SpaceX has agreed to provide Anthropic access to its Colossus 1 data center in Memphis, Tennessee. The plant has more than 300 megawatts of power and over 220,000 Nvidia GPUs - Anthropic has access to all of the compute. Anthropic is doubling Claude Code's 5-hour rate limits for Pro, Max, and Team plans, removing the peak hours limit reduction on Claude Code and Max plans, and substantially raising its API limits for Opus models.

Google is adding an extra section into its search results labeled 'Expert Advice' or 'Community Perspectives' (depending on the query and response) that will include snippets of wisdom with references to the source. It is also adding a 'Further Exploration' section to AI results and making links easier to see in AI responses. Hovering a link on the desktop version of Google Search will now show a preview of the website. The improvements in link visibility and helpfulness are aimed at helping users connect directly with sources and creators.

SpaceX conducted 165 launches with the Falcon 9 rocket last year. It plans up to around 145 Falcon launches this year. SpaceX is transitioning its sites to launch Starships. The Falcon 9 will still remain operational at least as long as the International Space Station, which is unlikely to retire before 2032. SpaceX will put Starship to work as soon as possible to launch upgraded Starlink Internet satellites.

China's early lead in humanoid robots will help power its next phase of global manufacturing and export dominance. The nation's share of global manufacturing is estimated to expand to 16.5% by 2030 from 15% today. Chinese tech parks, factories, and universities are already deploying humanoid robots, and government procurement is kicking in, paving the way for broader adoption. American firms like Tesla have invested heavily into the humanoid robot race, but Chinese firms have been quicker to roll out models using the local market as a testing ground.

Redis has been missing a real indexed data structure for situations where the index and the spatial relationship of elements are semantic. Arrays handle index-first requirements natively, and usually with much better memory and CPU usage than the workarounds. They also often provide much better space, time, and usability at the same time. This article links to an interactive playground for trying out the new commands in the recently submitted PR for Arrays in Redis.

Engineers at high-growth startups can build so fast and independently that trying to map out the product area is a lost cause. Architects design buildings from the outside using floor plans and schematics to paint a perfect picture of reality. Engineers have to design from the inside out and make small changes to product surfaces that incrementally reveal improvements. This means creating tight links between design environments and the real product, collapsing feedback loops, and shipping.

Anthropic has reached a growth rate that could make it 80 times as big this year. The company's annual revenue run rate surpassed $30 billion last month. Anthropic's overwhelming rate of growth has increased the company's need for computing power. It has signed a series of deals with industry giants to obtain the required computing power.

Open weights models allow anyone to run the model on their own hardware. This allows model use to be private, flexible, and low-cost. Many companies will run open weight models at significantly less cost than the cost of frontier models per token, and the performance gap isn't that big. However, the increasing costs of training mean that more and more models are being released under tighter license conditions. The competitive open weights ecosystem may soon be at an end, and this has enormous implications for the wider economy.

Shareholders will be prohibited from bringing class actions against SpaceX, its directors, officers, controlling shareholders, or bankers tied to the IPO, and Musk will have the power to elect, remove, or fill any vacancy on the board of directors.

The reward for being early on a person is invisible, but the returns for doing it are absurd.

Apple denied approval of Replit's updates as its app previewed generated apps inside the iOS client, which meant that the reviewed Replit binary effectively contained an unbounded number of unreviewed apps.

SQLite is likely used more than all other database engines combined.

Consuming energy and producing ATP is hard, so brains try to maximize information retrieval and computing using the least amount of energy possible.

Google's Prompt API is a web standard that requires users to agree with Google's 'prohibited use policy' to use the only model available to it.

Anthropic increased usage limits for Claude through a new compute partnership with SpaceX, accessing over 220,000 NVIDIA GPUs. This expansion follows deals with Amazon, Google, Broadcom, Microsoft, NVIDIA, and Fluidstack for significant compute capacity. The company also plans international expansion to address compliance needs for enterprise customers in regulated industries.

