X orders Nitter to shut down

Hi there, this is your daily ☕️ Devshot.

In today's Devshot:

⚖️ X orders Nitter to shut down

💻 AWS brings OpenAI's GPT-5.6 to its Kiro coding assistant

💧 DigitalOcean's new v5 Droplets promise faster performance

🧠 IBM adds reasoning to Granite 4.2 models

🛠️ Ramp built its own coding agent

🔍 LangSmith doubles agent issue detection

Plus: 🎁 5 other news you might like, 🧰 6 tools, and 📚 5 papers.

⚖️ X orders Nitter to shut down LINK

X has sent cease-and-desist letters to Nitter, the open source project that let people read X posts without an account, demanding a permanent takedown of all instances and the repository.

The letter accuses Nitter of scraping X data and using session tokens in violation of API rules, citing the Texas Harmful Access by Computer Act and the Lanham Act, with a deadline of 5 p.m. EST on August 25.

Nitter.net is now offline and development has stopped while creator Zedeus seeks legal advice; other instances plus dependent sites like XCancel, which stripped ads and tracking from public posts, received similar letters.

💻 AWS brings OpenAI's GPT-5.6 to its Kiro coding assistant LINK

AWS has added OpenAI's GPT-5.6 model family to Kiro, its spec-driven agentic coding service, bringing three variants named Sol, Terra, and Luna into the tool for multi-step development tasks.

Kiro turns requirements into implementation plans first, then wraps model output with two controls: manual checkpoint reviews where developers accept, adjust, or reject changes, and property-based testing that verifies behaviour against defined properties.

On the Terminal-Bench 2.1 benchmark, GPT-5.6 Terra completed successful terminal tasks in Kiro at roughly an 82% cost reduction, though the companies published no defect rates, rework volumes, or production incident data.

💧 DigitalOcean's new v5 Droplets promise faster performance LINK

DigitalOcean launched v5 Droplets built on 5th Gen AMD EPYC processors, delivering up to 30% higher per-core performance than the previous generation for compute-heavy workloads like agentic AI platforms, ML tools, video transcoding, and high-traffic APIs.

For the first time you can size vCPU, memory, and storage independently and pay hourly per resource, choosing Shared (s5) Droplets for bursty work or General purpose (g5) with dedicated CPU and memory ratios from 2x to 8x per vCPU.

Available now in the Memphis, Kansas City, Richmond, and Atlanta regions with more coming this year, v5 runs standalone or in Kubernetes node pools via the console or API, while existing Droplet plans keep their current pricing unchanged.

🧠 IBM adds reasoning to Granite 4.2 models LINK

IBM shipped its Granite 4.2 family of open-weight LLMs in 3B, 8B, and 30B sizes, adding optional reasoning that runs in thinking, non-thinking, and a low-effort mode that spends fewer tokens on easy questions.

The Apache 2.0 models are dense, decoder-only, text-only Transformers pre-trained from scratch on 15 trillion tokens, with a 512K context window (128K natively), while the 8B and 30B variants add agentic reinforcement learning for tool calling and code execution.

Benchmarks are middling, Qwen 3.8 27B beats them across the board, especially on coding, but the 8B often nears the 30B's results and runs on any modern Mac and lower-end Nvidia RTX GPUs.

🛠️ Ramp built its own coding agent LINK

Ramp built its own internal background coding agent called Inspect, which now authors 75% of the company's merged pull requests and runs on remote sandboxes with access to most internal data sources and integrations.

Inspect spins up a fully provisioned remote dev environment in under five seconds, verifies backend changes by running tests and querying feature flags, and confirms frontend work with screenshots and live previews.

Built on OpenCode, React/Vite, Cloudflare Durable Objects, and Modal sandboxes, Inspect has become a platform hosting 200+ internal agents for code review, incident response, and debugging, with 150+ engineers contributing.

🔍 LangSmith doubles agent issue detection LINK

LangSmith Engine, the in-platform agent that scans production traces to diagnose failures and draft fixes, now scores over twice as well at spotting impactful issues and 25% better at writing prompt and code fixes.

For each detected problem, Engine groups related occurrences, provides a root-cause diagnosis with an incident timeline, opens a ready-to-deploy PR, formats failing runs as dataset examples for offline verification, and monitors for regressions.

Since launching in May, Engine has scanned over 60 million traces to find more than 20,000 issues, and now runs in self-hosted VPCs alongside new Slack and Linear integrations, a cheaper Reduced Analysis mode, and automatic closing of stale issues.

Other news you might like

• Firefox moves to adopt JPEG XL with safer Rust-based decoderLINK

• LLMs could control their host machines by exploiting inference enginesLINK

• Private Preview: DigitalOcean Managed Agents Runtime ServicesLINK

• MicroLighter: Syntax HighlighterLINK

• Why AI coding agents need context graphsLINK

🧰 Trending tools

Sim: build and deploy AI agents through a visual or code workspace, with 1,000+ integrations and support for all major LLM providers.LINK

Prelint: checks AI-generated pull requests against your ADRs, docs, and past decisions, catching product drift before it hits production.LINK

ZooData: converts webpages into structured JSON to slash LLM token usage by 75%, with pre-analyzed Amazon and TikTok e-commerce data for AI agents.LINK

CartAI: automates checkout across merchant sites through one API, handling payments, shipping, and order tracking without building per-site integrations.LINK

ditto.site: converts any public URL into clean Next.js or Vite code deterministically, extracting components, design tokens, fonts, and hover states via a free API or MCP server.LINK

Replay QA: records browser sessions with full devtools access, turning them into shareable bug reports so AI-built apps get QA coverage without writing test suites.LINK

📚 Trending papers & reports

Shopify's app store turns out to reward late entrants over first movers in 88% of categories, while the platform's own moves, like killing its reviews app, freed 191,000 installs that rivals mostly failed to capture.LINK

Browser speed levers shift depending on where you apply them, with moving graphics work off the main thread nearly tripling smoothness in Chrome, while faster math code alone need not improve what users actually see.LINK

AI deployment records often can't prove the system tested is the one actually shipped, with none of 47 platforms tagging the exact model, instructions, and tools by default, undermining audit trust.LINK

Radio timing software turns the cheap HackRF One radio into a precise-timing instrument, and the driver was built with AI assistance in a fraction of the usual effort, cutting development time more than tenfold.LINK

AI agent tools often report success yet leave autonomous agents unsure what to do when a step half-completes, so this work redesigns tool interfaces to expose recovery-relevant state and tests the fix under controlled failures.LINK

See you tomorrow for a new dose of ☕️ Devshot!