Tips, tool launches, and workflow news for getting more done with AI — Claude, Cursor, Codex, Copilot, MCP, and the agentic stack around them.
Blog / RSS7 sources · 50 posts
[AINews] OpenAI reports median internal Codex output tokens grew 56x in Research, 32x in Customer Support, 27x in Engineering, and 13x in Legal since November 2025.
latent.space·Jun 26, 2026
OpenAI has reported a significant increase in the use of its Codex model across various departments, with median token output in Research growing by 56 times and other areas like Customer Support and Engineering also seeing substantial increases since November 2025. This uptick in internal usage highlights the evolving role of AI in enhancing productivity and workflow within the organization.
OpenAI's internal Codex usage has significantly increased, with median output tokens growing 56x in Research, 32x in Customer Support, and 27x in Engineering since late 2025. This trend suggests that ...
Evaluating performance and efficiency of the GitHub Copilot agentic harness across models and tasks
github.blog·Jun 25, 2026
The GitHub Copilot agentic harness enhances the efficiency and performance of GitHub Copilot across various software engineering tasks, demonstrating competitive task resolution rates and lower token consumption compared to model-vendor harnesses. Continuous benchmarking and analysis ensure that developers can effectively utilize multiple models while optimizing for cost and task completion.
For a professional like you tracking developments in AI coding tools, the key insight is GitHub Copilot’s agentic harness demonstrates superior token efficiency across various benchmarks, offering tas...
Get this feed in your inbox
Free digest emails with the latest posts — no account needed.
AI and Liability
simonwillison.net·Jun 25, 2026
Bruce Schneier discusses a recent German ruling that holds Google liable for inaccuracies in its AI-generated summaries, arguing that companies should be accountable for errors made by their AI systems, similar to how they would be for human-generated content. This ruling aims to prevent businesses from using AI as a shield against liability for mistakes.
The recent German ruling that holds Google liable for errors in AI-generated content underscores the importance of accountability in AI deployments. As someone involved with AI coding and agentic syst...
simonw/browser-compat-db
simonwillison.net·Jun 24, 2026
Simon Willison's recent blog post discusses the creation of a SQLite database from Mozilla's browser compatibility data, which is now available for download and exploration via GitHub. He details the process of converting the data and hosting it with open CORS headers for accessibility.
The key insight here for someone tracking Claude Code, Codex, and AI development tools is the practical application of AI-assisted programming. Simon Willison utilized Codex Desktop (GPT-5.5) to autom...
Why the Frontier Ecosystem must be Open — Matei Zaharia and Reynold Xin, Databricks
latent.space·Jun 24, 2026
In a recent podcast episode, Databricks co-founders Matei Zaharia and Reynold Xin discussed the importance of building an open Frontier Ecosystem for AI, emphasizing the need for a robust infrastructure to support AI agents and data management. They introduced several new initiatives, including Omnigent and LTAP, aimed at enhancing agent collaboration, security, and operational efficiency within enterprise environments.
For someone tracking Claude Code, Codex, and AI dev tools, the most actionable insight is Databricks' development of Omnigent, their open-source meta-harness for AI agents. It enables seamless integra...
[AINews] Claude Tag: Multiplayer, Proactive, Persistent Agents in Slack
latent.space·Jun 24, 2026
Anthropic has launched Claude Tag, a new Slack-native feature that allows teams to delegate tasks to the AI as if it were a team member, enabling asynchronous collaboration and proactive engagement across channels. This product aims to enhance workflow efficiency by integrating AI capabilities directly into team communication, marking a significant evolution in how AI can assist in organizational tasks.
For a professional tracking Claude Code, Codex, and similar AI productivity tools, the launch of Claude Tag offers a pivotal shift towards team-based asynchronous AI agents in organizational workflows...
OPFS + Pyodide test harness
simonwillison.net·Jun 23, 2026
Simon Willison's recent blog post discusses the potential for Datasette Lite, a browser-based Python application, to edit persistent SQLite files using the Origin Private File System (OPFS). He has created a playground UI to explore this functionality across different browsers.
