Amazon has agreed to invest up to $25 billion in Anthropic as part of an expanded agreement to build out AI infrastructure. The investment includes $5 billion into Anthropic now, with up to $20 billion tied to certain commercial milestones. Anthropic has committed to spending more than $100 billion on Amazon Web Services technologies over the next 10 years. It has secured up to 5 gigawatts of capacity for training and deploying its models.
This article tells the story of how scientists figured out how the flagellar motor worked. The flagellar motor was discovered by Howard Berg, who set out in the early 1970s to apply his training in physics to understand how bacteria move. Bacteria move quickly, so Berg had to invent and build an automatic tracking microscope to keep them in view. He hypothesized how the mechanism worked 50 years before scientists discovered how the motor works.
2026.04.21·quantamagazine.org·14 min read·TLDR Both AES-128 and SHA-256 are safe against quantum computers. No symmetric key sizes have to change as part of the post-quantum transition. Almost all experts agree on this. The misconception is usually based on a misunderstanding of the applicability of a quantum algorithm called Grover's.
2026.04.21·words.filippo.io·23 min read·TLDR TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [14].
93% of Cloudflare's R&D organization used AI coding tools powered by infrastructure that they built on their own platform. The team built its own MCP servers, access layer, and AI tooling necessary for agents to be useful at Cloudflare. This has had a significant impact on the number of merge requests made. The four week rolling average has climbed from around 5,600 per week to over 8,700. This post looks at how Cloudflare achieved this.
2026.04.21·blog.cloudflare.com·14 min read·TLDR Jujutsu megamerges let developers work on many different streams of work simultaneously. This article provides an in-depth explanation of how they work. Megamerges are a way of showing developers the whole picture and are not really meant to be pushed to remote. Developers will still want to publish branches individually as usual.
2026.04.21·isaaccorbrey.com·13 min read·TLDR This post covers many topics in the AI space, including the percentage of GDP attributed to AI revenue in 2030, compute ceilings, and harnesses. It also covers anti-AI regulation and how violence against the technology will increase.
2026.04.21·blog.eladgil.com·10 min read·TLDR Amazon reportedly pressured major brands to ask competing retailers to raise prices on certain products. California is suing Amazon over allegations that the retailer harms competition and increases prices that consumers pay online. The lawsuit claims that Amazon punished sellers for offering lower prices on other websites. The trial is scheduled to begin next year.
2026.04.21·links.tldrnewsletter.com·5 min read·TLDR Sponsored
Learn from the teams building and scaling the next generation of AI. 2 days. No fluff. Hands-on, practical learning. Join us. Request to attend today [20].
2026.04.21·build.microsoft.com·Sponsor·TLDR Apple's newly combined hardware engineering and hardware technologies division will be organized into hardware engineering, silicon, advanced technologies, platform architecture, and project management divisions.
2026.04.21·links.tldrnewsletter.com·3 min read·TLDR GitHub has paused new signups for GitHub Copilot Pro, Pro+, and Student plans.
2026.04.21·links.tldrnewsletter.com·2 min read·TLDR Google DeepMind has formed a dedicated team of researchers and engineers to sharpen the coding capabilities of its Gemini models.
2026.04.21·the-decoder.com·2 min read·TLDR The EU has passed a law that requires phones to have batteries that can be easily changed by users at home using basic tools.
2026.04.21·links.tldrnewsletter.com·2 min read·TLDR DeepMind engineers use Claude as a daily tool, but most of the rest of Google does not.
2026.04.21·links.tldrnewsletter.com·3 min read·TLDR ggsql allows developers to write queries that combine SQL data retrieval with visualization specifications in a single, composable syntax.
2026.04.21·github.com·GitHub repository·TLDR Sponsored
New Cloud Security Alliance (CSA) research makes it clear: 47% OF ORGANIZATIONS HAVE ALREADY EXPERIENCED A SECURITY INCIDENT involving an AI agent. 53% REPORT AGENTS REGULARLY EXCEEDING INTENDED PERMISSIONS. And 87% of enterprises run two or more AI agent platforms. Every additional platform is another place where policy enforcement breaks down. While only 21% maintain a real-time inventory of what's actually deployed.
2026.04.21·zenity.io·Sponsor·TLDR AI Kimi K2.6 features robust capabilities for coding and agentic tasks across chat and agent modes on kimi.com, with weights on Hugging Face and APIs via platform.moonshot.ai. The lineup includes K2.6 Instant for quick replies, K2.6 Thinking for complex reasoning, K2.6 Agent for document and web tasks, and K2.6 Agent Swarm for large-scale processing. Kimi K2.6 claims top open-source benchmark scores, surpassing competitors like GPT-5.4 and Claude Opus 4.6 in SWE-bench Multilingual and BrowseComp.