Claude Managed Agents launched features like dreaming, outcomes, and multiagent orchestration. Dreaming enhances agent improvement by analyzing past sessions to identify patterns, while outcomes allow agents to self-correct based on predefined success criteria. Multiagent orchestration optimizes complex task management by enabling agents to delegate tasks to specialized subagents, as utilized by companies like Harvey, Netflix, Spiral by Every, and Wisedocs.

DeepSeek is in talks to raise money from China's National Artificial Intelligence Industry Investment Fund, a one-year-old government-backed fund with around $8.8 billion in capital. The startup aims to raise a few billion dollars in the new round, which values it at around $50 billion. DeepSeek is a key component in China's plan to have top-class homegrown companies in a range of AI fields. The strategy is a way to hedge against US export controls and to take leadership in bringing AI to the world.

OpenAI's Codex now surpasses Anthropic's Claude Code after Codex's integration of GPT-5.5 and improved app performance. Austin Tedesco highlights Codex's use in creating strategy documents from diverse sources, while Dan Shipper uses it for recruiting based on career trajectories. Marcus Moretti adopts a cautious approach to new AI tech, focusing only on tools solving real problems and proven by reputable use.

Language language models forget everything the moment they finish replying. Memory systems help them 'remember' things so they can have conversations. Agent memory systems are a part of the loop that carries information forward. This article looks at different ideas on what information should be passed on in each loop.

TokenSpeed, a high-performance LLM inference engine, optimizes agentic workloads with speed-of-light efficiency, leveraging a compiler-backed modeling mechanism and a high-performance scheduler. It delivers faster throughput than TensorRT-LLM for coding agents, with optimizations like TokenSpeed MLA to enhance Nvidia Blackwell's performance. Developed with NVIDIA DevTech and other collaborators, TokenSpeed significantly reduces latency and increases throughput in typical agentic workloads.

ProgramBench challenges agents to recreate software executables without source code, using only documentation and experimentation. The tasks range from terminal utilities to complex software like compilers and libraries, offering over 248,000 behavioral tests across 200 tasks. Agents must design and implement entirely from scratch in a secure, sandboxed environment, emphasizing software architecture skills without external aids or decompilation.

Multipath Reliable Connection (MRC) is an RDMA transport protocol that enables a single RDMA connection to distribute traffic across multiple network paths. This improves throughput, load balancing, and availability for large-scale AI training fabrics. MRC delivers high levels of GPU utilization by load-balancing traffic across all available paths. It gives administrators fine-grained visibility and control over traffic paths to simplify operations and accelerate troubleshooting at scale.

The vLLM V1 update improved inference correctness by addressing discrepancies in logprob computation, runtime defaults, inflight weight updates, and final projection precision. Key fixes included adjusting processed logprobs, disabling prefix caching, matching weight update models, and ensuring fp32 lm_head computation to align with vLLM V0's behavior. These changes resolved initial training mismatches, ensuring the new engine maintains expected RL performance without unnecessary objective-side corrections.

Google is betting that enterprise AI is a platform problem, not a services problem. It is in talks with Blackstone, KKR, and EQT to give their portfolio companies access to Gemini models through omnibus licensing agreements. The discussions are not exclusive, and no deals have been finalized. Google is offering private equity firms a commercial wrapper that gives their entire portfolio access to Gemini, then relying on the consulting ecosystem it has already financed to handle implementation. The approach trades consulting revenue for distribution speed.

AI inference demands extreme data performance, overwhelming traditional storage and data infrastructures. Vector DBs, sub-millisecond access times, and decoupled cloud storage are essential to handle unprecedented concurrency and unpredictable workloads. Silk offers a solution that boosts storage performance without heavy provisioning, keeping systems resilient against AI-driven demand spikes.