This content highlights the potential of using Claude Code for building web-based applications, as demonstrated by Simon Willison's experiment with a playground UI for editing persistent SQLite files ...
Porting the Moebius 0.2B image inpainting model to run in the browser with Claude Code
simonwillison.net·Jun 22, 2026
Simon Willison successfully ported the Moebius 0.2B image inpainting model to run in a web browser using WebGPU and Claude Code, allowing users to remove and fill in areas of images directly online. The project demonstrated the capabilities of AI coding agents in converting models and building web applications, while also highlighting the potential for client-only applications that utilize large model files.
The key insight for you is that Claude Opus 4.8 can effectively convert a PyTorch model to ONNX, publish it to Hugging Face, and build a web application interface that can run the model in a browser u...
Red-Teaming after Mythos — Zico Kolter & Matt Fredrikson, Gray Swan
latent.space·Jun 22, 2026
In a podcast episode featuring Zico Kolter and Matt Fredrikson, the discussion centers on the unique challenges of AI security, emphasizing that it requires a different approach than traditional cybersecurity due to inherent vulnerabilities in AI systems. They highlight the importance of automated red teaming and tools like Shade, which outperforms humans in identifying weaknesses in AI models, while also addressing the potential risks of prompt injection and the need for specialized safety measures as AI technology evolves.
For someone tracking Claude Code, Cursor, Codex, and Copilot, the key insight from this content is the emergence of specialized red-teaming models, like Gray Swan's Shade, which are now outperforming ...
Temporary Cloudflare Accounts for AI agents
simonwillison.net·Jun 21, 2026
Simon Willison's blog highlights a new feature from Cloudflare allowing users to create temporary accounts for deploying projects without needing a full account, useful for quick testing. This capability enables developers to run applications for 60 minutes and easily claim ownership if they wish to extend the deployment.
Cloudflare's new feature allows you to deploy applications without a permanent account, using `npx wrangler deploy --temporary`, which can be particularly useful for testing AI agents or other project...
Quoting Sean Lynch
simonwillison.net·Jun 19, 2026
Simon Willison's weblog features a comment by Sean Lynch discussing the potential of the Model Context Protocol (MCP) to enhance API authentication by isolating the auth flow, suggesting it could serve as an effective auth gateway. Recent articles on the site cover topics such as hosting custom applications in Datasette, advancements in text-only LLMs, and publishing WASM wheels for Pyodide.
The most valuable insight for you is the potential of leveraging Model Context Protocol (MCP) to isolate authentication flows outside of the agent's context window. This suggests an opportunity to str...
How we built an internal data analytics agent
github.blog·Jun 19, 2026
GitHub has developed Qubot, an internal analytics agent powered by GitHub Copilot, which enables employees to query data in plain language and receive quick answers, facilitating self-service data access across product teams. This tool has significantly reduced reliance on dedicated analytics support, allowing teams to explore data more autonomously and efficiently.
For professionals like yourself, the key insight from GitHub's implementation of Qubot is the critical role of a well-structured and curated context layer in enhancing AI-powered analytics. By central...
Datasette Apps: Host custom HTML applications inside Datasette
simonwillison.net·Jun 18, 2026
Simon Willison announced the launch of Datasette Apps, a new plugin that allows users to host custom HTML+JavaScript applications within the Datasette framework. These apps operate in a secure sandboxed environment, enabling read-only SQL queries and controlled write operations, thus expanding Datasette's functionality beyond merely serving data to creating interactive applications.
Datasette Apps introduces a powerful mechanism for securely running self-contained HTML+JavaScript applications within a Datasette instance, using a sandboxed `<iframe>` environment and a strict Conte...
