@ClaudeDevs Isn't this graph engineering where the model draws the graph? Someone/something is creating a deterministic path that chains tools together?
@cu30rry_ The last couple of days I've been implementing the @poteto verification skills in a few projects and really starting to appreciate it's thoroughness; already caught quite a few bugs.
@brankopetric00 Karpathy made the flip side of this point earlier this year: building menugen, the hardest part by far wasn't the code, it was the DevOps around it. Still the post I keep going back to: https://t.co/YSRJCHZ9DR
I've been testing that leadership theory on small agent teams. A lead agent, two specialists, a real two-app architecture to document.
The lead's first miss was a classic new-manager one: it never sent work back. When something looked off, it fixed it itself.
@poteto From the talk, about a minute in: "there are so many parallels with, you know, management skills and how to, like, manage agents" https://t.co/tc4rhI1lVd
I have a theory, not uniquely my own. @poteto mentions it as well. Those with leadership experience (formal, or informal) manage multiple agents more efficiently than someone who doesn't.
As soon as you hit your target you start your fast skimming the rest of the article, if nothing catches your attention, move on to the next article; most articles have the key points early.
How to keep up (with any fast moving topic): give yourself a target for how many important details you want to learn from reading an article on a topic. My target is 2.
Read the article, as you're reading you'll naturally say to yourself, "Self, that's [interesting|important|crazy]", or something similar. That counts as 1.
https://t.co/eYjlQpGZIT
"Microsoft says the outage was not caused by a cyberattack but by overwhelmed servers responsible for establishing secure connections"
If it wasn't a "cyberattack". Then what caused such a heavy load to fubar the entire m365 infra?
Asking for a friend.
BREAKING: Microsoft services are beginning to come back online after a massive global outage left millions of users without access for nearly 18 hours.
The disruption affected businesses, schools and government agencies in at least 74 countries, including the United States.
Major services impacted included Outlook email, Microsoft Teams and Copilot.
Microsoft says the outage was not caused by a cyberattack but by overwhelmed servers responsible for establishing secure connections, preventing them from processing requests.
Download the FOX News app for the latest details: https://t.co/BAyV31y5kz
@t3dotcodes for the curious: shift+cmd+n is the default for a new thread that skips the project picker; cmd+n may ask which project first. https://t.co/ijx7sMr8Ot
@GergelyOrosz Isn't coding what enables all the other capabilities?
i.e. "I want to generate a bunch of images" => code/script to generate the images.
The normal non-swEng user doesn't care it wrote code; Don't they just care about getting their images with the least amount of token burn?
Ways for agent to agent communication:
t3 code - has internal messaging between threads
claude code - sendMessage tool can be used to have multiple session's talk to each other
Subagents - native pretty much everywhere these days
I'm sure there's others...
@ClaudeDevs Embedded sites inside the chat is the part I didn't expect. Curious how auth works when the embedded app needs the user's own session rather than the connector's.
1st mistake I made in my stack: I <cough>claude</cough> rewrote a two-line upstream principle into twenty because the file next to it was long. If the upstream is tiny, yours should be too. A principle that costs context every time the mode reads it is working against itself.