.”
What do people’s feed game look like these days? Start Discover, switch to Following? Something else entirely?
Doubt I would have stumbled across this book if not for @wired.com but I’m ripping through it
www.wired.com/story/plaint...
Enjoyed The Economist on Tyler Cowen - “the man who wants to know everything”
Tyler Cowen, the man who wants to know everything
economist.com/1843/2025/02...
from The Economist
“The future of AI competition will be about 'power dominance' - do you have access to enough electricity to power the datacenters used for increasingly large-scale training runs”
eta-publications.lbl.gov/sites/defaul...
“Product reviews are for winning trust”
Recommended watch for product leaders and senior ICs
youtu.be/W1coI_d9MsQ?...
Rolling out a generative AI POC, frantically trying to collect user metrics to justify additional investment, checks survey results:
"What's the prompt driving this?"
🙃
Starting "Bubble and the End of Stagnation" this morning.
"A bubble is therefore not simply a collective delusion but an expression of a future that is radically different from now."
Of all the books I've read on the Twitter story, up to and including the Musk takeover, this one was my favorite.
Well deserved
You're going to love it!
I think the risk can be mitigated if you're hyper focused on a specific, achievable use case, but that's more of the "low hanging fruit" approach and does little to answer the broader questions around agentic implementations
Given your experience, are you betting on generalizable agent architectures or highly specialized implementations?
Great thread on agents highlighting a few limitations:
1) Most companies are rushing in without clear goals
2) We lack good ways to measure successful agent implementations
3) It's hard to get lots of good training data to match real world situations (for eval specifically)
I sometimes wonder to what degree it's necessary/beneficial for non-technologists to have intuition around AI fundamentals, capabilities, and limitations.
Dario was cool, but I'm enjoying listening to the Amanda Askell portion of the Lex Fridman podcast. Interesting perspective on Claude's personality straight from Anthropic's "prompt whisperer"
https://youtu.be/ugvHCXCOmm4?si=93THamoDxiZOfEZx
Croissant seems pretty good for cross posting so far
Just made my way through this this morning and interesting discussion - sticking with both for now but Bluesky definitely has the vibes for now
If it starts making car noises, hop off. It's broken
What a chart.
www.economist.com/interactive/...
Recently: Wow, recent LLMs can sort of play chess! They fall apart after the early game, but they can do something! Amazing!"
dynomight.net/chess/
Something weird with LLMs and chess...
"Before September 2023: Wow, recent LLMs can sort of play chess! They fall apart after the early game, but they can do something! Amazing!
September-October 2023: Wow! LLMs can now play chess at an advanced amateur level! Amazing!
(Year of silence.)
I'm not the biggest Lex Fridman fan, but will definitely listen to the 5 hour interview with Dario Amodei
youtu.be/ugvHCXCOmm4?...
Character Limit
Moonbound
The Real North Korea
Stalin’s War
Those are my last 4 from those genres!
That feeling of fear/pride when my product team is doing so much user discovery that I can’t even sit in quietly and collect my own insights on it all anymore 😂
What do you like?
Hi, PM leader at Walmart 👋
Wonder if the question is: does automation need to save time to be valuable, or is there inherent value in making processes more predictable, consistent, and partition-able?
The coffee maker example - interaction is still required, but you don’t have to actively monitor the brewing process.