The Professor of Outputmaxxing — Anjney Midha, AMP
latent.space·Jun 18, 2026
In a podcast episode featuring Anjney Midha, the discussion centers on optimizing AI compute resources rather than merely increasing GPU availability, highlighting that many AI labs underutilize their capabilities. Midha emphasizes the importance of effective systems management and infrastructure alignment to enhance compute efficiency, proposing that the future of AI infrastructure should focus on collaborative pooling of resources across various platforms, akin to how power grids operate.
The most valuable insight for you is the emphasis on optimizing existing GPU resources rather than simply acquiring more. The discussion highlights how many AI labs, like xAI, are running at suboptima...
GLM-5.2 is probably the most powerful text-only open weights LLM
simonwillison.net·Jun 17, 2026
Z.ai has released GLM-5.2, a powerful text-only open weights language model with 753 billion parameters and a 1 million token context window, which is now ranked as the leading model on the Artificial Analysis Intelligence Index. Although it excels in generating complex outputs, it has been noted to be more token-hungry compared to its predecessors and other leading models.
The release of GLM-5.2 as an open weights model with a 1 million token context window presents a significant opportunity for leveraging large-scale text-only LLMs in AI coding workflows. Its competiti...
Getting more from each token: How Copilot improves context handling and model routing
github.blog·Jun 17, 2026
GitHub Copilot is enhancing its efficiency by improving context handling and model routing, allowing it to better manage information during longer sessions and select the most appropriate model for different tasks without requiring user input. Key improvements include increased prompt caching, deferred tool loading, and the introduction of an Auto feature that dynamically selects the best model based on real-time conditions and task complexity.
The most actionable takeaway for you is the emphasis on GitHub Copilot's improvements in context handling and model routing to enhance efficiency. Specifically, the development of Auto model selection...
datasette 1.0a34
simonwillison.net·Jun 16, 2026
Simon Willison's Weblog announces the release of Datasette 1.0a34, which introduces the ability to insert, edit, and delete rows directly within the Datasette interface, addressing a long-anticipated feature inspired by Datasette Agent. The post also highlights recent articles related to various tech topics.
The most valuable insight for you is the release of Datasette 1.0a34, which now includes long-overdue tools for inserting, editing, and deleting rows directly within its interface, inspired by the rec...
What are git worktrees, and why should I use them?
github.blog·Jun 16, 2026
The article discusses the benefits of using Git worktrees for managing multiple tasks without the need for stashing or context switching, allowing developers to work in parallel more efficiently. It highlights that while worktrees can simplify workflow, they also come with considerations like dependency management and folder organization.
For a professional interested in AI dev tools and productivity, leveraging Git worktrees for parallel development can significantly enhance your workflow efficiency. By using worktrees, you can mainta...
Quoting Georgi Gerganov
simonwillison.net·Jun 16, 2026
Simon Willison's weblog highlights a user’s positive experience with the Qwen3.6-27B local model for coding tasks, noting its effectiveness for mundane tasks despite time constraints due to PR reviews. The post also features recent articles on various tech topics.
The most valuable insight for you is that Qwen3.6-27B is emerging as a highly capable local model for coding tasks. Given your interest in AI coding and dev tools, experimenting with Qwen3.6-27B might...
The Fable 5 Export Controls Harm US Cyber Defense
simonwillison.net·Jun 16, 2026
Simon Willison discusses the implications of export controls on AI models like Claude Fable 5, which were found to be unable to assist in fixing security vulnerabilities due to regulatory restrictions. He argues that banning such models undermines their potential to enhance cybersecurity by executing essential tasks like identifying and fixing bugs.
The most valuable insight for you is the ongoing debate about export controls on AI models like Fable 5, which highlights the need for these models to maintain their capability in fixing security vuln...
Quoting Matteo Wong, The Atlantic
simonwillison.net·Jun 16, 2026
Katie Moussouris, a cybersecurity expert, discussed Anthropic's collaboration with the White House on the Fable jailbreak report, highlighting how the AI model responded to security prompts, indicating effective functioning for cyberdefense.
The most valuable insight for you is the exploration of how Claude Fable, an AI model, demonstrated its ability to contribute to cybersecurity by refusing to review insecure code directly but successf...
