Apple partners with Google to power Siri with Gemini; Samsung targets 800M AI devices by 2026.

AI Digest — March 8, 2026

Your daily briefing on AI developments and open-source project releases.

🔖 Project Releases

Claude Code — v2.1.71 (March 7, 2026)

The latest Claude Code release landed just yesterday with a solid mix of new features and stability fixes.

New features: The standout addition is the /loop command, which lets you run prompts or slash commands on a recurring interval — for example, /loop 5m check the deploy will re-run every five minutes. Alongside this, cron scheduling tools were added for recurring prompts within a session, and voice mode got a rebindable push-to-talk keybinding (default: space) via keybindings.json. The bash auto-approval allowlist was also expanded with commands like fmt, comm, cmp, seq, and others.

Key fixes: A nasty stdin freeze in long-running sessions where keystrokes would stop being processed has been resolved. Voice mode users no longer suffer a 5–8 second startup freeze caused by CoreAudio initialization. The startup UI freeze triggered by simultaneous OAuth token refreshes is also gone. Other fixes address /fork conversations sharing the same plan file, oversized images breaking context, and plugin installations being lost when running multiple Claude Code instances.

Performance: Startup time is faster thanks to deferred native image processor loading, and bridge session reconnection after laptop wake now completes in seconds rather than hanging.

Claude Code is also now available as a coding agent for all GitHub Copilot tiers — Copilot Pro, Business, and Enterprise — as of late February 2026.

Full release notes →

Beads — v0.59.0 (March 6, 2026)

Steve Yegge’s Beads — the Dolt-powered, version-controlled SQL database designed as a “memory upgrade” for coding agents — shipped a fresh release two days ago.

What’s new: This release focuses on infrastructure cleanup and reliability. The daemon infrastructure has been fully removed, and the doctor command now supports warning suppression via config. Fresh clone state detection on the Dolt server has been improved, and the Nix flake was modernized for nixpkgs-25.11. A notable addition is sync.git-remote surfacing in the fresh clone error flow, making it easier to troubleshoot setup issues.

Bug fixes: Batched SQL IN-clause queries were corrected to prevent full table scans — a meaningful performance improvement for larger databases. The circuit breaker system was refined with a reduced 5-second cooldown and active TCP health probes. Various configuration routing issues were also addressed.

Pre-compiled binaries are available for Linux, macOS (Intel & Apple Silicon), Windows (AMD64 & ARM64), Android/Termux, and FreeBSD.

Full release notes →

OpenSpec — v1.2.0 (February 23, 2026)

OpenSpec, the spec-driven development (SDD) framework by Fission AI, last released v1.2.0 about two weeks ago. No new release this week.

Highlights from v1.2.0: The biggest addition is Profiles, letting you choose between a core install (4 essential workflows) or a custom setup where you pick exactly which skills to include — managed via openspec config profile. The new Propose workflow lets you create a complete change proposal with design, specs, and tasks from a single request, eliminating the need to run new then ff separately.

Two new AI tools gained support: Pi (pi.dev) and AWS Kiro IDE, both with prompt and skill generation. The init command now auto-detects existing tool directories (.claude/, .cursor/, etc.) and pre-selects them. The sync command was improved to prune deselected workflows, keeping your project directory clean.

OpenSpec now supports 21 AI tools in total, including Claude Code, Cursor, Windsurf, Gemini CLI, GitHub Copilot, Amazon Q, and many more.

Full release notes →

📰 Top AI News

Apple’s Reimagined Siri Debuts with iOS 26.4

Source: Crescendo AI

Apple has officially announced that a completely reimagined, AI-powered Siri will debut this month alongside iOS 26.4. In a notable strategic move, Apple is partnering with Google to leverage its 1.2 trillion parameter Gemini AI model, running on Apple’s Private Cloud Compute infrastructure to maintain strict privacy standards. This represents a significant shift in Apple’s AI strategy and could reshape how hundreds of millions of iPhone users interact with their devices daily.