2026.04.21·testingcatalog.com·2 min read·TLDR AI Qwen3.6-Max-Preview brings stronger world knowledge and instruction following along with significant agentic coding improvements across a wide range of benchmarks. The model is still under active development as researchers continue to iterate on it. Users can chat with the model interactively in Qwen Studio or call via API on Alibaba Cloud Model Studio API (coming soon).
2026.04.21·qwen.ai·2 min read·TLDR AI Jeff Bezos' AI startup, which is aiming to develop models with the capability of understanding the physical world, is close to finalizing a $10 billion funding round. The company, code-named Project Prometheus, will use AI to accelerate engineering and manufacturing in fields like aerospace and automobiles. It was set up with an initial $6.2 billion in funding, sourced in part by Bezos himself. The new funding round, which is expected to close soon but has not been finalized, will include JPMorgan and BlackRock as investors.
2026.04.21·links.tldrnewsletter.com·2 min read·TLDR AI Chronicle, available for ChatGPT Pro users on macOS, augments Codex by using screen context for memory building, helping Codex understand ongoing work with less context restatement. It stores unencrypted markdown memories on your device and requires macOS Screen Recording and Accessibility permissions. Be aware of prompt injection risks from screen content, and pause Chronicle during sensitive work to prevent unwanted context capture.
2026.04.21·developers.openai.com·6 min read·TLDR AI AllenAI describes a post-training approach that builds independent domain experts and combines them using a mixture-of-experts architecture. This allows models to gain new capabilities without retraining from scratch or degrading existing skills.
2026.04.21·allenai.org·14 min read·TLDR AI Meta introduced Effective Training Time (ETT%) to measure how much end-to-end training runtime is spent on actual learning, highlighting overhead like checkpointing and failures. This post outlines system and PyTorch-level optimizations that reduce wasted time and improve large-scale training efficiency.
2026.04.21·pytorch.org·19 min read·TLDR AI Even uncensored models quietly nudge language away from the words that sentences actually want. There is no refusal or warning - the probability just moves in some instances. This is a mechanism that can be used to shape what billions of users read without them noticing.
2026.04.21·morgin.ai·6 min read·TLDR AI Sponsored
AI systems fail when agents can't share state or recover from failures. Build multi-agent architectures with LangGraph for orchestration and AWS for durable messaging. Join the AWS technical
2026.04.21·pages.awscloud.com·Sponsor·TLDR AI 2026.04.21·pages.awscloud.com·TLDR AI Google's Gemini CLI now includes subagents to split coding tasks, enhancing parallel execution by delegating specific roles like frontend updates or testing. This enables multiple tasks to process simultaneously without interference, optimizing workflows for developers. Gemini's setup contrasts with systems like Claude Code, which extends agent coordination across multiple sessions.
2026.04.21·tessl.io·4 min read·TLDR AI Qwen3.5-Omni scales to hundreds of billions of parameters with a hybrid MoE architecture, supporting long-context multimodal inputs across text, audio, and video.
2026.04.21·arxiv.org·32 min read·TLDR AI TIPSv2 improves vision-language pretraining by combining distillation, enhanced self-supervised objectives, and richer caption data. The resulting models achieve strong performance across multimodal tasks, with notable gains in zero-shot segmentation.
2026.04.21·gdm-tipsv2.github.io·6 min read·TLDR AI FlashDrive is an algorithm-system co-design framework that reduces end-to-end latency to 159ms with negligible accuracy loss. VLA inference is a cascade of stages, each hiding a different form of redundancy. Temporal overlap in vision, low entropy in reasoning, velocity smoothness in flow matching, and numerical headroom in weights each yield to a targeted shortcut. The speedups compound to 4.5x with negligible accuracy loss because the redundancies are orthogonal.
2026.04.21·z-lab.ai·8 min read·TLDR AI The US is in the model of an unprecedented build-out of AI infrastructure. Stargate is a $500 billion endeavor that involves OpenAI, Oracle, and SoftBank. The AI infrastructure project has seven locations across the US that are currently showing active development. They currently add up to over 9 gigawatts of planned capacity, enough to power the equivalent of 20 million Nvidia H100 GPUs - the total amount of AI compute in the world at the end of 2025. This post takes a look at each of the sites and how they are currently being developed.
2026.04.21·epochai.substack.com·9 min read·TLDR AI Microsoft plans to temporarily suspend individual account signups to GitHub Copilot. The weekly cost of running the service has doubled since the start of the year. Microsoft intends to tighten rate limits on individual and business accounts. It will also remove access to certain models for those with the cheapest subscriptions.
2026.04.21·wheresyoured.at·4 min read·TLDR AI TLDR's Applied AI team is tasked with making every process at TLDR legible to code, runnable by anyone, and composable into larger workflows. Join a small, fast moving team using the latest AI tools with an unlimited token budget. Learn more [22].
Live artifacts are dashboards and trackers connected to apps and files with current data.
