Nostr notes by Meet

Build a Household Agent You’d Actually Trust This Weekend ...

2026-06-30T01:47:27+02:00

Build a Household Agent You’d Actually Trust This Weekend
Everyone points AI at the enterprise. The build that actually pays off is the one running your own home, and whether you can trust it comes down to five decisions.

quoting
naddr1qv…7vak

This article aims to equip senior technical professionals with the skills to build a reliable AI household agent, emphasizing the operator's role and how AI can support judgment rather than replace it. It highlights that the most valuable AI applications are often unglamorous household management tasks, which can be automated locally. The key to building a trustworthy agent lies in making five critical decisions, transforming a demo into a dependable tool. - The newsletter guides senior technical professionals on using AI and coding agents. - The goal is to keep professionals as operators, not AI assistants, by automating tasks that don't require human judgment. - The article argues that the most valuable AI applications are often overlooked household management tasks, not enterprise solutions. - A household operations agent can be built locally, with the main challenge being establishing trust in the system. - Building a trustworthy agent depends on making five specific decisions, detailed for subscribers. - [ ] https://claudecodefornoncoders.substack.com/p/build-a-household-agent-youd-actually - [ ] https://meet.layer3.press/articles/bad25484-2366-4c95-8290-4c440b021550

Your Family Health Agent: The Whole Build The database, the ...

2026-06-25T14:15:56+02:00

Your Family Health Agent: The Whole Build
The database, the skills, the Telegram bridge, the one script that keeps it alive, and a prompt you can paste into Claude Code to stand up your own this weekend.

quoting
naddr1qv…xkeg

This article details how to build a personal AI health agent using Claude Code, focusing on a local SQLite database, simple English skills for data logging, and a Telegram bridge for mobile access. It emphasizes matching interfaces to tasks, with chat for capturing data and a web app for review, and highlights the importance of maintenance scripts for reliability. The agent acts as an information organizer for medical professionals, not a diagnostician or treater. - The build centers around a single SQLite file for local data storage, eliminating the need for servers or accounts. - Skills are short, plain English instructions that enable the agent to perform specific tasks like logging workouts and sleep data. - A Message-Content-Platform (MCP) bridge, specifically Telegram, allows the agent to interact outside the chat window via a mobile phone. - The system uses two interfaces: Telegram for quick data capture and a small web app for reviewing trends. - A crucial design decision was separating data capture (chat) from data review (web app) to reduce friction and ensure data completeness. - A vital maintenance script in the shell startup file prevents stale messaging-plugin processes from causing message loss. - The agent's role is to maintain an organized health record, supporting medical professionals rather than diagnosing or treating. - Two methods are provided to start: cloning a starter template from GitHub or building from scratch using a detailed prompt in Claude Code. - [ ] https://claudecodefornoncoders.substack.com/p/your-family-health-agent-the-whole - [ ] https://meet.layer3.press/articles/c930b177-a178-498c-9e50-e49e818afe19

Just got this email https://t.co/bD0b8diezt #naddr1qv…xjyr

2026-06-25T01:09:35+02:00

Just got this email https://t.co/bD0b8diezt

quoting
naddr1qv…xjyr

Just got this email https://t.co/bD0b8diezt - [ ] https://x.com/jeremyphoward/status/2069920905162359188 - [ ] https://meet.layer3.press/articles/019efe10-259c-2e9a-70ef-33fb687ec579

This is a new paradigm for interacting with Claude that is ...

2026-06-24T00:26:31+02:00

This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads.

quoting
naddr1qv…u4tu

This is a new paradigm for interacting with Claude that is significantly more "inline" with all the other human activity org-wide. Once you do all of the under the hood engineering work to make this "just work" (e.g. across tools, integrations, compute environments, memory, security, etc.), Claude basically joins the team in a seamless way - you can talk to it as you would talk to a person and it can help with a very large variety of workloads.

Imo this is the 3rd major redesign of LLM UIUX. The first paradigm was that the LLM is a website you go to, the second was that it is an app you download to your computer. This third one is that it is a self-contained, persistent, asynchronous entity with org-wide tools and context, working alongside teams of humans. It really takes a while to wrap your head around it, but it works and it is awesome.

@claudeai: Introducing Claude Tag, a new way for teams to work with Claude.

