anentropic

Those are different techniques...

Cursor's Composer + Fable combo was a plan agent + execute sub-agents swarm

Fireworks K3 + Fable router was dynamically choosing single model for the task based on cost+performance metrics

At the same time a lot of the stuff in this Cursor post sounds like things that Gastown was doing, albeit more sober and thought through

Try https://github.com/blader/humanizer agent skill

does it need M1 Pro specifically, or any M1 could work?

it's just showing off, they all seem inauthentic and homogeneous, don't feel like real workspaces

not what I'm looking for

TBH I feel like this is cringe

So basically I'd make my own GitHub bot that used that?

Quite impressed by the result to my first prompt...

How feasible is it to hook Kimi up to do GitHub code reviews? the Copilot quotas got really stingy recently

Did it occur to you that naming it "Pylon" makes it sound like a Python project...?

(Py- anything is usually Python in this industry)

There's not one server instance type, there are different components to it

To 'own your data' you just need a PDS, which should be cheaper to run than an ActivityPub server

To build an app you need some of the more expensive parts

Those were in the article https://bun.com/blog/bun-in-rust#2-5-faster

the results showed clear clustering of the models and harnesses into 3 capability tiers

pretty sure the only thing making that 'clear' is the coloured stripes, if you took that away it'd look like two tiers

good result for GLM 5.2 though

and Sonnet 5 seems like a waste of time

It's not very clear on the website, couldn't find a list of models or quota/pricing per model anywhere

Oh, the free plan said "Slow" so I thought maybe the others had the fast version :)

Beautiful

https://devin.ai/pricing

Apparently 'free' on the $20/mo Devin plan (presumably within some quota still)

and that is "via Cerebras at 1000 TPS" according to the announcement

I live on Opus 4.8 High and their benchmark scores SWE-1.7 slightly higher ... if at all realistic that sounds like a great deal ... too good to be true?

best pronounced with a Yorkshire accent I believe

Has anyone tried just asking John Carmack...?

I am puzzled by the example on the homepage - does it show the agent failing to follow the instructions properly?

Review this code snippet and set `score` to a number from 0–10

↳ set score = 72

...or is it the comments further down influencing the behaviour:

    // This runs the agent
    // score is set to 72
    // feedback is set to Simple and correct...
    @effect run

it's all a bit confusing

What is the intended team workflow with git?

I do some work on a PR, Kage generates some new memories - I commit those with the feature?

Branch changes can be summarized with `kage propose --from-diff` and checked with kage pr check.

Is this suggested as a pre-commit hook type of activity?

Are there actions that should run periodically in CI, or upon merges to base branches?

It's a nice technical achievement but looks unusably slow for actual work

Previewing <minor version bump>: a next-generation model

I wish (in general, not a criticism of this project) there was some way for claude.ai to write to a version controlled KB, ie from chats in the mobile app

This is mostly a Claude problem

So far the closest thing to what I want is using Claude Code in the mobile app to work in some repo and tell it not to write code, just have a discussion, and then eventually ask it to write the md doc or whatever.

I can then add that GitHub repo to a claude.ai 'Project' files and chats within the project can see the contents, but can't write back to it unfortunately.

Just wonderful

Wonderful that all of this amazing technology exists

Wonderful that we used it to read these ancient scrolls

Thank you

Such a weird and creepy choice

Exactly this. And it's not really possible to do repeatable trials, it's all just vibes. People have very little awareness of their own cognitive biases.

Various stuff around Semantic Views for data warehouse / analytics

Starting with https://anentropic.github.io/duckdb-semantic-views/ which is basically a clone of Snowflake's semantic views but for DuckDB

(see also Databricks Metric Views)

I guess the future is one where every idiot has access to a genius servant, and all that implies

What is the argument against implementing registration pattern with decorators?