OpenSRE watches your CI/CD pipeline, works out why the build broke and starts fixing it, and it's fully open source. If your Friday keeps getting eaten by red builds, this is worth a look: github.com/Tracer-Cloud/open…
19
Ideogram 4.6 does native 4K, multi-turn editing and precise local edits through an API, from $0.008 a picture. It tops the multi-step editing benchmark. If you brief a designer for every product variant, give this ten minutes of your Friday. ideogram.ai/models/4.6/
17
Alibaba just open-sourced its code review tool: rule checks catch the obvious bugs, AI writes the line-by-line comments. They've used it internally for two years. Cheaper bugs are caught bugs. github.com/alibaba/open-code… #AI #DevOps
20
A frozen decision model now generates coherent 512px images in 78 milliseconds, up to 122 times faster than today's image models. Real-time visuals inside your product just stopped needing a data centre. arxiv.org/abs/2611.24738
17
Frontier models asked to design a network without gradient descent all chose a 1970s idea instead: about 1,800 times fewer parameters and up to 17 times faster on GPU. Sometimes progress is subtraction, not scale. arxiv.org/abs/2611.22055 #AI #ML
24
AI agent bills just got cheaper. Tencent's Hy4 flash runs at about a twentieth of the cost of the full model, and it's free in WorkBuddy until 30 November. Worth a look before you budget next year's AI spend. tencent.com/tencent-launches…
34
Anyone can hide instructions in a PDF your agent reads. SPILLWAY v2 stops that: it verifies tool calls against typed fields instead of trusting the model, and blocked every injection in testing. On PyPI today. arxiv.org/abs/2611.07222 #AI #Security
1
1
43
Anyone with an AI agent can now reverse-engineer your app or website. rea handles binaries, Electron apps and sites, and it picked up 19k GitHub stars in a week. If your product is your moat, start treating it like one. github.com/morluto/rea #AI #Security
1
60
Google, Microsoft and Mozilla just shipped a skill pack for coding agents: current web best practices, compatibility data, and a scanner for legacy code. If your AI-generated pages feel stuck in 2023, this is why. github.com/GoogleChrome/mode…
38
IBM built a format that stops prompt injection: data and instructions stay separate, and attacks fell from nearly 100% to under 0.1%. If your team pastes documents into chatbots, read it. huggingface.co/blog/ibm-gran…
AI agents can now pay at checkout. ChatGPT Wallet gives them stand-in card numbers so they never see yours, with a fingerprint check each time. Machine buying arrived before shops were ready. community.openai.com/t/intro…
1
32
Gemini 4 Hydrogen now runs Google's AI Mode, and it cites more links per answer, including niche pages rankings would bury. Answer one question thoroughly and your door just got wider. blog.google/innovation-and-a… #SEO #AI
44
llama.cpp just hit 1.0.0. The local inference engine under half the AI tools you use cut its old API, added speech, and promised proper versioning. When a foundation layer matures, everything on top gets safer. github.com/ggml-org/llama.cp…
1
1
61
Repairing a broken reasoning model used to mean millions of tokens of agent edits. A new paper does it in four forward passes for about 90 cents of GPU time, and beat the agent's result. That changes the maths on running small models. arxiv.org/abs/2611.14210 #AI
51
Your best model is wasted writing plans for cheap ones to follow. shadcn's improve skill flips it: the strong model audits your codebase and writes markdown specs, the cheap executor builds them. Less spend, same quality. github.com/shadcn/improve
34
NVIDIA donated OpenShell, its agent-security runtime, to CNCF: sandboxed tools and policy checks applied as your agents run. When agent security becomes open infrastructure, skipping it is now negligence. blogs.nvidia.com/blog/open-s… #AI #DevOps
n8n just open-sourced a personal agent that signs into websites, runs jobs in parallel and remembers months of context — but every send, pay or delete needs your approval first. Autonomy with the brakes still attached. github.com/n8n-io/nod #AI #Automation
1
2
66
Meta built a model for the boring half of AI agents: tool calls, routing, extraction. Muse Ember costs $0.10 per million tokens, matching the industry floor. The commodity layer is set, so your edge is what you build on top. research.meta.ai/blog/muse-e…
1
45
NVIDIA researchers compiled an LLM's reasoning into a CPU-executable cache. Same answers on hard tasks, 16 to 291 times cheaper per call. If your agents keep re-solving the same problems, that is your margin hiding in plain sight. arxiv.org/abs/2611.09891
31
If your safety scores come from multiple-choice tests, they're flattered. Models commit to an answer letter before refusal ever kicks in, so the benchmark measures knowledge, not safety. Worth knowing before you pick a model on them. arxiv.org/abs/2611.08639
25
