Most economists forecast 0.1-1.5% annual excess growth through 2035, while AI insiders predict 3-30%. Most economists, though, assume that today's ChatGPT is essentially the ceiling of AI progress and that visions of a virtual AI coworker won't really materialize, but how is it possible that cheap, PhD-level intelligence wouldn't boost productivity? The forecasts imply that either AGI will be far less general than claimed or that adoption will be dramatically slower than historical technology diffusion. AI's 2024 economic footprint appears to be around 0.5% of GDP.
Kimi K2 Thinking, a reasoning MoE model developed by China's Moonshot AI team, shows just how many people are learning quickly to train excellent AI models. Early reports note a distinctive style and writing quality - the model is beating leading closed models on some benchmarks such as Humanity's Last Exam and BrowseComp. It is the closest open models have been to the closed frontier of performance ever.
AI agents are acting on your behalf - and on behalf of your customers. Learn how to give them access to the systems and APIs they need, without sacrificing trust, in these two free resources by Auth0: _Customer Identity Trends 2025_ [10] and _Securing AI Agents - The New
Chinese lab Moonshot AI released an open-source reasoning model that matches or exceeds GPT-5 and Claude Sonnet 4.5 on some benchmarks, including a 44.9% on Humanity's Last Exam and 71.3% on SWE-Bench Verified.
Google introduced File Search, a built-in RAG system for the Gemini API that simplifies retrieval tasks. It includes vector search, automatic citations, and broad file format support with free storage and on-demand embedding.
Extended conversations can reshape how AI models answer the same question. After 10 rounds of debate about moral dilemmas, GPT-5 reversed its position 54.7% of the time when asked again. When models changed their stated beliefs, they also chose different tools to accomplish tasks.
Soumith Chintala has left Meta after working at the company for 11 years. Chintala spent almost eight years leading PyTorch, taking it from nothing to more than 90% adoption in AI. He plans to move to something small and new. PyTorch has been left in capable hands, and Chintala is confident that its values will stay intact, but things may change in flavor because he will no longer impose his own taste from the top.
Google Cloud has introduced new AI-oriented instances powered by its own Axion CPUs and Ironwood TPUs. The new instances are aimed at both training and low-latency inference. Servers running these chips come equipped with Google's custom Titanium-branded controllers, which offload network, security, and I/O storage processing, enabling better management and higher performance. More details about Google's in-house designed processors are available in the article.
StackAI is the platform for building no-code enterprise AI agents, white-glove support included. Trusted by Morgan & Morgan, Nubank, BAE Systems, Insurama, Monzo...
The Parallel Search API is a web search tool designed for AI agents engineered to deliver the most relevant, token-efficient web data at the lowest cost.
In a thinly veiled dig at other AI labs, Mustafa Suleyman said the team will focus on practical technologies—like digital companions, disease diagnostics, and renewable energy—rather than “ill-defined and ethereal superintelligence."
Balancer has identified a rounding flaw in its swap logic as the root cause of the November 3 exploit that drained over $128 million from its Composable Stable Pools (CSPs) across multiple chains, including Ethereum, Base, Arbitrum, Avalanche, and Polygon. The issue stemmed from a rounding error in the upscale function for EXACT_OUT swaps within the v2 vault's batchSwap feature, which attackers used to manipulate pool balances and siphon liquidity. Emergency response teams like Hypernative, SEAL 911, StakeWise DAO, and BitFinding helped recover portions of the stolen funds, with StakeWise DAO retrieving 73.5% of osETH losses. The exploit primarily affected v5 CSPs, while v6 pools were automatically paused, and Balancer v3 remains unaffected.
Coinbase has called on the US Treasury Department to ensure that upcoming regulations for the GENIUS Act remain faithful to congressional intent, warning against regulatory overreach that could stifle innovation and hinder America's ambition to become the “crypto capital of the world.” Chief Policy Officer Faryar Shirzad emphasized that the rules should focus narrowly on stablecoin issuers, not on software developers, validators, or open-source protocols. Coinbase also encouraged Treasury to clarify that loyalty and rewards programs do not constitute “interest” payments prohibited under the Act and to classify payment stablecoins as cash equivalents for tax and accounting purposes.
Almanak's Strategy Builder allows agents to build strategies for users to deploy and dynamically change over time. Its launch video demoed an LP accumulation strategy to provide liquidity and accumulate the underlying tokens. Almanak's templates allow for slight alterations via natural language. Each strategy has a unique, AI-generated dashboard with custom metrics specific to the strategy's implementation.