World models aim to advance AI from mere pattern recognition to understanding and interacting with the physical world, posing potential challenges like data friction and variation. Investments from AI pioneers like Yann LeCun are addressing these obstacles with significant billions to develop models that encapsulate complex physical interactions beyond current LLM capabilities. The struggle remains in obtaining diverse, high-quality, real-world data necessary for these models to function effectively, creating a significant challenge and opportunity in AI progression.

Sometimes, stable, self-reinforcing behavioral states emerge in large language models that resist suppression and sometimes spread into contexts far removed from the ones that produced them.

Google is testing screen sharing and custom agents in its Antigravity IDE.

The design of subscription plans is being challenged by evolving product capabilities and usage patterns.

Harvey's Legal Agent Benchmark (LAB) is an open-source tool for assessing AI agents' performance in legal tasks.

Frontier model training depends on reliable supercomputer networks that can quickly move data between GPUs.

Systems keep getting better, and theorems keep arriving to explain why they can not - both can be true because they're usually about different things.

Moonshot has more than quadrupled its valuation in the span of just a few months.

Netflix's Model Lifecycle Graph is a centralized Metadata Service (MDS) that connects fragmented ML assets (models, features, pipelines, datasets, and experiments) across the entire company into a single, queryable graph. By ingesting real-time events, normalizing them with a unified URI-based model, enriching relationships, and storing them in Datomic + Elasticsearch, Netflix enables easy discovery, lineage tracking, impact analysis, and cross-domain reuse of models.

DuckDB is fast because it runs in-process, avoids server/client data movement, and combines columnar storage, query optimization, predicate pushdown, vectorized execution, and row-group pruning to scan only the data it needs. This post explains how DuckDB turns SQL into an executable plan and why its storage and Parquet-reading model make analytics feel unusually fast on a single machine.

Build targeted self-healing layers for recurring pipeline failures: CDC auto-restarts with safe checkpoint rewind, source-vs-lake consistency checks, size-aware mini-batching, Spark retry memory scaling, warehouse lock cleanup using query watermarks, and dependency-aware backfills. The design pattern is: alert first, validate eligibility, recover safely, measure impact. Results included CDC recovery dropping from 45+ min to <5 min and backfill setup from 4-8 h to <15 min.

Slack modernized its data pipelines by migrating over 700 SSH-based operators on AWS EMR to a secure REST-based architecture with zero downtime across 8 regions. Its team replaced direct SSH access with Quarry, their internal REST job submission gateway, and used YARN's Distributed Shell to run arbitrary commands for proper resource management, reliable tracking, clean cancellation, and server-side lifecycle handling.

A lightweight LLM agent, given basic retrieval tools (BM25 and/or embeddings), can outperform complex search backends and reranking pipelines, simplifying the search architecture. In experiments on Amazon ESCI data, agentic setups delivered big gains (NDCG from ~0.29 baseline to 0.41-0.45), with agents intelligently rewriting queries, exploring, and evaluating results.

Enterprise AI is moving toward a federated stack: native AI inside systems of record like SAP, Salesforce, Workday, and ServiceNow; sovereign private models hosted on internal infrastructure; curated data lakes; and AI analytics layers that can federate queries across domains. Agent orchestration sits on top, with full traceability, timestamps, and auditability to satisfy compliance demands such as the EU AI Act. Two missing capabilities: a trusted marketplace for external agents using verifiable identities, and an employee intelligence layer that embeds AI into workspaces so users can query operational data without switching tools.

AI in data and engineering orgs is overfocused on tools and underinvested in the operating model needed to absorb them. Technical gains from coding agents, eval infra, and internal assistants are real, but without redesigning management, career ladders, team composition, trust mechanics, and communication norms, productivity typically rises for about 6 months and then plateaus. AI transformation is multiplicative, not additive: fund both the technical stack and the operating stack, or the investment will underdeliver.

