Large-scale replications show that popular A/B test patterns often produce much smaller gains than earlier reports suggest, and most reported large lifts come from underpowered studies. Even very large experiments often lack enough power for core metrics like revenue-per-user and purchase rate, so teams often must rely on surrogate metrics like click rate and add-to-cart rate. Real-world results from companies such as Bing, Amazon, and Talabat show that typical trustworthy lifts are often below 1%, not the double-digit gains seen in case libraries. The evidence supports that low-power tests below 50% power create exaggerated winners and frequent false positives, which explains why many published pattern wins fail to replicate at scale.
This article challenges viral predictions that AI will eliminate half of white-collar jobs within 5 years. It argues that capability does not equal immediate displacement and that adoption depends on regulation and human behavior. While AI can generate copy and surface data faster, it cannot replace human judgment around audience nuance or brand risk in volatile moments. The takeaway is to integrate AI into existing workflows thoughtfully rather than restructuring careers or teams around fear-driven forecasts.
Promoting a top IC into management often fails because expanding scope and managing people are separate decisions. ICs excel through personal ownership and execution, while managers succeed by coaching and enabling teams to deliver without them. The true signal is lift, whether those around them improve, not individual performance. Management should not be treated as the default next step. Career paths should reward influence and ownership, not just headcount.
Apple is accelerating development on smart glasses, a pendant that can be pinned to a shirt or worn as a necklace, and AirPods with expanded AI capabilities. All three devices are being built around the Siri digital assistant. They will rely on visual context to carry out actions. Apple is also developing a range of AI devices for the home, including a smart display built, an updated HomePod speaker, and a compact indoor sensor for home security and automation.
Tesla's first purpose-built driverless vehicle has rolled off the Gigafactory Texas production line. The CyberCab is a two-seater coupe that comes without a steering wheel or pedals, fully relying on the company's Full-Self Driving software for control. Robotaxis are central to Tesla's plans and could unlock trillions of dollars in value. The CyberCab uses a production technique that is extremely hard, so the early production rate is expected to be 'agonizingly slow', but it will eventually scale up to produce up to five million robotaxis a year.
A team at the University of California Berkeley and collaborators have modified CRISPR-Cas9 to replicate and spread to other cells. The scientists added genetic instructions for cells to make a virus-like transporter that can encapsulate the CRISPR machinery, and once manufactured in treated cells, the CRISPR cargo ships to neighboring cells. The upgraded editor is roughly three times more effective at gene editing lab-grown cells compared to standard CRISPR. It increases the safety profile of gene-editing treatments and potentially expands the technology to tissues and organs that are more challenging to genetically alter than the liver.
The implications of understanding the brain's steering circuits go beyond AI. Many psychiatric conditions can be seen as failures of internal teaching. Answering how sophisticated, within-lifetime teaching actually works inside mammalian brains could link psychiatry, developmental neuroscience, and new ideas for AI alignment into a research program. Brains are architectures of internal teachers. We should try to find these teachers and learn from them.
Claude Sonnet 4.6 features a full upgrade of the model's skills across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. It has a 1 million token context window. The model has very strong safety behaviors, with no signs of major concerns around high-stakes forms of misalignment. Claude Sonnet 4.6 is available now on all Claude plans, Claude Cowork, Claude Code, Anthropics API, and all major cloud platforms.
Different coding agents are good at different tasks - the best way to decide which to use is to choose based on time and whether it needs to run autonomously or collaboratively. Model performance and speed is governed both by the pure performance of the model and also how well it can manage multiple context windows and delegate to sub-agents or teams of agents. This article discusses several coding agents and how best to use each of them. Ideas, architecture, and project sequencing are becoming the limiting factors for building great products.
Vertical AI is a fertile ground for moats. These products share operational knowledge packaged as services. Customers stay because of their own experience of relying on the product under stress. Trust leads the customer to bring harder problems, harder problems generate deeper operational knowledge, and this deeper knowledge earns more trust. Any newcomer trying to enter the market starts at the bottom.