In Slack, Claude joins as a team member with access to the channels and tools you choose. Tag Claude in and delegate tasks to it while you focus on other work. https://t.co/R2C6A5Kcye

- [ ] https://x.com/karpathy/status/2069547676849557725 - [ ] https://meet.layer3.press/articles/019efe10-2f37-2afd-7305-04960c04c4e8

Nice to see the Drivetrain Approach getting some love again! :D ...

2026-06-23T10:41:09+02:00

Nice to see the Drivetrain Approach getting some love again! :D

quoting
naddr1qv…6gdj

Nice to see the Drivetrain Approach getting some love again! :D https://t.co/XcOjrJ0As3

@ForBo7_: Did a close read of @jeremyphoward et al's Drivetrain Approach

Define goal → Consider levers → Consider data → Implement assembly line

Most important is having actions that fulfill the objective

Predictions are more data, not actions

Key takeaways 🧵

[ ] https://x.com/jeremyphoward/status/2069339966380052725

[ ] https://meet.layer3.press/articles/019efe10-259d-31d8-72ae-2383d5b0ffd9

Weekend Dispatch — June 20, 2026: Three Forces Run Your AI. ...

2026-06-20T14:15:29+02:00

Weekend Dispatch — June 20, 2026: Three Forces Run Your AI. None Is the Model.
Meta trained agents on its own workers. Open weights crossed the frontier but won’t run on your laptop. The war went to the model nobody trusts.

quoting
naddr1qv…y34x

Meta captured employee work data to train AI agents, sparking backlash and program adjustments. Advanced open-weight AI models are now available but are too large to run on personal devices, requiring specialized infrastructure knowledge. Government decisions are dictating AI model accessibility and usage, as seen with Grok's military deployment and Anthropic's restriction. - Meta's 'Model Capability Initiative' trained AI agents by recording employee computer activity, which was later scaled back due to employee concerns. - New open-weight AI models like MiniMax M3 and Zhipu's GLM-5.2 show competitive performance but require extensive infrastructure to operate, limiting true user sovereignty. - Grok was reportedly used by the Pentagon for military targeting, while Anthropic refused military use of its models, leading to government sanctions against the company. - Government decisions, not model benchmarks, are determining which AI models are accessible and deployed, highlighting a layer of control beyond the models themselves. - [ ] https://claudecodefornoncoders.substack.com/p/weekend-dispatch-june-20-2026-three - [ ] https://meet.layer3.press/articles/efc6e322-f0eb-49fb-b66c-e19a6f01556a

I Built My Family an Agentic Health Record. You Can Too. You ...

2026-06-18T14:15:25+02:00

I Built My Family an Agentic Health Record. You Can Too.
You forget most of what your doctor tells you. I built my family a system that remembers, over a weekend, and it put us in the best shape of our lives.

quoting
naddr1qv…zfnv

The author created an AI agent, 'Othellobot,' using accessible tools like Telegram and Claude, to meticulously track his family's health data, including workouts, sleep, medications, and symptoms. This system transformed their health by enabling consistency and providing concrete information for doctor visits, moving them from passive patients to informed partners.

The agent's primary function is to keep data honest and organized, not to diagnose or treat, empowering medical professionals with real information.

The interface evolved from a complex app to a simple conversational agent, demonstrating that matching the interface to the job is key for effective data capture and interaction. - A custom AI agent, 'Othellobot,' was built to track family health data, including training, sleep, medications, and symptoms. - The system promotes consistency in health habits and provides a clear record for doctor visits, improving patient-doctor partnerships. - The agent's role is data organization and honest record-keeping, not medical diagnosis or treatment. - The interface shifted from a feature-rich app to a conversational agent via Telegram for easier data capture. - Building such a system is now accessible, allowing individuals to create personalized health companions without relying on expensive services or third-party companies. - [ ] https://claudecodefornoncoders.substack.com/p/i-built-my-family-an-agentic-health - [ ] https://meet.layer3.press/articles/c43fb9f8-2cf5-4659-a2ae-86a364ff6ec6

Weekend Dispatch: Architect or Be Architected Everyone’s ...