Gartner: Global AI Spending to Hit $2.5 Trillion in 2026

Source: The Motley Fool

Gartner projects worldwide AI spending will increase 44% this year to reach $2.5 trillion. The surge is being driven by cloud computing infrastructure, advanced chip demand, and enterprise software adoption. This figure underscores that AI investment is accelerating rather than cooling off, with infrastructure providers and chip manufacturers positioned as primary beneficiaries.

DeepSeek Prepares V4 Model Launch

Source: LLM Stats

DeepSeek is preparing to release its highly anticipated V4 large language model — its first major launch since January 2025’s V3/R1 waves that disrupted the industry. Given how much DeepSeek’s previous releases shook up pricing and performance benchmarks, the AI community is watching this one closely. Details on architecture and capabilities are still emerging.

GLM-5: Zhipu AI’s 744B Parameter Open Model

Source: LLM Stats

Zhipu AI released GLM-5, a 744 billion parameter mixture-of-experts model with 44B active parameters, a 200K context window, and an impressive 77.8% score on SWE-bench Verified. Released under the MIT license, it’s one of the largest and most capable open-weight models available. This continues the trend of Chinese AI labs pushing the frontier of open-source LLMs.

NVIDIA Partners with Alpamayo for Autonomous Vehicle AI

Source: Crescendo AI

NVIDIA announced a strategic partnership with Alpamayo, a specialist in AI-driven simulation, to streamline autonomous vehicle development. The collaboration leverages NVIDIA’s DRIVE Orin and Thor platforms alongside Alpamayo’s high-fidelity digital twin technology. The partnership signals continued investment in simulation-first approaches to self-driving, where virtual testing can dramatically reduce the time and cost of real-world validation.

Samsung Targets 800 Million Gemini AI Devices

Source: Crescendo AI

Samsung Electronics announced plans to double its footprint of mobile devices equipped with Google’s Gemini AI to 800 million units by the end of 2026. The push extends generative AI features beyond flagships into mid-tier and budget smartphones and tablets, democratizing access to on-device AI for a far broader consumer base.

China’s Five-Year Plan Centers Quantum and AI

Source: The Quantum Insider

China has placed quantum technology and artificial intelligence at the center of its next five-year national development strategy. The plan signals a concerted push to secure global leadership in emerging computing and advanced science, with dedicated funding streams and talent programs. This has implications for the global AI race and technology competition between major powers.

Apple Launches M5 Pro and M5 Max MacBook Pro

Source: Apple Newsroom

Apple introduced the MacBook Pro with M5 Pro and M5 Max chips, delivering up to 4x performance gains over the previous generation. Of particular interest to the AI community: new Neural Accelerators in the GPU enable professionals to run advanced LLMs on-device while maintaining exceptional battery life. This makes local AI inference increasingly practical for developers and researchers.

🧭 Key Takeaways

Claude Code’s /loop command and cron scheduling mark a shift toward autonomous, recurring agent workflows — your coding assistant can now monitor and act on its own schedule.
The open-weight LLM race intensifies with Zhipu AI’s 744B parameter GLM-5 under MIT license, and DeepSeek’s V4 on the horizon. Competition from Chinese labs continues to push capability and accessibility forward.
Apple is going all-in on AI partnerships — Gemini-powered Siri in iOS 26.4 and Neural Accelerator-equipped M5 chips signal that on-device and cloud AI are converging in Apple’s ecosystem.
AI spending is accelerating, not plateauing — Gartner’s $2.5 trillion projection (up 44% YoY) confirms that enterprise adoption is still in growth phase, not consolidation.
The agent tooling ecosystem is maturing — between Beads’ version-controlled agent memory, OpenSpec’s 21-tool SDD framework, and new cron/loop features in Claude Code, the infrastructure for autonomous coding agents is getting serious.

Generated on March 8, 2026 by Claude