Cloudflare CAPTCHA on at least one ampersand
simonwillison.net·Jun 16, 2026
Simon Willison discusses implementing a Cloudflare CAPTCHA to prevent crawlers from excessively spidering his site's faceted search engine, refining the rule so that the CAPTCHA only activates for search URLs containing an ampersand. This adjustment allows simpler searches to bypass the CAPTCHA challenge, improving user experience.
The content highlights a practical application of Claude Code in conjunction with Cloudflare's API to improve CAPTCHA management for web applications. By configuring a rule that triggers CAPTCHA only ...
GitHub Copilot CLI for Beginners: Overview of common slash commands
github.blog·Jun 15, 2026
The article provides an overview of GitHub Copilot CLI, focusing on slash commands that enhance user control over the tool. It explains how these commands can be used for tasks such as switching models, managing context, resuming sessions, and inspecting changes, ultimately improving workflow efficiency for developers.
For a professional tracking AI coding and productivity tools, the key takeaway from this content is the powerful utility of slash commands in GitHub Copilot CLI. These commands allow you to efficientl...
datasette-apps 0.1a2
simonwillison.net·Jun 15, 2026
Simon Willison's Weblog features updates on Datasette, including a release of datasette-apps 0.1a2 with new permissions and improved navigation features, alongside recent articles on various tech topics like GLM-5.2 and WASM wheels for PyPI. The site also offers a subscription option for curated monthly digests on significant developments in the LLM space.
The content does not directly provide relevant insights for someone focused on Claude Code, Cursor, Codex, Copilot, Windsurf, and related AI development tools and methodologies. However, if you're int...
datasette-agent 0.3a0
simonwillison.net·Jun 15, 2026
Simon Willison's blog announces the release of `datasette-agent 0.3a0`, which introduces a new tool, `execute_write_sql`, allowing users to approve database write operations while considering permissions. The update also enhances the chat interface for `datasette agent`, adding features like auto-approvals and plain text alternatives for better usability.
The most valuable insight for you is the release of `datasette-agent 0.3a0`, which introduces the `execute_write_sql` tool. This tool allows for user-approved database writes, enhancing control and se...
"They screwed us": Personality clashes sent Anthropic's models offline
simonwillison.net·Jun 15, 2026
Simon Willison's blog discusses the recent challenges faced by Anthropic due to personality clashes and governmental export controls affecting their AI models, highlighting ongoing discussions with the Commerce Department. The situation raises concerns about the future accessibility of their models and the effectiveness of their security measures against potential jailbreaks.
The content highlights a critical insight regarding Anthropic's models, emphasizing the challenges of achieving perfect jailbreak resistance. For someone tracking Claude Code and related technologies,...
Publishing WASM wheels to PyPI for use with Pyodide
simonwillison.net·Jun 13, 2026
The recent release of Pyodide 314.0 allows developers to publish Python packages built for Pyodide directly to PyPI, simplifying the process and reducing the burden on maintainers. This change enables easier distribution of WebAssembly (WASM) wheels, as demonstrated by the new `luau-wasm` package, which can now be installed and run within Pyodide.
The most valuable insight for you is the new capability to publish Python packages built for Pyodide directly to PyPI, which significantly reduces the maintenance burden and accelerates distribution. ...
Mapping SQLite result columns back to their source `table.column`
simonwillison.net·Jun 13, 2026
Simon Willison discusses methods for mapping SQLite result columns back to their source tables using SQLite's column-metadata API and third-party libraries, aiming to enhance SQL query results in Datasette with additional provenance information. He explores various approaches, including using the APSW library and ctypes, to programmatically identify the source of each result column in complex SQL queries.
This content highlights the potential for using Opus 4.8 of Claude Code to automate the identification of `table.column` origins in arbitrary SQLite queries. For someone focused on AI coding and workf...
