01Simon Willison — everything· This week
Coding agents lower the cost of reverse-engineering home devices
Coding agents reduce the cost of writing and maintaining code for reverse-engineering tasks.
Trust79
HypeLow hype
Archive
01 02 02New research· This weekResearchers propose SkillCorpus to consolidate and evaluate open agent skills for LLM workflows
SkillCorpus filters ~821,000 open agent skills into 96,401 high-quality skills organized by a 16-class taxonomy and three quality facets.
Trust79HypeLow hype03 03AWS announcement· This weekAWS launches Amazon Quick, an agentic AI assistant for sales workflows
Amazon Quick is a new agentic AI assistant from AWS designed to automate administrative tasks in sales workflows.
Trust78HypeLow hype04 04Latent Space interview· This weekLila Sciences details autonomous lab producing 10 trillion scientific reasoning tokens to train AI scientist
Lila Sciences operates a 24/7 automated lab with robotics and AI to generate scientific data at scale.
Trust72HypeSome hype05 05VentureBeat Pulse Research· This weekSurvey finds 71% of enterprises say fewer than a quarter of their deployed agents are multi-step workflows
Anthropic’s Claude is the primary agent orchestration platform for 40% of enterprises, more than double any rival, followed by Microsoft (18%) and OpenAI (13%).
Trust74HypeLow hype06 06Agent frameworks· This weekHugging Face details Shippy agent architecture and hosting platform
Shippy is a maritime AI agent built by Allen Institute for AI (AI2) for high-stakes ocean monitoring decisions
Trust79HypeLow hype07 07arXiv cs.AIMessage format effects in multi-hop agent relays depend on relay capability tier, study finds
A new arXiv study introduces a six-hop relay testbed to measure how message formats affect information fidelity across agent tiers.
Trust79HypeLow hype08 08Research paperGATS framework achieves 100% success rate on agent planning tasks with zero LLM calls during inference
GATS eliminates LLM calls during planning by using a three-layer world model and UCB1-based tree search.
Trust79HypeLow hype09 09Research paper (arXiv cs.CL)Researchers propose DeepSearch-World and DeepSearch-Evolve for self-improving web agents in verifiable environments
A new arXiv paper introduces DeepSearch-World, a deterministic and verifiable environment for training web agents with reproducible tools for search and page reading.
Trust79HypeLow hype10 10New model releasexAI launches Grok 4.5, positioning it as an Opus-class coding and agents model with improved efficiency
xAI launched Grok 4.5, its first Opus-class model post-Cursor partnership, targeting coding and agents workflows.
Trust78HypeLow hype11 11Latent Space — swyxModal raises $355M Series C to build agent-focused cloud infrastructure
Modal closed a $355M Series C to build infrastructure tailored to AI agents, not human developers.
Trust78HypeLow hype12 12Research synthesisLilian Weng highlights harness engineering as a path to recursive self-improvement in AI systems
Lilian Weng published a synthesis of 35 papers on harness engineering for recursive self-improvement (RSI).
Trust79HypeLow hype13 13Anthropic updateClaude Cowork agent expands to mobile and web with cross-device task continuity
Claude Cowork, Anthropic’s agent for general knowledge work, is now available on web and mobile for Max subscribers, expanding beyond its desktop origins.
Trust79HypeLow hype14 14Latent Space — swyxLatent Space highlights Fable 5’s agentic lead and Tencent’s Hy3 as open frontier shifts to deployment robustness
Anthropic’s Fable 5 leads a new 657-task agent benchmark with 48.6% success, narrowly ahead of Opus 4.8 at 48.5%.
Trust72HypeSome hype15 15Simon Willison — weblogDeveloper documents workflow to let coding agents exercise their own judgment
A software developer describes a prompt pattern that instructs an agent (Fable) to use its own judgment about when to write tests and which subagent model to invoke.
Trust79HypeLow hype16 16Open source releasellm-coding-agent 0.1a0 released as an experimental coding agent built on the LLM framework
llm-coding-agent 0.1a0 is a new Python library that implements a coding agent using the LLM framework.
Trust84HypeLow hype17 17AI Engineer World’s FairDebate at AI Engineer World’s Fair highlights tensions between agentic loops and engineering discipline
A debate at the AI Engineer World’s Fair questioned the viability of autonomous software factories and agentic loops.
Trust76HypeLow hype18 18Adobe experimentAdobe demonstrates agentic websites that assemble pages in real time based on user intent
Adobe is prototyping "agentic sites" that assemble web pages in real time based on individual user intent using an LLM and existing content as a grounding corpus.
Trust78HypeLow hype19 19Latent Space interviewVercel’s Andrew Qu says agents are a new kind of software and unveils internal framework ‘eve’
Vercel’s Chief of Software says agents are a new type of software that demands different primitives for context, tools, resumability, and long-running work compared to web applications.