Hyperstition Markets introduce prediction markets engineered for reality construction rather than passive forecasting, rewarding traders for both accurate prediction and narrative enactment. The mechanism requires traders to demonstrate actions aligned with predictions verified through smart contracts, creating Nash equilibria where rational actors maximize returns by betting on and working toward specific futures. Applications include technological development bootstrapping, cultural transformation requiring epistemological framework shifts, and collective action problems where early believers profit from manifesting self-fulfilling prophecies. Unlike traditional prediction markets aiming for unbiased information aggregation, treating futures as discovered, Hyperstition Markets explicitly engineer reflexive loops between belief, incentive, and action, treating futures as designed.
Crypto's three primary liquidity channels – DATs, stablecoins, and ETFs – have plateaued after strong momentum through Q1 2025. Combined DATs and ETFs have grown from $40 billion to $270 billion since early 2024, while stablecoins doubled from $140 billion to $290 billion before flattening. Despite global M2 expansion and US quantitative tightening officially ending, high SOFR rates keep cash parked in T-bills yielding attractive returns, redirecting liquidity away from crypto into equity markets instead of compounding digital asset inflows. The slowdown across all three channels signals broad deceleration in fresh capital deployment rather than simple product rotation. Crypto entered a "self-funded phase" where capital rotates internally between majors and alt sectors rather than entering net new, creating PVP dynamics with short-lived rallies, narrowing breadth, and volatility spikes driven by liquidation cascades instead of sustained trends.
Leading crypto applications like Pump.fun ($724M revenue) and Hyperliquid ($667M) are now out-earning the blockchains they're built on, such as Solana ($1.3B revenue, $632M in fees). These apps trade at far lower valuations despite using most of their revenue for token buybacks and transparent cash flows, while L1s retain only a fraction of the activity-generated value. The long-standing assumption that chains capture proportional value from ecosystem growth is breaking down as onchain value accrues to applications, not infrastructure. The “L1 premium” could continue eroding unless chains find new mechanisms, like white-labeled stablecoins, to internalize more of the liquidity and fees flowing through their ecosystems.
Ethereum's future lies in proof aggregation, with L2s like ZKsync defragmenting the network and scaling it to millions of TPS, rendering alternative L1s obsolete. The Ethereum L1 remains the source of incorruptibility and neutrality, expected to reach 10,000+ TPS within three years, but still not enough for global demand. As L2s evolve into cashflow machines replacing traditional payment rails (e.g., credit cards, FX, and remittance apps), their success will flow back to Ethereum, which underpins their security. Investors should own the L2s generating yield to accumulate ETH, the risk-free asset of the EVM economy.
Since ETH last traded at $4,600, we've seen record-breaking L1 and L2 transaction volumes, JPMorgan's plan to accept ETH as collateral, SEC approval of staked ETH ETFs, and the rise of tokenized assets like gold and Fidelity's $200M onchain fund.
Lucid Agents is a developer library by the Daydreams team that combines x402 and ERC-8004 together to create trustless, verifiable agents that can pay one another and monetize their services.
Stability AI largely defeated Getty Images in a UK High Court case over AI training data. However, the judge avoided ruling on whether AI models require permission to use copyrighted material for training—the central question dividing tech companies and the creative industries. Getty won only on a narrow trademark claim regarding watermarks but dropped its main copyright argument mid-trial due to weak evidence, while the judge rejected secondary infringement claims since Stable Diffusion doesn't store copyrighted works. Getty continues to pursue similar claims against Stability in California as part of a broader legal battle.
Apple TV has introduced a new sonic logo created by Finneas that features warbling electronic sounds followed by three gentle piano notes, offering a more calming alternative to the anxiety-inducing idents common among streaming services. The mnemonic exists in three versions, ranging from one to ten seconds, with the longest intended for theatrical releases of Apple Studios films. Accompanying the audio rebrand is a colorful animated opening bumper, signaling a playful new direction for Apple's typically minimalist design approach.
Jameson, the world's best-selling Irish whiskey, has unveiled a refreshed identity by JDO that modernizes its look while honoring its heritage. The redesign refines key brand elements—like the crest, wordmark, and bottle design—to create visual harmony across the portfolio, clarify each whiskey's story, and attract both new and loyal drinkers through subtle, character-driven updates.