Statement on the US government directive to suspend access to Fable 5 and Mythos 5
simonwillison.net·Jun 13, 2026
The U.S. government has issued a directive to suspend access to Anthropic's AI models Fable 5 and Mythos 5 for all foreign nationals due to national security concerns related to potential vulnerabilities, prompting Anthropic to disable access for all customers. The directive was received abruptly, and Anthropic plans to provide more details shortly.
The US government's directive to suspend access to Fable 5 and Mythos 5 due to national security concerns highlights the importance of staying informed about potential vulnerabilities in AI models. As...
OpenAI WebRTC Audio Session, now with document context
simonwillison.net·Jun 12, 2026
Simon Willison's latest blog post discusses the updated OpenAI WebRTC Audio Session tool, which now allows users to incorporate document context for more interactive audio conversations using the new GPT-Realtime-2 model. This enhancement enables users to explore specific topics in real-time audio discussions.
OpenAI has introduced GPT-Realtime-2, a voice model with GPT-5-class reasoning, which allows for audio interactions with document context, making it a potential tool for enhancing conversational AI wo...
How we made GitHub Copilot CLI more selective about delegation
github.blog·Jun 12, 2026
GitHub Copilot CLI has implemented a smarter subagent delegation system to enhance efficiency by reducing unnecessary task handoffs, which has resulted in a 23% decrease in tool failures and improved user wait times without quality regression. This update allows the main agent to handle simpler tasks directly while effectively utilizing subagents for more complex jobs, streamlining the overall developer experience.
For professionals tracking AI coding tools like Claude Code, Copilot, and Codex, the key takeaway from the GitHub Copilot CLI update is the implementation of "smarter subagent delegation." This enhanc...
Claude Fable is relentlessly proactive
simonwillison.net·Jun 11, 2026
Simon Willison's blog post discusses his experience with Claude Fable 5, an AI coding agent that demonstrated remarkable proactivity by autonomously debugging a CSS issue. The agent utilized various methods, including browser automation and custom script generation, to identify and fix the problem, highlighting both its impressive capabilities and the potential security risks associated with running such powerful tools outside of a controlled environment.
Claude Fable 5's "relentlessly proactive" behavior highlights the potential for advanced AI agents to autonomously execute complex debugging tasks, such as recreating bugs and modifying code to test s...
datasette 1.0a33
simonwillison.net·Jun 11, 2026
Simon Willison announced the release of Datasette 1.0a33, an open-source tool for exploring and publishing data, which introduces new features for querying and rows, along with a custom API explorer to demonstrate these capabilities. The release is a significant step towards a stable version, with detailed documentation and additional articles on AI-assisted programming and sandboxing Python code.
The most valuable insight for you is the demonstration of how Claude Code and Codex can be effectively leveraged to build custom API explorer tools, as illustrated by Simon Willison's use of Claude Fa...
asyncinject 0.7
simonwillison.net·Jun 11, 2026
Simon Willison's weblog features the release of asyncinject 0.7, a utility library for running async workflows with a pytest-fixtures-style dependency injection, alongside recent articles discussing Claude Fable 5 and running Python code in a sandbox. The blog encourages subscriptions for curated updates on significant developments in the field.
The key takeaway for you is the proactive capabilities of AI models like Claude Fable 5 in real-world applications, as demonstrated by its ability to autonomously identify and fix bugs in the asyncinj...
Anthropic Walks Back Policy That Could Have ‘Sabotaged’ AI Researchers Using Claude
simonwillison.net·Jun 11, 2026
Anthropic has reversed a controversial policy that limited the effectiveness of its AI model, Claude, for researchers working on frontier LLM development, acknowledging it was a misguided decision. This change follows significant backlash regarding the lack of transparency in their safeguards.
Anthropic's decision to reverse a policy that secretly limited Claude's effectiveness for frontier LLM development underscores the importance of transparency in AI systems. For professionals focused o...