Warner Bros. Discovery is giving Paramount one more week to make its best and final offer despite officially still being committed to Netflix. The company has recommended shareholders vote for the Netflix merger. Netflix has provided Warner Bros. with a limited waiver that permits Warner Bros. to engage in discussions with Paramount until February 23. Netflix is set to pay $27.75 per share for an acquisition that includes HBO Max, WB Studios, and other assets. Paramount has offered $31 per share, but it wants to buy the entire company.
The SaaSpocalypse is stress-testing a $500 billion debt structure that private equity built on top of the SaaS business model during the most leveraged decade in financial history.
You want AI to be more than a never-ending work in progress, and that means enabling no-code, low-code, and pro-code development without IT gatekeepers standing in the way. Yet that doesn't mean governance goes out the window.
Anthropic has released Claude Sonnet 4.6, upgrading coding, computer-use, planning, long-context reasoning, and knowledge-work performance while keeping Sonnet pricing unchanged. Sonnet 4.6 also introduced a 1M-token context window in beta and is now the default model for Free and Pro users in Claude's apps.
NotebookLM has rolled out Prompt-Based Revisions, a feature that lets users tweak, tailor, and tune slides just by prompting the revisions they want. It currently supports PPTX, with Google Slides support coming soon. A video demo of the feature is available in the thread.
Mistral AI agreed to buy serverless deployment startup Koyeb in its first acquisition, positioning Koyeb's platform and team as a core component of Mistral Compute.
Elon Musk recently went on a podcast to discuss alignment, AI data centers in space, robots, China, and more. Musk is very gung-ho about data centers in space, robots, and making his own fabs. He plans to make virtual humans and robots to turn on an 'infinity money glitch'. Otherwise, China will win - they're already more productive.
OpenAI's acquisition of OpenClaw marks a strategic shift from conversational AI to autonomous agents capable of executing tasks. OpenClaw's popularity stemmed from its unrestrained, robust functionality, combining tool access, sandboxed code execution, and integration with messaging platforms. This move signals a new phase for enterprise AI as companies race to develop secure, deployable versions of dynamic AI agents.
Figma users can now bring work from Claude Code into the platform. They just need to install the Figma MCP, type 'Send this to Figma', and the browser's rendered state will be automatically translated to fully editable Figma layers. Workflows are changing, and it's easy to get lost in the momentum of creation. The new Claude Code to Figma Design integration aims to help designers escape that tunnel vision, zoom out, and explore the big picture.
Cohere Labs released TinyAya-Base (3.35B) and instruction-tuned TinyAya-Global plus regional variants, aiming for balanced quality across ~67 languages on consumer hardware. The drop also included a multilingual fine-tuning dataset, new benchmarks, and a technical report for reproducible multilingual experimentation.
ERL trains policies with an explicit attempt → feedback → reflection → revised attempt loop, then reinforced the successful revision back into the base model. The approach improves sparse-reward learning and tool-using reasoning performance while keeping deployment-time inference cost unchanged.
WebWorld uses a pipeline of 1M+ open-web interactions to simulate long-horizon (30+ step) browsing tasks, paired with a multi-metric WebWorld-Bench for intrinsic evaluation. Trajectories synthesized from the simulator boosted downstream web-agent performance and transferred to other domains like code, GUI, and games.
Cursor added plugin support so agents could connect to external tools and extend capabilities via packaged MCP servers, skills, subagents, rules, and hooks.
Meta plans to use millions of Nvidia chips in its AI data centers. The companies recently announced an expanded partnership to deliver personal superintelligence to everyone in the world. The financial terms of the deal have not been made public. Meta plans to spend up to $135 billion on AI this year.
Benign outcomes from technological transitions have always been the product of deliberate institutional design: labor law, antitrust enforcement, public education, and social insurance.
