1Claude Code gets local scheduled tasks.

Claude Code desktop now supports local scheduled tasks — set up recurring jobs that run automatically as long as your machine is awake. Think: daily code audits, periodic test runs, or scheduled PR sweeps without leaving your editor. Simple idea, but it removes the friction of remembering to run routine dev workflows. (13,547 likes | 1,022 RTs) Read more →

OpenAI Codex lands on Windows. Full native app with a sandboxed agent environment and PowerShell support for Windows developer workflows. If you've been stuck on the web version, this closes the gap. (4,544 likes | 566 RTs) Read more →

GPT-5.4 agentic prompting guide updated: OpenAI published new patterns for building reliable agents — covering tool use, structured outputs, verification loops, and long-running workflows. If you're building agents on the OpenAI API, this is the reference doc to bookmark. (860 likes | 60 RTs) Read more →

Claude community meetups — fully funded: Anthropic is bankrolling Claude meetups worldwide — funding, swag, and monthly API credits for organizers. If you've been wanting a local AI dev community, the barrier just dropped to zero. (11,310 likes | 824 RTs) Read more →

2Claude Memory hits the free tier.

Anthropic just made persistent memory available to all free-plan users — Claude remembers your preferences, projects, and context across conversations. You can also import saved memories and export them anytime. This was the single biggest gap between free and paid tiers; closing it signals Anthropic is competing on stickiness, not just capability. (38,414 likes | 2,711 RTs) Read more →

3Agentic code review ships in Claude Code.

A team of AI agents now runs deep reviews on every PR in Claude Code. Anthropic built it internally first — engineer code output is up 200% this year, and human reviews had become the bottleneck. Early users report it catches real bugs that experienced reviewers miss. This isn't linting; it's structural, logic-level review. (5,374 likes | 328 RTs) Read more →

Claude Code adoption across enterprise: A roundup shows how Ramp, Rakuten, Brex, Wiz, Shopify, and Spotify are all using Claude Code in production engineering workflows. The pattern: teams are replacing ad-hoc Copilot usage with structured agentic coding. (1,534 likes | 127 RTs) Read more →

Dario Amodei issues a statement: Anthropic's CEO published a statement — details at the link. (5,124 likes | 690 RTs) Read more →

AI in the Iran conflict — and Anthropic's legal fight: MIT Tech Review covers how AI models are being used in the Iran conflict theater, plus an escalating legal battle involving Anthropic suing the US government. The intersection of frontier AI and geopolitics is getting very real. Read more →

Autoresearch: Sparks of Recursive Self-Improvement: Latent Space covers new work on AI systems that can recursively improve their own research capabilities. It's a small step toward self-improving AI — not AGI, but the kind of incremental progress that compounds. Worth tracking. Read more →

Qwen 3.5 drops in 9B and 4B: Qwen3.5-9B and Qwen3.5-4B are both multimodal (image-text-to-text) models now trending on HuggingFace. The 9B model has 1.22M downloads already; the 4B variant at 560K. Open-weight multimodal at these sizes makes local deployment practical for real applications. (680 likes | 329 likes) 9B → 4B →

NVIDIA's AI inference at planetary scale: Latent Space interviews Nader Khalil (Brev) and Kyle Kranen (Dynamo) ahead of GTC on how NVIDIA is thinking about agent inference infrastructure — running AI engineers at scale, not just training them. If you're building inference-heavy products, this is your pre-GTC homework. Read more →

Pokémon Go's mapping data powers delivery robots: Niantic's decade of AR mapping data is being repurposed to give delivery robots centimeter-accurate world models. The irony: millions of players catching Pikachu were unknowingly building the navigational backbone for autonomous last-mile delivery. Read more →

Persistent Memory vs. Context Window: When Claude "remembers" something across conversations, that's persistent memory — stored facts about you that get injected into future sessions. This is fundamentally different from the context window, which is the temporary working memory within a single conversation. Context windows reset every chat; persistent memory survives across them. The tradeoff: memory is selective (the system decides what's worth storing), while context is complete but temporary. Understanding this distinction helps you work with AI more effectively — put critical instructions in the current conversation, and let memory handle preferences and background.

• "AI Engineer will be the last job": Latent Space reflects on the jobs debate. Link
• Energy intelligence for data centers: Loudoun County's data center boom meets AI power demands. Link

Anthropic is quietly building the full developer lifecycle, not just the model. Memory for free users, agentic code review, scheduled tasks, community meetups — in a single week, Anthropic shipped four features that have nothing to do with model benchmarks. The strategy is clear: while OpenAI competes on reasoning scores and Google on multimodal breadth, Anthropic is betting that developer workflow integration is the real moat. The code review feature is especially telling — they built it because their own engineers were drowning in PRs, which means it's solving a real bottleneck, not a demo problem. When the company building the model is also the company using it hardest, the feedback loop gets very tight. For builders choosing a platform, the question is shifting from "which model is best?" to "which ecosystem makes me most productive?" (5,374 likes | 328 RTs) Read more →

Until next time ✌️