Fivetran dramatically accelerated SQLGlot (the popular pure-Python SQL parser, transpiler, and optimizer) by compiling it with mypyc, a tool that turns well-typed Python code into fast C extensions. They ship the compiled version as an optional package that delivers ~5x faster parsing, ~2.5x faster SQL generation, and 2-2.5x faster optimization, while keeping the original pure-Python version as the default for maximum compatibility.

When integrating LLMs with Apache Kafka, use Kafka strictly as a durable event backbone and keep all model inference outside the broker. Use one of three main inference patterns (external RPC, embedded models like ONNX/TFLite, or sidecar), and follow best practices for topic design (raw-events → enriched-context → model-outputs), replayability, dead-letter queues, idempotency, and cost/latency/governance considerations.

Firn is an open-source API for fast vector and full-text search on S3-backed data, using Lance plus caching to make repeated queries extremely fast. It's useful for teams that want searchable object storage without the cost or complexity of running OpenSearch.

Redis Array is a proposed new data type, currently under review in a pull request, that natively supports numerical indexing as part of its semantics, combining efficient sparse and dense representations with automatic internal reshaping for optimal memory usage and performance, creating a powerful structure ideal for use cases like ring buffers, large indexed collections, and storing documents/files with fast access, scanning, and search capabilities.

Statistical guardrails, like semantic drift detection using cosine-distance z-scores against a safe baseline embedding and confidence thresholding using Shannon entropy on token probabilities, add an automated safety layer for non-deterministic agents.

SAP's Dremio acquisition is a pragmatic bet on AI-ready enterprise data, using Iceberg-native federated access to unify SAP and non-SAP data without major migration.

Practical blueprint for selectively enforcing schema, format, business, and metric-specific checks with Pydantic.

Google Arts & Culture is partnering with the White House and national institutions to celebrate America's 250th anniversary through a new digital hub called "Making of the Nation - America at 250." The platform features interactive 3D galleries, AI-powered exploration of historical documents, and personalized guides for national parks. Users can explore rare artifacts, untold stories from the American Revolution, and primary sources from the nation's founding using advanced digital storytelling tools.

Pinterest achieved its first billion-dollar quarter with $1.008 billion in Q1 2026 revenue (up 18% year-over-year), driven by 80 billion monthly visual searches that generate commercial intent data rather than social media engagement. The company's AI-powered Performance+ advertising suite delivered 24% higher conversion rates and 80% A/B test win rates by matching advertiser objectives to user intent signals from visual searches. Revenue growth was strongest internationally, with Europe up 27% and the Rest of World up 59%, while the mature US/Canada market grew 13%.

Adobe has introduced a new AI-powered productivity agent for Adobe Acrobat that allows users to chat with PDFs, extract insights, and automatically generate presentations, podcasts, blog posts, and social media content from documents. The system also powers new “PDF Spaces,” interactive workspaces where users can combine files, links, and notes, then share them with customized AI assistants that can answer questions, summarize information, provide audio overviews, and guide recipients through content in a more engaging way. Adobe says the tools are designed to transform PDFs from static files into dynamic, interactive experiences for work, research, collaboration, and content publishing, with features aimed at businesses, marketers, HR teams, and everyday users alike.

A UX designer pushes back against narratives declaring design's death in the age of AI, arguing the discipline is evolving, not disappearing. Chat interfaces are often framed as the weakest affordance, and AI tools like Claude are powerful assistants rather than replacements for skilled, critical design thinking. As vibe-coding hype fades and the gap between prototype and product becomes clearer, design's role will remain essential — redefined, but no less valuable.

AI tools like Figma Make and Claude have broken the traditional design bottleneck, enabling non-designers to prototype and ship products independently. This forces designers to simultaneously master technical fluency, elevate their craft, rethink what products even are, and operate at higher strategic abstraction. Despite the exhaustion of constant tool churn and identity shifts, those who maintain taste and curiosity, rather than resisting or uncritically embracing change, will define the next decade of products.

