Hi there, this is your daily ☕️ Devshot.
In today's Devshot:
💧 Anthropic launches Claude Haiku 5.5
🎨 Developer builds free Adobe clones in Rust
🐧 Google AI flags 500 Linux kernel CVEs
🧰 AWS open-sources macOS AI agent sandbox
🧠 Meta's CRAM compresses Linux memory in RAM
🔄 Percona builds proxy for Valkey Cluster
Plus: 🎁 6 more stories you might like, 🧰 6 tools, and 📚 5 papers.
💧 Anthropic launches Claude Haiku 5.5 LINK
- Anthropic released Claude Haiku 5.5, its fastest small model yet, aimed at high-volume work like summaries, compactions, database queries, and classification, and priced roughly 75% below Haiku 4.5 to run on average.
- It's the first Haiku-class model with an adjustable effort setting, letting you trade cost against intelligence, and it delivers its best value on prompts up to 100,000 tokens, which covered around 90% of requests to the previous Haiku.
- Haiku 5.5 is available now on AWS, Google Cloud, and Azure, the Python and TypeScript SDKs add beta computer use and browser use, and Anthropic halved Sonnet 5.5's cache-read price to cut most agentic work by about 20%.
🎨 Developer builds free Adobe clones in Rust LINK
- A Reddit user has released PhotoCraft, LightCraft, ImageCraft, FilmCraft and other free, open-source clones of Adobe's most-used creative apps, all built in Rust with Claude and posted on GitHub despite their original post being removed.
- The suite aims to copy Adobe's familiar layouts and workflows into one unified package, and is built agent-native, exposing its tools and menus to AI over a JSON control channel and an MCP server so models can create art directly.
- Commenters warn the vibecoded code, hundreds of thousands of lines from a solo developer, poses privacy and malicious-code risks that one person cannot realistically verify, echoing a Sonar survey where 96% of developers said they don't fully trust AI-generated code.
🐧 Google AI flags 500 Linux kernel CVEs LINK
- Google's Sashiko, an open-source agentic reviewer for Linux kernel patches powered by Gemini and other LLMs, has been cited on nearly 500 kernel CVEs in its first year of operation.
- Since launching under a year ago, the tool has completed over 191,000 patch reviews across 99 mailing lists, running tens of millions of autonomous Git lookups along the way.
- Its reports have drawn more than 7,600 replies from over 1,100 kernel developers; Google is now building a local terminal review mode and a persistent bug database.
🧰 AWS open-sources macOS AI agent sandbox LINK
- AWS has open-sourced Strands Box, a sandbox that confines autonomous AI agents using OS-level isolation plus contextual, deterministic policy enforcement that the agent can't talk its way around.
- Box pairs the Dogwood Local Engine's temporal awareness with Strands Shell and Monty for Python interpreters, checking tool calls against prior activity, so an agent could post to Slack but no more than three times every ten minutes.
- Available on GitHub now for macOS only, with Linux in development and Windows unscheduled; it supports any agent harness, but AWS stresses developers still decide what access to grant and where human review is needed.
🧠 Meta's CRAM compresses Linux memory in RAM LINK
- Meta engineers led by Gregory Price built CRAM, a Linux memory-compression scheme that keeps compressed data in RAM instead of routing it through swap like ZRAM or zswap, claiming up to 452x faster reads.
- CRAM exposes compressed memory as a private NUMA node rather than a block device, letting Linux keep normal migration and ballooning semantics; in read-only tests it hit 489 million ops/sec against ZRAM's 1.1 million, running at DRAM speed.
- Enabling writes drops the lead to 5.4x at 20% writes, since data must page fault and migrate back before modification, and CRAM still can't reliably track logical RAM capacity, an unsolved problem its "Chicken Bit" only halts cascading failures for.
🔄 Percona builds proxy for Valkey Cluster LINK
- Percona has built Valkey-proxy, an open source proxy that lets applications written for a single Redis instance connect to a distributed Valkey Cluster without being rewritten to handle multiple nodes, routing, and command differences.
- The proxy sits between the application and the cluster and understands the full Valkey protocol, unlike alternatives such as Envoy, which Percona says can't support everything Valkey does; it targets self-hosted and on-premises operators.
- Approved in late September to join the Valkey project, the code should be fully public by the end of October, with a release candidate in December and general availability by early 2027; Freshworks is an early design partner.
Other news you might like
- Microsoft just released a new operating system and it is not Windows LINK
- Grab Redesigns Counter Service Storage for 50% Lower P99 Latency LINK
- ClickHouse Agents: Claude-powered agentic analytics, now in public beta LINK
- Raspberry Pi desktop returns to PCs and Macs after years of waiting LINK
- FEX 2610 Tagged LINK
- Pulumi Google Cloud Provider Version 10.0.0 LINK
🧰 Trending tools
Dograh: open-source voice agent platform with a visual flow builder, telephony, human transfer, and self-hosted QA monitoring across 30-plus model integrations LINK
tiun: an AI-native backend bundling authentication, payments, customer database, and analytics so teams ship and scale without stitching separate services together. LINK
Jev: adds fast, structured AI-driven decision-making to software automation, letting developers build reliable workflow logic without manually writing rules LINK
Switch: open-source workspace connecting people and AI agents across Slack, Teams, Discord, and Mattermost, preserving context during handoffs without migration or vendor lock-in. LINK
akta.pro: private company data API with event tracking, news, and entity resolution, helping finance and GTM teams source deals and streamline outreach LINK
HarnessRouter: open-source API connecting 20+ agent harnesses like Codex and Claude Code through one interface, self-hosted via Docker with your own keys and state. LINK
📚 Trending papers & reports
City-map views of the Linux kernel turn its 40 million lines of code into an interactive 3D cityscape, letting teams visually track how files and directories grow and change over time, including the new Rust additions. LINK
Coding agents still need human software engineers, the paper argues, because AI tools generate plausible but unverifiable code and lack the persistent domain knowledge, design judgment, and process oversight that only trained engineers provide. LINK
AI agent grading has reliability gaps, as audits of two popular benchmarks for testing software agents found the automated checkers wrongly passed 21 tasks that actually saved incorrect data, overstating how well agents really perform. LINK
Doc2Spec turns plain-English coding rules into formal, machine-checkable specifications automatically, catching 142 previously unknown rule violations in real software and cutting the costly manual work of writing verification rules by hand. LINK
Reusable code building blocks let AI coding agents assemble entire software repositories far better, lifting the top model's delivery score by ~61% and raising all 13 tested models by roughly 0.15 on average. LINK
See you tomorrow for a new dose of ☕️ Devshot!