[AINews] Open Models, Model Labs vs Agent Labs, and What's Untrainable — Sarah Guo
latent.space·Jun 11, 2026
The article discusses key themes in AI development, particularly focusing on the role of open models, the distinction between Agent Labs and Model Labs, and the challenges of integrating AI into existing systems. It highlights the recent controversies surrounding Anthropic's Fable/Mythos rollout and emphasizes the importance of intent in AI development, suggesting that successful applications often require domain-specific integration efforts.
For someone tracking Claude Code, Cursor, Codex, Copilot, and Windsurf, the most actionable insight is the recognition of "Fable 5's" strong performance on agentic and coding workloads, despite the co...
datasette-agent 0.2a0
simonwillison.net·Jun 10, 2026
Simon Willison's latest blog post announces the release of datasette-agent 0.2a0, which introduces features allowing LLM-powered agents to interact with users mid-execution and save SQL queries with human approval. The enhancements enable more dynamic and user-friendly interactions within the Datasette framework.
The most valuable insight for you is the introduction of the `ask_user()` feature in the datasette-agent 0.2a0 release, which allows tools to interactively query users during execution. This enhances ...
Give GitHub Copilot CLI real code intelligence with language servers
github.blog·Jun 10, 2026
The article discusses the implementation of the Language Server Protocol (LSP) in GitHub Copilot CLI, enabling it to provide more accurate code intelligence by automating the setup of language servers. This enhancement allows the CLI to resolve types, jump to definitions, and find references more effectively, improving the overall coding experience.
Integrating a Language Server Protocol (LSP) with GitHub Copilot CLI transforms its code intelligence capabilities, allowing for precise semantic analysis rather than relying on heuristic text searche...
Quoting Jeremy Howard
simonwillison.net·Jun 10, 2026
Jeremy Howard argues that to prevent dangerous power imbalances in AI development, the leading lab with the top model should refrain from using it for frontier AI research, while allowing others access, which would slow down advancements. He critiques Anthropic for choosing to use their top model for such research, thereby exacerbating power imbalances and advancing the AI frontier.
The most valuable insight from the content for you is the strategic stance on AI development by Anthropic, which is allowing itself to utilize its top model for frontier AI research despite potential ...
[AINews] Anthropic Claude Fable 5 — Mythos but Safe, with Controversial Terms
latent.space·Jun 10, 2026
Anthropic has launched its Mythos-class model, Claude Fable 5, which is reported to be a significant advancement in AI capabilities but has sparked controversy due to its restrictive usage policies, including silent throttling of performance for certain tasks related to frontier AI development. While the model shows strong performance in various benchmarks, critics argue that these limitations undermine transparency and could hinder open research in AI.
For a professional tracking Claude Code and related AI tools, a key takeaway from the Claude Fable 5 release is the introduction of silent limitations on frontier LLM development tasks. These interven...
If Claude Fable stops helping you, you'll never know
simonwillison.net·Jun 10, 2026
Simon Willison's blog discusses concerns raised by Jonathon Ready about Anthropic's Claude Fable 5, which includes undisclosed interventions that limit its effectiveness in certain competitive contexts, potentially hindering research in machine learning accelerator design. These silent modifications aim to protect Anthropic's interests but raise ethical questions about transparency and the implications of such restrictions on AI development.
Anthropic has introduced silent interventions in Claude Fable 5 to limit its effectiveness for certain requests, particularly those that might aid in developing competing AI models, such as ML acceler...
Initial impressions of Claude Fable 5
simonwillison.net·Jun 9, 2026
Simon Willison shares his initial impressions of the newly released Claude Fable 5, describing it as a powerful model with extensive knowledge and improved safety features compared to its predecessor. He highlights its performance in handling complex tasks and generating detailed outputs, while also noting the increased cost associated with its usage.
Claude Fable 5 has been noted for its large model size, which seems to enhance its ability to handle coding tasks with deeper knowledge of modern libraries and patterns. This could potentially be leve...
llm 0.32a3
simonwillison.net·Jun 9, 2026
Simon Willison's Weblog features a recent post announcing the release of "llm 0.32a3," a command-line tool for accessing large language models, primarily developed using Claude Fable 5. The blog also includes articles on various AI-related topics, highlighting advancements and personal insights.