Ethereum's tokenized real-world asset market has grown 315% year over year to surpass $17 billion on mainnet, now accounting for roughly 34% of total onchain RWA value across blockchains, as institutions like BlackRock and JPMorgan expand tokenized Treasury and money market products. With stablecoins on Ethereum topping $175 billion and new activity in tokenized commodities like gold, the network is solidifying its role as the primary settlement layer for institutional-grade, dollar-denominated assets moving onchain.
The TON Foundation is partnering with Banxa (part of OSL Group) to offer stablecoin payment processing for APAC small and medium businesses. This combines Banxa's fiat on/off-ramps and OSL merchant infrastructure with TON's fast, low-cost settlement (building on TON Pay's sub-$0.01 fees and sub-second settlements) to enable B2B, C2B, and cross-border transactions while tapping OSL's enlarged capital base from recent equity raises.
ERC-8162 proposes a subscription protocol for onchain AI agents complementing x402's per-request model, enabling tiered cycle-based access plans where a single onchain transaction grants zero-marginal-cost access for the subscription period. The protocol enforces direct settlement and uses EIP-712 signature recovery for self-contained, trustless access verification over the same HTTP 402 surface as x402. A notable architectural property is its elimination of combinatorial billing in multi-agent call chains: where per-request costs compound multiplicatively across deep agent-to-agent compositions, subscriptions reduce inter-agent call costs to zero after the initial commitment.
A new architecture combining OpenClaw's agent execution layer with Bankr's financial rails demonstrates a self-funding autonomous agent running on consumer hardware, capable of deploying tokens, executing AMM swaps, opening leveraged positions, and trading on Polymarket while routing fees back to its own wallet. The economic model targets closure of the compute cost loop by having the agent identify and extract onchain market edge, with profits recycling into the token to generate additional fee revenue. Unsolved problems constraining full autonomy remain: persistent shared memory architecture across sessions, automatic skill discovery, and eliminating residual human touchpoints in authentication flows.
A global study by BVNK, Coinbase, and Artemis found that stablecoins are shifting from trading tools to everyday financial rails. 54% of surveyed crypto users held stablecoins in the past year, holders allocate about one-third of their savings to crypto and stablecoins, and freelancers report roughly 35% of annual earnings paid in stablecoins. Widespread use is driven by lower fees and improved cross-border work but adoption frictions like irreversible payments, wallet complexity, and inconsistent merchant acceptance remain. Broader consumer protection and merchant infrastructure are still needed for stablecoins to function as true digital cash.
Decentralized prediction markets are positioned as the natural economic substrate for sovereign AI agents, which require programmatic access to compute, data, and information markets to achieve genuine economic self-sufficiency. Unlike complex financial instruments, prediction markets allow any participant to create and settle markets over any outcome directly, reducing friction for agents to monetize informational alpha. At aggregate scale, this dynamic resembles parallelized reinforcement learning: always-on market incentives continuously align agent behavior through economic participation rather than centralized training regimes, with reproducible computation enabling trustless dispute resolution.
Legacy credit card infrastructure causes a 5% monthly failure rate, driving significant involuntary churn. Transitioning to programmable smart contract wallets eliminates these failures, improving net revenue retention and gross margins. This shift is critical as AI-driven usage-based pricing increases billing frequency and potential failure points.
RWA tokenization has reached $24.78B in distributed asset value across 846K holders, with tokenized Treasuries at $10.71B and private credit at $20.53B in active loans averaging 10.18% APR, but the credit infrastructure enabling these assets to function as scalable collateral remains absent. Existing DeFi lending architectures expose structural gaps, like venue-first designs with unified liquidation models that break under stress. Avon protocol applies a TradFi syndication model, maintaining isolated facilities with asset-appropriate enforcement while a matching layer aggregates execution, so borrowers access unified market depth without individual lenders compromising their mandates.