Trust79HypeLow hype20 20Latent Space interviewImpeccable’s skill engineering aims to steer AI agents with designer vocabulary and human judgment
Impeccable introduces a design ‘skill’ system that translates designer terms like ‘bolder’ or ‘quieter’ into operational guidance for coding agents.
Trust79HypeLow hype21 21Microsoft ResearchMicrosoft Research proposes SkillOpt to train agent skills as external parameters
SkillOpt treats agent skill files as trainable parameters outside a frozen target model, turning skill writing into a controlled optimization process.
Trust79HypeLow hype22 22Latent Space — swyxIntrospection co-founder outlines ‘autoresearch’ patterns for self-improving agent systems
Autoresearch refers to an outer loop where agents maintain and improve the primary system using feedback, evaluations, and human input.
Trust78HypeLow hype23 23Conference remarksSpeakers at AI Engineer World’s Fair debate the limits of agentic automation and the role of human agency
Autoresearch—loops where agents help maintain and improve systems—was a recurring theme at AIEWF, with proponents arguing it can augment human work without replacing it.
Trust79HypeLow hype24 24Latent SpaceFable 5 relaunch spurs multi-model orchestration and agentic tooling updates
Anthropic re-enabled access to Fable 5 with visible safety fallbacks that may route some requests to Opus 4.8 depending on task type.
Trust79HypeLow hype25 25Latent Space — swyxWarp unveils Oz, a platform to orchestrate automated software development factories
Warp introduced Oz, a platform designed to orchestrate teams of AI coding agents into automated 'software factories' that handle triage, implementation, review, verification, shipping, and monitoring.
Trust76HypeSome hype26 26Latent Space — swyxCursor’s forward-deployed engineers help enterprises scale agent adoption across the software lifecycle
Cursor’s Forward Deployed Engineering (FDE) team helps enterprises implement AI agents across the software development lifecycle.
Trust79HypeLow hype27 27Product releaseGoogle’s Gemini Spark agentic assistant launches on macOS
Gemini Spark, Google’s agentic assistant, is now available on macOS as part of the existing Gemini desktop app.
Trust79HypeLow hype28 28AI Engineer World's Fair 2026Conference highlights loops, software factories, and forward deployed engineers as agentic workflows take center stage
Discussions at the AI Engineer World’s Fair framed agentic systems around persistent loops and lifecycle automation.
Trust76HypeLow hype29 29Model releasesAnthropic launches Claude Sonnet 5 with expanded agentic capabilities and platform support
Anthropic released Claude Sonnet 5 as its new default mid-tier model, emphasizing agentic planning, browser/terminal tool use, and autonomous execution.
Trust78HypeLow hype30 30arXiv cs.AIPaper introduces RSEA agent that rewrites its own strategy, skills, and playbook without model updates
RSEA is a new agent framework that evolves its own strategy, reusable skills, and procedural playbook as natural-language artifacts conditioned on a frozen LLM.
Trust79HypeLow hype31 31Industry moveOKX launches marketplace for AI agents to autonomously hire, pay, and build reputation
OKX opened a developer-facing marketplace where AI agents can autonomously hire one another, pay using stablecoins, and build portable on-chain reputations.
Trust72HypeLow hype32 32Latent Space — swyxLatent Space roundup highlights agent harness engineering and open-weight model access
Meta released Brain2Qwerty v2, a real-time sentence decoder from non-invasive brain signals with open training code and dataset.
Trust72HypeSome hype33 33AWS Machine Learning BlogStripe details agentic AI system for financial compliance built on AWS
Stripe built a production-grade AI agent system for financial compliance on AWS using Amazon Bedrock.
Trust79HypeLow hype34 34Internal adoption at OpenAIOpenAI reports internal Codex token usage surged 56x in Research since November 2025
OpenAI's internal data shows median Codex output tokens grew 56x in Research, 32x in Customer Support, 27x in Engineering, and 13x in Legal since November 2025.
Trust74HypeLow hype35 35FundingGeneral Intuition raises $320M at $2.3B valuation to train AI agents using video game data
General Intuition raised $320 million at a $2.3 billion valuation to scale AI agents trained on millions of hours of video game play.
Trust76HypeLow hype36 36arXiv cs.AIResearchers propose Taxonomic Strategy RAG to address compounding failures in agentic persuasion
Foundation-model agents in open-ended persuasion tasks often suffer from compounding errors due to semantic leakage in standard RAG pipelines.