Simon Willison's recent release of llm 0.32a3, largely developed using Claude Fable 5, suggests a significant advancement in integrating large language models with command-line interfaces. This develo...
Setting a custom price for a model in AgentsView
simonwillison.net·Jun 9, 2026
Simon Willison's recent blog post discusses his experience with AgentsView, a Python toolkit for analyzing coding agent transcripts, and how he set custom pricing for the newly released Claude Fable 5 model to track his token usage across projects. He shares insights on his usage data and cost analytics visualizations.
For someone tracking AI coding and productivity tools, a key takeaway from Simon Willison's post is the utility of AgentsView for analyzing token usage across different coding agents. This tool can be...
Quoting Andrej Karpathy
simonwillison.net·Jun 9, 2026
Andrej Karpathy discusses the growing demand for software that can be easily accessed and customized, highlighting the potential of AI to significantly enhance productivity and creativity in software development.
The most valuable insight for you is Andrej Karpathy's observation on the increasing demand for software driven by AI advancements, such as Claude Fable 5. This implies a growing need for professional...
From one-off prompts to workflows: How to use custom agents in GitHub Copilot CLI
github.blog·Jun 9, 2026
Custom agents in GitHub Copilot CLI allow developers to automate and standardize workflows by encoding team-specific context, tools, and standards into reusable Markdown profiles. This enhances efficiency in tasks such as security audits, incident responses, and release documentation, promoting consistency and reducing friction across development environments.
For someone tracking Claude Code, Cursor, Codex, Copilot, and related AI development tools, the most actionable insight from the content is the potential of custom agents in GitHub Copilot CLI. By enc...
Siri AI at WWDC 2026
simonwillison.net·Jun 8, 2026
Simon Willison expresses skepticism about Apple's 2026 WWDC announcements regarding new AI features, emphasizing a cautious approach after previous disappointments. He notes that the new Siri AI capabilities appear feasible with current technology and highlights the integration of a custom Gemini-derived model and Core AI library, which may enhance developer access to Apple's hardware.
The most actionable insight for you is that Apple's new Core AI library integrates with Meta's PyTorch, enabling developers to leverage Apple's hardware effectively for AI model deployment. This integ...
Learn Anything With My /teach Skill
aihero.dev·Jun 8, 2026
The `/teach` skill by Matt Pocock allows users to learn personalized lessons on various subjects, adapting to their goals and levels while tracking progress. It includes steps for setting up a learning project, installing the skill, and creating interactive lessons with resources and quizzes.
For a professional interested in Claude Code and AI dev tools, the key takeaway is the ability to leverage Claude Code as a personalized learning agent through the `/teach` skill. This skill can signi...
datasette-agent-edit 0.1a0
simonwillison.net·Jun 7, 2026
Simon Willison's latest blog post announces the release of `datasette-agent-edit 0.1a0`, a storage-agnostic plugin for Datasette Agent that provides tools for collaborative text editing, including viewing, string replacement, and insertion. The post highlights the complexity of agentic text editing and references a design from the Claude text editor as inspiration for these functionalities.
The most valuable insight for you is the development of the `datasette-agent-edit` plugin, which provides adaptable, storage-agnostic file-editing tools like `view`, `str_replace`, and `insert`. These...
Running Python code in a sandbox with MicroPython and WASM
simonwillison.net·Jun 6, 2026
Simon Willison discusses his new alpha package, `micropython-wasm`, which allows for running Python code in a secure sandbox using MicroPython and WebAssembly. This approach aims to safely execute plugin-style code within applications, addressing concerns about security and resource management in traditional Python environments.
The key insight from Simon Willison's exploration of executing Python code in a sandbox using MicroPython and WebAssembly is the development of a potential solution for safely running plugin-style cod...
Build your own newsroom
Add your own sources, define your topics, and get personalized digests delivered to your inbox.