YC's Fall Request for Startups dropped one week before the application deadline.
"Multiplayer AI" and "Self-Maintaining APIs" are two sides of the same problem: keeping multiple distributed agents up to date, within and across organizations.
@svpino The worst part is that models are giving you the illusion of thinking for you. It's easy to fall for it, and you need to watch yourself constantly.
Outdated / stale text files are becoming a new class of bugs in the agentic world.
Frontier models have gotten really good at following hundreds of instructions at once and finding the most relevant pieces of information.
Even one out-of-date sentence in an ocean of your .md files can alter your agent's trajectory substantially, and you are no longer able to read it all.
Make sure your agents invest time to keep docs / memory / skills / specs current. It will save you hours of frustration.
They are your compilers now.
@tobi yes, but the problem is how it's done, not if it's done, at least for coding tasks.
Unless you have a full test suite to test against, you just count on an agent to do it the "right" way.
Today, weโre introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set.
[schema] makes an LLM think like a physicist. ๐งต
Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.
This is a 17-place jump from Kimi-k2.6 (#18 -> #1).
In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.
The full model weights will be released by July 27.
Congrats to the @Kimi_Moonshot team on this major milestone!
The great switch from SaaS to internal tool stack accelerates.
Make sure your agents know where they are (your company context) and point them to your databases.
Build a self-improvement engine of top.
Divide and conquer with Fable until it's in the subscription.
I just found out that deep nested subagents work really well with Fable 5 for coding tasks.
At each level Fable divides a problem into smaller sub-problems until it decides problems are well defined and small enough for single context window. Then run the subagents with the right model and pass the necessary context + success gate.
Finally it recursively compiles the complete solution.
With the right prompt there may be even more to it
I'm officially resurfacing with my life in San Francisco.
Unofficially, I got my EB1 Green Card, already been here for a couple of weeks and sorted out life - lease, DL, insurances.
Now it's time to build.