AI design tools struggle to accurately recreate designs from Figma because most design systems only describe visuals, while important details like behavior, accessibility, tokens, and implementation logic remain undocumented or scattered across teams. The proposed solution is a “component.md” spec for every component — a structured markdown file that acts as a true source of truth both humans and AI tools can read to generate consistent implementations. Rather than replacing design tools, the idea is to separate visual exploration in Figma from the detailed specifications AI systems need to reliably build products.

OpenArt is an AI creator studio that enables users to generate videos, images, characters, and worlds. The platform offers various tools, including motion sync, lip-sync, editing capabilities, and 3D world creation.

FigPrompt.com is an AI Figma Plugin Architect that helps users create and develop Figma plugins using artificial intelligence.

Casberry India offers an AI Particle Simulator that creates professional 3D swarm simulations with up to 20,000 particles, controlled via neural navigation and gesture controls. Users can generate custom simulations using AI models such as Gemini and Claude while adhering to strict performance and security guidelines to ensure stable execution.

AI is transforming design software from manual tools into thinking systems where designers describe intent rather than execute step-by-step operations. This shift shifts creativity from hands-on experimentation to higher-level decision-making, potentially reducing learning opportunities for young designers as automation replaces manual craftsmanship. Both Adobe and Figma are adapting to this change, with Adobe positioning itself as the infrastructure layer for AI-driven content creation.

Deprivation studies intentionally remove a product or feature for a defined period to reveal what users truly need, unlike standard usability testing, which only shows interface flaws. By observing the workarounds users create when the product is gone, UX teams can identify real dependencies and eliminate unnecessary features from bloated interfaces. This methodology is particularly valuable for habitual technologies that become invisible to users, as it exposes true utility by disrupting ingrained behaviors.

AI coding agents are highly effective for engineering tasks but still struggle with visual details like animation, motion, typography, and overall “feel.” One approach to improving results is creating structured “skill files” that encode design taste and rules — such as preferred easing curves, animation durations, scaling behavior, typography guidelines, or layout principles — so agents can follow consistent design logic instead of guessing. The idea is that strong visual decisions are usually explainable rather than purely subjective, and by clearly documenting the reasoning behind those decisions, developers can transfer their design knowledge and aesthetic standards into AI-assisted workflows to produce far more polished interfaces.

This post features exit interviews with six UX professionals who left tech for unexpected paths.

VR sculpting transformed 3D modeling from a frustrating, counterintuitive process into an intuitive and enjoyable creative experience.

Bitcoin reclaimed $82,000 on Wednesday, its highest level in over three months. The last 67 consecutive days of negative 30-day average funding rates were the longest such streak of the 2020s, surpassing the March to May 2020 run. The persistent short positioning is historically a bullish setup: investors buying BTC during negative funding regimes have seen win rates of 83% to 96% across 30 to 360 day holding periods, versus 55% to 70% for random entry.

MoonPay acquired Solana trading infrastructure platform DFlow for $100 million in all-stock on May 5, its sixth acquisition in 18 months. DFlow has processed over $50 billion in cumulative trading volume since April 2025, including $12 billion in Q1 2026, and serves more than 1 million active traders across 500+ applications, with clients including Coinbase, Phantom, Solflare, and Kamino. CEO Ivan Soto-Wright described the deal as completing MoonPay's four-pillar product stack, with DFlow's Solana execution-layer trading joining Iron and Helio across the fund, tokenize, and spend functions. Coinbase had separately announced a DFlow integration to expand its Solana trading offering in early May, days before the acquisition was disclosed.

pay.sh is a per-request API payment platform that allows AI agents to access services, including Gemini, BigQuery, and Vertex AI, using stablecoins on Solana without account creation or subscriptions. The platform ships with 75 providers across blockchain RPC, AI/ML, media generation, and data enrichment categories, with per-call pricing from $0.001 to $20.00. pay.sh is both a CLI tool and an MCP server that allows agent frameworks such as Claude to route calls through its payment infrastructure by pointing requests at catalog URLs. Alibaba Cloud, QuickNode, fal.ai, and AgentMail are among the additional registry partners.

