Monday 13 July 2026 | Join Free | Upgrade

Hi there, this is your daily ☕️ DevShot.

In today's DevShot:

🖱️ Cloudflare's new tool watches your mouse to catch bots

💻 Meta's new coding AI undercuts rivals by 75 percent

🤖 OpenAI's GPT-5.6 trained its own smaller model

🪞 Anthropic's new Claude Reflect shows exactly how you use AI

Plus: 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.

📰 Top News

🖱️ Cloudflare's new tool watches your mouse to catch bots LINK
  • Cloudflare has launched Precursor, a bot detection system that replaces annoying CAPTCHA puzzles by continuously watching how visitors move their mouse, type, scroll, and interact with a page throughout an entire browsing session.
  • The system studies signals like mouse movement, scrolling rhythm, typing cadence, clipboard activity, and page visibility, but Cloudflare says it does not record the actual keys pressed, only the timing and rhythm.
  • Website operators enable Precursor from the Cloudflare dashboard, which injects a JavaScript bundle that collects interaction signals and sends them for analysis, affecting bot scores and challenge decisions as suspicious behavior follows the session.
💻 Meta's new coding AI undercuts rivals by 75 percent LINK
  • Meta released Muse Spark 1.1 on Thursday, a multimodal AI model built for agentic coding that takes on rivals from OpenAI and Anthropic while promising much lower costs to win over users.
  • The model handles multistep reasoning, manages digital workflows, fixes bugs, and assists with large code migrations, and Meta will charge $1.25 per million input tokens and $4.25 per million output tokens.
  • CEO Mark Zuckerberg marked the launch by posting on X for the first time in three years, calling Spark "a strong agentic and coding model at a very low price" and promising "more to come soon."
🤖 OpenAI's GPT-5.6 trained its own smaller model LINK
  • OpenAI says its new flagship model, GPT-5.6 Sol, post-trained the smaller model Luna on its own, optimizing it for specific skills and behaviors after Luna's initial pre-training with only an under-specified prompt from a researcher.
  • Through the Codex platform, Sol was told to find the right training configurations, pick suitable GPUs, launch the training script, and check everything was running, work that senior researchers previously handled as a team.
  • GPT-5.6 Sol scores 16.2 points higher than GPT-5.5 on OpenAI's aggregated RSI index, and researchers using it saw average daily token output per person more than double the previous peak set by GPT-5.5.
🪞 Anthropic's new Claude Reflect shows exactly how you use AI LINK
  • Anthropic added a built-in browser to Claude Code that lets the AI read, click, and type on external websites, letting the coding agent interact with web apps and documentation from start to finish inside one app.
  • The browser runs inside Claude Code's existing workflow, so the agent can fetch live documentation, test web UIs, and fill forms without leaving the coding session, using the same tools it already uses to preview local apps.
  • This is part of a broader push, alongside Claude Cowork expanding to mobile and web, to make Claude an agent that acts on the real world through clicks and typing, not just a chat assistant.

📚 Tutorials & Advice

> How to build a Claude Code connector secured with WorkOS AuthKit: Explains how to add real login security to a Claude Code tool server by letting AuthKit handle token issuance and consent, so developers skip building their own login system.
> Article: Removing a Hidden Round Trip from a Multi-Region AWS API: Adopting a signing method valid across multiple AWS regions eliminated a pre-flight region-lookup request, cutting startup latency and enabling clean cross-region failover.
> Operating OpenTelemetry at scale with OpAMP: OpAMP (Open Agent Management Protocol) lets a central backend remotely configure, update, and monitor thousands of data-collection agents consistently, with safe rollback when a change breaks.
> How to Encrypt Terraform State Files: Encrypt Terraform state files both in transit and at rest to protect the passwords and infrastructure details they store in plain text.
> ReSharper 2026.2 is out with a feature developers have been begging for: The ReSharper extension for VS Code now includes full debugging, letting developers set breakpoints and inspect variables without leaving their editor.

🔗 Other News & Articles

  • Ethereum Foundation Shuts Down Protocol Support Team After Half-Decade Run LINK
  • Modernizing the Meta Ads Service With an Open-Source Kernel Scheduler LINK
  • A tiny London startup built a CUDA compiler that reportedly beats AMD's own tools on AMD hardware LINK
  • Cloudflare Identifies Race Condition in hyper’s HTTP/1 Implementation LINK
  • German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German LINK
  • OpenAI, Microsoft & Anthropic agree on who runs the agent. They disagree on what you can take back. LINK
  • sqlite-utils 4.1 LINK

📄 Papers & Resources

> Robot motion planning gets faster and simpler with a new single-stage training method that generates good action sequences quickly, without the costly two-step teacher-student process older approaches require. LINK
> Faster video generation now matches or beats existing tools on quality, by teaching a slow video-making model to produce results in far fewer steps, cutting the time needed for real-time use. LINK
> Factory robots doing precise contact work get significantly safer and more reliable after a new fine-tuning method cuts dangerous force spikes, reducing force readings above 60 N by 46 times. LINK
> Self-driving car sensors work better when a radar and camera are combined and share a continuous map of the scene, rather than treating object detection and surroundings as separate tasks. LINK
> Text-to-video generation gets 45% cheaper to run using a new architecture that replaces the standard, computationally heavy approach, making high-resolution AI video creation far more practical. LINK

🧰 Latest Tools

Opper AI: lets developers access 300+ AI models through one API key, with automatic routing, EU-based inference, and OpenAI SDK compatibility. LINK
TryCase: gives AI coding agents disposable Linux desktops to run, test, and verify apps, returning screenshots, recordings, and logs as proof. LINK
PMB: gives AI coding assistants persistent project memory via local SQLite, storing decisions, goals, and lessons learned across sessions without cloud dependencies. LINK
Cloud World Model: simulate multi-cloud architectures across AWS, GCP, Azure, OCI, and DigitalOcean to predict cost, performance, and resilience before provisioning real infrastructure. LINK
Aura: Agents + Git + Intent Open Source: a Git-native IDE that tracks AI agent code changes at function level and verifies task completion before committing. LINK
ClinePass: an open-source VS Code agent that autonomously handles coding, terminal commands, and browser tasks to complete development work end-to-end. LINK

Not subscribed to ☕️ DevShot yet? Subscribe for free

Keep reading