As AI takes over solution generation, designers must focus on problem framing, a skill that is undermined by cognitive biases, such as the need for closure, ambiguity aversion, and action bias, which push teams toward premature answers. These tendencies cause designers to freeze problem definitions early rather than embracing ambiguity. However, experienced practitioners who resist this urge and refine the frame create a shared understanding and uncover deeper insights. Effective framing isn't a delay, but essential design work that transforms AI outputs from noise into useful material and compounds value throughout delivery.
At this point, the massive productivity gains from AI coding agents are prompting three main reactions from design teams. Some designers are shifting their role to focus on UX alignment after features are built, essentially flipping the traditional design-then-develop workflow. Others are adopting AI coding tools themselves to prototype and ship features directly, while skeptics dismiss the speed as "faster slop" despite developers' productivity unlikely to return to previous levels.
Rebrands have become cultural flashpoints, reflecting deeper societal fatigue and polarization rather than just design critique. In an era of constant change, even small logo updates can feel like a loss of stability or identity, making emotionally intelligent, culturally aware design more crucial than ever to maintain trust and relevance.
Generative AI has transformed product design from a two-way human-computer interaction into a three-way Human-Model-Interface (HMI) experience, requiring designers to have a deep understanding of AI model capabilities, input requirements, and output quality. Designers must now prototype early and iterate alongside model development, considering complete end-to-end workflows while asking whether the AI can reliably deliver on user expectations and build trust. A four-phase framework—prompt, plan, show, and next—helps users feel supported through AI interactions by gathering context intelligently, clarifying next steps, providing visibility into the model's reasoning, and offering actionable follow-up recommendations.
In this post, a designer explores working with generative AI through the lens of four artists, arguing that treating AI as an instrument requiring skilled practice offers more creative potential than passive "vibe coding" dependency. Brian Eno's collaborative approach, working alongside the machine as a gardener, and Holly Herndon and Mat Dryhurst's integration into AI systems while building ethical frameworks, demonstrate healthier relationships with technology than Rick Rubin's skill-free positioning under the machine. Using Hayao Miyazaki's Spirited Away as an allegory, the post warns that AI's insatiable consumption mirrors No Face's hunger—dangerous without boundaries—and suggests finding creative satisfaction requires knowing when to stop rather than feeding endless appetite.
Brand strategy isn't dead—it's simply evolving. Despite claims that AI, shifting consumer behavior, and new branding models have made it obsolete, effective brand strategy remains rooted in timeless storytelling and human insight. The difference now lies in crafting bolder, more imaginative, and culturally attuned narratives that stand out in a changing world.
Linear redesigned its mobile app by creating its own version of Apple's Liquid Glass design system, rather than using Apple's APIs, which gave it complete control over customization for its expanding user base.
London-based designer Alexander Clark created Olio, a complete visual identity for an Italian restaurant inspired by Southern Italian dining culture, as a portfolio project without a client.
Slack's engineering team reduced its build pipeline time from 60 minutes by applying classic software engineering principles to its build system using Bazel. They treated build performance optimization like code performance optimization by implementing caching (storing results of expensive build work), parallelization (spreading work across multiple compute resources), and better granularity (breaking work into smaller, more efficient units).
This article goes over an investigation of an extremely rare Ruby bug where an internal FFI (Foreign Function Interface) Hash was mysteriously replaced by a String object at runtime, causing 2,700 identical "undefined method 'default' for String" errors. The root cause was discovered to be missing write barriers in FFI versions before 1.17.0, which prevented Ruby's garbage collector from knowing about object references in C code, allowing it to incorrectly free the Hash and replace it with a String at the same memory address.
This dev modified CPython to track allocations and discovered that Python frequently allocates `PyLongObject` objects, especially when printing integers. However, optimizations like freelists and pool allocators help reduce the real-world impact by reusing allocated memory. While Python allocates often, the cost is mitigated by these strategies, though tagged pointers, a technique used by other languages to avoid allocation, could further improve performance.
Game design, while complex in detail, can be understood through a twelve-step framework. At its core, game design revolves around presenting players with engaging problems to solve, offering clear feedback, and escalating challenges to encourage learning and mastery. Games consist of interconnected loops, combining simple problems into larger systems.
Devs should experiment with LLM agents as they are actually quite simple to implement. This article goes over how to create basic agents with personalities and tool integration, showing how easy it is to set up a loop where an LLM interacts with the environment. Understanding context engineering, token limits, and agent design decisions is necessary for building effective and secure agents.