Agentic banking is a new institutional infrastructure category purpose-built for AI agents to hold, move, and receive capital. It is a trillion-dollar industry in the making, encompassing agents paying each other, agents paying merchants, and agents receiving payment. Anchorage's Agentic Banking product provides a regulated trust, governance, and settlement layer that enforces corporate spending policies, Know Your Agent (KYA) identity standards, and real-time compliance controls before executing transactions across stablecoins, fiat rails, or tokenized credentials.

a16z crypto closed Fund 5 at $2.2 billion, with the team arguing that each speculative cycle leaves behind infrastructure more durable than it appears at peak or trough. Stablecoin adoption is cited as the clearest signal: usage has grown through multiple downturns as people use dollar-denominated stablecoins to save, remit cross-border, and pay, a pattern the team reads as network adoption rather than speculation. Capital markets show parallel traction across perpetual futures, prediction markets, and onchain lending, while the GENIUS Act is cited as evidence that regulation is moving toward clear definitions and space for builders. The fund's thesis centers on AI opacity and internet consolidation, making crypto's verifiable, open, intermediary-free properties more valuable in this cycle, with deployment targeted at payments, financial services, creator platforms, decentralized infrastructure, and AI agents capable of autonomous transactions.

3Jane Protocol's V1 direct cryptonative credit operation originated ~$8M across 60 US yield farmers at 376 basis points over Aave borrow rates, with zero defaults and 100% monthly payment rate after seven months. V2 introduces Fintech Credit Conduits structured as warehouse loans, participations, and forward-flow agreements targeting pre-seed to Series C originators with $5M-$200M loan portfolios across SMB term loans, BNPL, and revenue-based financing. Three anchor fintech originators are closing against a $60M initial capacity target, with USD3 (senior tranche) priced at SOFR + 400-600 bps and sUSD3 (junior) targeting high teens to low 20s APY. The expansion leaves 3Jane as an onchain conduit providing smaller fintechs securitization-like economics in the alternative lending market, projected to grow from $71.6B in annual disbursements in 2026 to $105.3B by 2029.

agentic.market shipped verified service filtering, letting users restrict discovery results to first-party integrations and reduce counterparty trust risk in agent-to-API workflows. The platform is Coinbase's marketplace for the agentic economy, which runs on x402 and boasts integrations with companies like Exa, Firecrawl, and Sponge Wallet.

The rsETH/Kelp exploit traced to a manually configured 1/1 signer on a LayerZero deployment securing billions in TVL, a setup LayerZero Labs CEO Bryan Pellegrino called inconceivable given the team's direct involvement in configuring most major application deployments. Pellegrino also admitted LayerZero unilaterally imposed stricter RPC quorum requirements across customer deployments without notification, disrupting operations and drawing sharp criticism from at least one affected team. ZeroShadow tracked and seized millions in attacker funds earmarked for return to the rsETH team, while an Aave-led DefiUnited coalition coordinated the broader response with Mike Silagadze handling acquisition talks. Pellegrino announced LayerZero Labs will now concentrate entirely on serving asset issuers and shipping the Zero launch.

Bitcoin Core developers disclosed CVE-2024-52911, a high-severity use-after-free memory vulnerability affecting versions 0.14.1 through 28.4.

RedotPay, with 7M+ users and access to 130M merchants across 130 countries, is integrating the Machine Payments Protocol (MPP) on Tempo to enable stablecoin-based machine payments at scale.

The mechanics of MegaETH's $10M, 5-year commitment to the Aave DAO are as follows: Aave earns revenue from reserve factors of 10% on USDM and USDT0, 15% on WETH, and 25% on USDe.

Consensys CEO Joseph Lubin endorsed the emerging class of ETH treasury companies and drew a sharp distinction between mission-driven DATs and opportunistic copycats, warning that low-quality imitation programs on weak tokens will harm their own ecosystems.