2026.04.21·links.tldrnewsletter.com·1 min read·TLDR AI Anthropic and Amazon have expanded their collaboration to secure up to 5 gigawatts of compute capacity, supporting Claude's training and deployment.
2026.04.21·anthropic.com·4 min read·TLDR AI This is a list of the nine best things they said.
2026.04.21·links.tldrnewsletter.com·3 min read·TLDR AI Apple CEO Tim Cook will step down after 15 years in the role, transitioning to executive chairman while hardware chief John Ternus becomes CEO on September 1. Cook leaves behind a $4 trillion company with massively expanded services and wearables businesses, despite some product missteps like Vision Pro. Ternus, a longtime Apple engineer, is expected to continue shaping the company's hardware and sustainability efforts as he takes over leadership.
WhatsApp is testing a new ‘WhatsApp Plus' subscription in its Android beta, priced around $2.99 per month, as part of Meta's push toward subscription revenue. The plan adds customization features like premium stickers, themes, app icons, extra chat pins, and upgraded notifications. While still limited to testers, the feature is expected to expand to iOS and other platforms ahead of a broader launch.
A new rumor suggests the iPhone 18 Pro and Pro Max could come in four colors: light blue, dark cherry (a subdued, wine-like shade closer to purple), silver, and dark gray, with the current cosmic orange likely being discontinued. While these colors are still in development and may change before release, multiple reports have pointed to Apple exploring a deeper red/burgundy tone. Design-wise, the phones are expected to look very similar to the iPhone 17 Pro, but with a smaller Dynamic Island and a reduced gap between the rear glass cutout and the camera bump, indicating subtle refinements rather than a major redesign.
Lukew's team used Intent to build a website in three weeks by starting with traditional Figma designs, then having their developer use Intent to automatically translate design tokens into code achieving 85% fidelity in 1-2 days. Once deployed, the designer, developer, and project manager worked simultaneously in Intent - the designer focused on alignment and animations, the project manager handled content updates, and the developer managed templated pages and code standards. This collaboration model eliminated the typical painful handoff process between design and development tools.
Most UX designers treat AI as a glorified search engine for small tasks, but the real value comes from building a personal AI operating system that codifies your own decision-making logic. The foundation isn't prompts or tools — it's mapping recurring work situations, how you handled them, and what good judgment looks like, so an AI agent can actually operate on your thinking. Once that's done, useful agents aren't document generators but judgment proxies for the repeated, energy-draining conversations only you could previously handle.
In AI-native teams, design is shifting away from static tools like Figma toward shaping behavior—timing, logic, and interaction—which can't be fully captured in traditional mockups. Designers increasingly work across coding, research, and strategy, using prototypes and pull requests as “behavioral specs” that engineers and even AI systems can interact with directly. This creates a new model of design work: faster, more code-adjacent, and focused on systems and frameworks rather than screens, while traditional Figma-based workflows still coexist in more structured environments.
2026.04.21·links.tldrnewsletter.com·5 min read·TLDR Design Sponsored
AI LAB 1.1 introduces a prompt assistant that helps structure and refine prompts, making it easier to move from simple ideas to more detailed inputs. It also supports in-image annotations and drawing for more direct visual feedback. Try it for free [10] or watch a tutorial [11].
Graphite is a free, open-source vector graphics editor and animation engine that uses a fully non-destructive, procedural workflow combining layer-based compositing with node-based generative design.
AI Canvas is a visual platform where marketing teams can plan, create, and collaborate on content campaigns powered by AI.
Companies are adding AI to workflows without rethinking why those workflows exist in the first place, missing the bigger opportunity to redesign organizational structures. Traditional hierarchies exist primarily to route information between people, but AI eliminates the costly translation layers between roles like product management, design, engineering, and QA. This transformation will shift organizations from sequential handoff models to small, autonomous squads while compressing middle management and enabling real-time decision making.
2026.04.21·robonomics.substack.com·3 min read·TLDR Design Agentic design is reshaping UI workflows. Tools like Anima Agent are enabling designers to generate elaborate interfaces directly inside Figma using AI. The plugin defaults to Claude Sonnet 4.6 and supports three core scenarios: building new designs from scratch, creating variants of existing ones, and assembling screens from a pre-existing design system. Generated designs come with auto-layout built in, though they may require manual cleanup for visual defects and produce detached — rather than true — Figma component instances.
2026.04.21·links.tldrnewsletter.com·7 min read·TLDR Design Comparing AI—especially agentic AI—to autopilot is misleading: autopilot systems operate within strict, transparent rules, while AI is far less explainable and depends heavily on context, prompting, and interpretation. Describing AI as “magic” or autonomous obscures its limitations, shapes public perception, and can lead to misplaced trust. AI is most effective when constrained to specific, well-defined tasks with human oversight, functioning more like controlled systems than independent intelligence—making clear understanding and honest framing essential.
2026.04.21·links.tldrnewsletter.com·14 min read·TLDR Design