☕️ Microsoft study finds AI coding agents lift pull requests 24%

Hi there, this is your daily ☕️ Devshot.

In today's Devshot:

💻 Microsoft study finds AI coding agents lift pull requests 24%

🍎 Apple buys observability startup SigScalr

☁️ Amazon's new Graviton5 chip beats Intel but trails AMD

🤖 Moonshot AI's Kimi K2.7 now works in Copilot

Plus: 🎁 12 other news you might like, 🧰 6 tools, and 📚 5 papers.

💻 Microsoft study finds AI coding agents lift pull requests 24% LINK

A large-scale Microsoft study found that developers using command-line AI coding agents merged about 24% more pull requests, the code changes submitted for review, than they otherwise would have during a four-month period.

The rollout covered Anthropic's Claude Code and GitHub Copilot CLI, and the gain only appeared with regular use: engineers using the tools five or more days a week saw a lift above 50%, versus roughly 15% for three-day users.

Copilot CLI users saw about 2.2 times the pull-request lift of Claude Code users in comparable weeks, though researchers said that came from Microsoft's own setting and should not count as a general ranking of the two tools.

🍎 Apple buys observability startup SigScalr LINK

Apple has acquired SigScalr, the company behind SigLens, an application monitoring tool that tracks and debugs how large numbers of interrelated apps run over time, according to a filing with the European Union.

SigScalr was a small company with between two and ten employees, founded in May 2021, which raised $1.76 million in a 2024 pre-seed round and released SigLens as open source on February 20, 2024.

Neither company has said what happens next, but SigScalr's website is down while its code stays on GitHub, and SigLens could be folded into Xcode, echoing Apple's June 2026 purchase of Swift tool Play.

☁️ Amazon's new Graviton5 chip beats Intel but trails AMD LINK

Amazon's new Graviton5 chip, tested in the M9g EC2 instances, outperformed Intel's Xeon 6 Granite Rapids in benchmarks but fell behind AMD's 5th Generation EPYC "Turin" processors in the cloud comparison.

The m9g.4xlarge came in cheapest at $0.78272 per hour, undercutting Intel's m8i.4xlarge at $0.84672 and AMD's m8a.4xlarge at $0.97376, with all four instances carrying 64GB of memory.

Graviton5 shifts from Neoverse-V2 to Neoverse-V3 cores and DDR5-5600 to DDR5-8800 memory, while Intel's M8i uses physical cores plus HT siblings, giving it half the physical cores of the AMD and Graviton offerings.

🤖 Moonshot AI's Kimi K2.7 now works in Copilot LINK

Moonshot AI's Kimi K2.7-Code has become the first open-weight model added to GitHub Copilot's model picker, sitting next to closed options from OpenAI, Anthropic, Google, and Microsoft when GitHub rolled it out on July 1, 2026.

The integration routes prompts and completions through Microsoft Azure infrastructure rather than Moonshot's own servers, which answers enterprises' biggest worry about sending source code to a Beijing-based lab at inference time.

GitHub launched it for Copilot Pro, Pro+, and Max subscribers across VS Code, Visual Studio, JetBrains, Xcode, and more, then added Business and Enterprise plans on July 7, where administrators must switch the Kimi K2.7 Code policy on.

Other news you might like

• Kubernetes Dashboard to Headlamp: A Step-by-Step GuideLINK

• Building an end-to-end reliability testing strategy with Grafana CloudLINK

• How to Build More Resilient Local-First Applications With AT Protocol InfrastructureLINK

• How to shrink the token budget without shrinking the teamLINK

• How to Be a Good Open Source MaintainerLINK

• Microsoft makes Windows Subsystem for Linux more stable with architecture tweakLINK

• The git history command deserves more attentionLINK

• In-N-Out Animation using sibling-index()LINK

• Flickr’s optimistic committingLINK

• How DoorDash Built an AI Shopping Assistant That Doesn’t Rely on the LLM AloneLINK

• How Microsoft Ships AI Agents at Enterprise ScaleLINK

🧰 Trending tools

Loomal: lets developers monetize APIs and tools instantly by accepting USDC payments from AI agents, with no API keys or invoices required.LINK

Marked QL: a macOS Quick Look extension that previews Markdown files in Finder with syntax highlighting, math, diagrams, and tables without opening an editor.LINK

RunInfra: a deployment platform that automates AI model quantization, benchmarking, and serving with optimized vLLM configurations and pay-per-token pricing.LINK

CodeMote: lets you control your local dev environment remotely from iPhone, managing terminals, diffs, and agent actions without routing code through third-party clouds.LINK

Nixmac: detects your existing Mac customizations and converts them into reproducible, version-controlled Nix configuration files automatically.LINK

ServiceBeard: converts customer emails into tracked issues via IMAP/SMTP, syncing directly with GitHub, GitLab, or Linear without per-seat helpdesk costs.LINK

📚 Trending papers & reports

Climate pattern search tools now let scientists visually inspect and verify whether AI-generated data groupings reflect real weather phenomena or just modeling artifacts, closing a critical gap in Earth science discovery.LINK

Video generation speed gets a boost from a new technique that skips redundant calculations more accurately, making AI video tools faster without needing lengthy setup or special training data beforehand.LINK

Routing inside multi-agent systems now has a formal economic theory proving that a central coordinator can optimally allocate tasks across many specialized models, guaranteeing efficient, stable outcomes.LINK

Robot bin-picking systems now work reliably on cheap hardware, achieving up to 600 picks per hour with 96 to 99% success, making automated warehouse picking far more affordable for smaller manufacturers.LINK

Combining vision-language models by merging their internal components rather than retraining from scratch lets developers blend complementary strengths cheaply, without the conflicting updates that normally derail such attempts.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Devshot!