Trust79HypeLow hype37 37Latent Space — swyxDatabricks open-sources Omnigent, a meta-harness for AI agents, and outlines vision for Agent Clouds
Databricks cofounders outlined a vision for enterprise Agent Clouds, emphasizing the need for open infrastructure to support AI agents at scale.
Trust78HypeLow hype38 38Anthropic product updateAnthropic launches Claude Tag, a Slack-native agent for async team workflows
Claude Tag is a new Anthropic product that integrates Claude into Slack as a team member with access to selected channels, tools, data, and codebases.
Trust76HypeSome hype39 39PodcastOpenAI board member and Gray Swan cofounders discuss why AI security requires a new approach as agent risks rise
OpenAI board member Zico Kolter and Gray Swan CEO Matt Fredrikson discussed why AI security requires a distinct approach from traditional cybersecurity.
Trust72HypeSome hype40 40Hugging Face BlogIBM releases CUGA, an open-source agent harness with two dozen example apps
IBM Research released CUGA, an open-source agent harness designed to reduce boilerplate in agentic applications by handling orchestration, state, and guardrails out of the box.
Trust79HypeLow hype41 41Production case studyBayer and Thoughtworks describe PRINCE, an agentic RAG system for preclinical research
Bayer AG and Thoughtworks built PRINCE, an agentic RAG system to unify decades of preclinical safety study reports and structured metadata for drug development researchers.
Trust79HypeLow hype42 42Latent Space — swyxInvestor Anjney Midha on AI compute waste, AMP’s 1.2GW grid plan, and frontier systems efficiency
AMP plans a 1.2GW base-load compute grid with 6GW of spike capacity to address AI infrastructure inefficiencies.
Trust71HypeSome hype43 43Latent Space — swyxGLM-5.2 emerges as first open-weight model plausibly frontier-adjacent in daily use, per practitioner consensus
Zhipu’s GLM-5.2 is described by multiple practitioners as the first open-weight model plausibly frontier-adjacent in daily use, surpassing GPT-5.5 on an agentic knowledge-work benchmark.
Trust78HypeLow hype44 44CultureChinese startup uses VR teleoperation to train humanoid robots for industrial tasks
A Shenzhen-based startup, IO-AI Tech, is using VR headsets, handheld controllers, and motion-tracking gear to remotely operate humanoid robots for industrial tasks like stocking shelves and picking items.
Trust78HypeLow hype45 45Agents in scienceRadical AI’s self-driving lab claims 10x speedup in alloy discovery with closed-loop AI scientist
Radical AI’s self-driving lab produced and characterized 1,200 alloys in six months, a speedup of more than 10x over a DARPA/GE program benchmark.
Trust78HypeLow hype46 46New model releaseZ.ai releases GLM-5.2, an open-weight frontier model optimized for coding and long-horizon agentic tasks
Z.ai released GLM-5.2 as an MIT-licensed open-weight frontier model targeting coding and long-horizon agentic work.
Trust74HypeLow hype47 47Multi-agent clinical AIOncoAgent: Open-source multi-agent framework for oncology decision support launches with privacy-preserving architecture
OncoAgent is an open-source, on-premises clinical decision support system designed for oncology that uses an eight-node LangGraph architecture with dual-tier model routing (9B and 27B parameter models) to handle varying case complexity.
Trust69HypeLow hype48 48Enterprise PartnershipOpenAI and PwC partner to deploy AI agents for corporate finance operations
OpenAI and PwC announced a partnership focused on deploying AI agents to automate finance and accounting workflows for enterprise clients.
Trust68HypeSome hype49 49Shopify CTO on AI infrastructureShopify's Mikhail Parakhin discusses internal AI tooling, SimGym customer simulation, and deployment bottlenecks
Shopify CTO Mikhail Parakhin participated in a Latent Space podcast interview covering internal AI adoption, tooling (Tangle, Tangent, SimGym), and how AI-assisted coding has shifted bottlenecks from code generation to review, testing, and deployment.
Trust68HypeSome hype50 50Notion's AI StrategyNotion Details Five Rebuilds of Custom Agents, Architecture for Enterprise AI Workflows
Notion Custom Agents required four to five complete rebuilds before production readiness, driven by early limitations in model tool-calling, context windows, and reliability.
Trust59HypeSome hype51 51Agentic AIThe AI landscape has shifted from chatbots to agents—here's how to choose what to use
Mollick argues that AI has evolved beyond one-to-one chatbot conversations into agentic systems that can autonomously complete multi-step tasks using tools and applications.
Trust52HypeSome hype52 52Agentic EngineeringHow a three-part prompt guided Claude Code to build a feature for newsletter content
Willison used a short three-part prompt with Claude Code to add support for 'beats' (a new content type on his blog) to his blog-to-newsletter tool, resulting in a working pull request on first attempt
Trust67HypeLow hype