DevTools Updates and Consensus Experiments #72

July 9, 2026 — 2 min read

Today's Letter

Vercel said Flags SDK and Vercel Flags now evaluate multiple feature flags in bulk about 10x faster.
The change reduces microtask queue overhead and creates fewer promises, with larger gains as the number of flags increases.
The recommended pattern is `await evaluate([flagA, flagB])` instead of `Promise.all([flagA(), flagB()])` when reading multiple flags.
The same `evaluate` API also accepts an object form, allowing named-key results through object destructuring.
`precompute()` inherits the performance improvement automatically, according to the changelog.
Vercel published the update on July 8, 2026 and said users need the latest `flags` and `@flags-sdk/vercel` packages to use it.

Source: vercel.com

Hugging Face introduced a deep-link integration that opens supported model pages directly into Amazon SageMaker Studio workflows.
Two new actions, Customize on SageMaker AI and Deploy on SageMaker AI, carry the selected model context into Studio without a second model search.
New Studio environments created through this flow are auto-provisioned with pre-configured permissions for customization, training, notebooks, and endpoint deployment.
AWS also creates and attaches the managed policy AmazonSageMakerModelCustomizationCoreAccess for serverless customization jobs including SFT, DPO, RLVR, and RLAIF paths.
Studio now shows GPU quota visibility for G5 and G6 instance types inside the instance picker, with direct links to Service Quotas when more capacity is needed.
The post says the flow is available now for supported models and is intended to reduce the setup steps previously needed across the AWS console, IAM configuration, domain creation, and quota checks.

Source: huggingface.co

Cloudflare introduced Meerkat, an experimental global consensus service for control-plane state.
It targets services that read and update shared state across 330+ data centers.
Meerkat is built on QuePaxa, a consensus algorithm published by EPFL researchers in 2023.
Cloudflare says QuePaxa differs from Raft by allowing all replicas to accept writes.
The design avoids leader-dependent write stalls and timeout-driven failover delays.
Cloudflare says past incidents in leader-based systems motivated the new approach.
Initial use is small internal control-plane state such as database leadership and placement data.
Cloudflare describes this as the first industrial QuePaxa deployment at global scale.
Meerkat remains in development and is planned as an internal-only system for now.

Source: blog.cloudflare.com

Primary-source retrieval failed because Google News returned a 503 block page instead of the article body.
Secondary coverage describes GitLost as an attack path that could expose private repository data through GitHub's AI agent.
The original linked article points to Noma Security, but the source text was not available in this step.
Under the current pipeline rules, the core claims cannot be treated as verified without a reachable primary source.
This item is not suitable for publication until the original post or another reliable primary source is fetched.

Source: noma.security

Jocoletter curates AI, software, and product trends for developers and builders.

#Cloudflare #HuggingFace #NomaSecurity #Vercel