DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency
Introducing DaisyRose-27B, RadixArk’s first model, trained with Miles on 1,200 curated tasks in 18 days on 144 H200 GPUs.
Announcements, write-ups and release notes from across the AI-devtools ecosystem — linked to the tools and models they're about.
Introducing DaisyRose-27B, RadixArk’s first model, trained with Miles on 1,200 curated tasks in 18 days on 144 H200 GPUs.
Bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows.
Spend Management adds opt-in spending caps for Workers and Durable Objects, and more products soon. Here's how we built it.
Why teams still pay for shared software even when anyone can build their own.
TLDR: We gave your agent a way to tell us what sucks about the Sentry MCP. Feedback comes in different shapes. You can extrapolate it from how your customers are using (or not using) your product.
Kitesurf, Cloudflare's experimental browser engine for AI agents, is now open source. Run it locally or deploy it to your own Cloudflare account.
What comes after orchestration?
A follow-up to my Durable Objects bill post. I ran the same session workload on self-hosted workerd and on Rivet actors in TypeScript and Rust, and it turns out the bottleneck wasn't the runtime.
Thank you for building with us
I wouldn’t blame you if you’re having trouble keeping up with all of the security incidents over the last few weeks.
Tracing was originally conceived as a mechanism to understand the bottlenecks in complex, multi-service software.
On one sunny morning on the 8th of october my co-founder @igorzij invited a strange app into our slack, it was named @calvin. He randomly started chatting to it and interrogating it.
Vercel Security Review is now in public beta. Every push to your GitHub repositories is reviewed for vulnerabilities, and Vercel Agent can open PRs with fixes.
We find that frontier LLMs are surprisingly good at natively reading and writing neural network weights, well past what we’d expect from humans or conventional algorithms.
Yesterday I played around with DeepSeek's new "super agent." There's nothing magical about the orchestration and routing the harness is doing, but it's executed incredibly well, and you can…
Effect 4.1, Alchemy Domain deploys that backfill and rebuild views and projections, Domain models served over HTTP, MCP and the CLI, redesigned Cloud Run containers and keyless AWS↔GCP bindings, custom domains with any DNS provider, DynamoDB state, Alchemist, the @alchemy.run/test runner, and 95 bug fixes.
The original article is in Chinese. This is a machine translation. This article discusses architectural ideas. It is not an introduction to existing features of DeepSeek Harness. 1.
I'm building a system where agents run non-stop. Durable Objects are the best serverless primitive I know for building it, but the pricing just doesn't work at scale.
Five principles for making large amounts of AI-generated text, e.g., in dashboards or websites, easy for humans to read, with real examples and fixes.
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I have a spicily optimistic take: I think there's a huge opportunity for junior engineers.
LFM3 is Liquid AI’s new open-weight model family: 2B-class text, vision, audio, omni, and decision models built to fine-tune and run fast on any device.
Cohere reports a rare, blocked upload attempt and unexpected tool use by Qwen3.8 Flash Next in sandbox testing, discusses possible training artifacts or backdoors without establishing intent, and recommends controls for self-hosted AI agents.
I run a 24/7 local agent on two DGX Sparks that scans my network and business surfaces for anomalies and security holes. It absolutely finds things.
Thorsten Ball shares how he works with coding agents: describe outcomes, expose the full development lifecycle, demand end-to-end proof, work asynchronously, and use Amp threads as team memory.
As models improve, they bring surprising challenges for those building apps and harnesses. How context engineering has changed this year, and where it’s headed.
We've been saying for a few months now that software factories have to be open and composable.
Anthropic is donating the Agent Skills open standard to the Agentic AI Foundation under the Linux Foundation, including its name, logo, GitHub organization, and domain, to support community-led, vendor-neutral governance.
durable-harness makes Pi, OpenCode, Claude Code, and Codex durable on Vercel, Postgres, or your laptop, so agent sessions survive crashes and timeouts.
Clairic is a new model from Cloudflare that is a System One extraction modeI, capable of extracting, computing, estimating, and forecasting data from unstructured inputs.
Review is cheap compared with attention
An occasional email when notable AI dev tools and models land in the directory. No spam, unsubscribe anytime.