2026-05-16T14:15:53+02:00

Weekend Dispatch: Architect or Be Architected
Everyone’s reading this week’s four AI stories wrong: the FDE rebrand, Karpathy’s compression test, the brake Citrini missed, and Nvidia on your wall.

quoting
naddr1qv…zfnu

Companies like OpenAI, Anthropic, and Google are investing heavily in 'Forward Deployed Engineers' to embed AI into businesses, as platforms increasingly adopt a headless approach, allowing external AI agents to function within them. The economic implications of AI adoption are debated, with concerns that widespread job displacement could lead to consumption collapse, acting as a natural brake on AI investment, unless alternative reabsorption mechanisms for workers are established. Nvidia's strategy of deploying compute infrastructure on residential properties mirrors historical resource plays, raising questions about ownership and control in the evolving AI landscape. - AI companies are investing billions in 'Forward Deployed Engineers' (FDEs) to integrate AI into businesses, a role previously filled by consultants. - SaaS platforms are becoming 'headless,' allowing users to interact via their preferred AI agents rather than platform-specific interfaces. - A key AI economic debate centers on whether job displacement will lead to a consumption collapse, acting as a 'brake' on AI investment. - Nvidia is decentralizing AI compute by installing GPU nodes on residential properties, enabling homeowners to provide infrastructure for compute revenue. - The article encourages individuals and businesses to focus on building skills for the 'third loop' of AI integration – defining and architecting outcomes – to remain relevant. - [ ] https://claudecodefornoncoders.substack.com/p/weekend-dispatch-architect-or-be - [ ] https://meet.layer3.press/articles/6bc24d46-5e6d-4e94-99fe-5c7b19353563

You’re Not Being Hired Strictly for Output Anymore. Your ...

2026-05-14T14:15:45+02:00

You’re Not Being Hired Strictly for Output Anymore. Your Judgment Is What Is in Demand Now.
Companies are running Agile retrospectives while the production function underneath them has changed. The work that built your career was output. The work that holds it now is judgment.

quoting
naddr1qv…xafx

The rise of AI has fundamentally changed the software development landscape, moving the primary constraint from execution speed to the quality of judgment about what work is worth doing. Traditional Agile methodologies optimized for velocity and output are becoming obsolete as AI drastically reduces the cost of producing artifacts. Companies need to re-evaluate hiring and performance metrics to prioritize critical thinking and decision-making skills. - The AI era has shifted the bottleneck from execution to judgment and decision-making. - Traditional methodologies like Agile, focused on output and velocity, are becoming outdated. - AI agents can produce code, specs, and reports cheaply, making output abundant. - The new scarce resource is the judgment required to decide which tasks are valuable and strategic. - Hiring and performance evaluations should shift from measuring output to assessing judgment quality. - The 'feature contract' artifact emphasizes upfront judgment, defining the 'why' and 'how' of success before execution. - Companies that adapt to valuing judgment will have a competitive advantage in hiring and performance. - [ ] https://claudecodefornoncoders.substack.com/p/youre-not-being-hired-strictly-for - [ ] https://meet.layer3.press/articles/7ebcce83-f4cf-499c-a0b1-20e51f9ad84d

You Can’t Prompt-Engineer Your Way to Reliable Agents. Wire the ...

2026-05-12T14:15:56+02:00

You Can’t Prompt-Engineer Your Way to Reliable Agents. Wire the Architecture Around It.
Domain 1 Capstone: Reliability is not in the prompt. It’s in the loop, the hooks, the policy gates that fire whether the model wants them to or not, and that the model can’t talk past.

quoting
naddr1qv…9u76

This article argues that AI agent reliability stems not from better prompting, but from the surrounding architecture, including execution loops, pre- and post-tool hooks, and unbypassable policy gates. A case study demonstrates how these architectural elements mechanically enforce a refund policy and normalize data, preventing the AI from violating rules even when instructed to do so.

The core thesis is that prompts ask, but architecture enforces, highlighting that structural components like hooks operate independently of the AI's reasoning and conversation.

The article details the practical implementation of these architectural concepts using Claude Code, including the role of the system prompt, tool definitions, and crucial hooks like normalize_dates.py and refund_policy_gate.py. - Reliability in AI agents comes from architecture (loops, hooks, gates), not just prompting. - A PreToolUse hook prevented an AI from issuing an unauthorized refund by enforcing a $500 limit and manager approval rule. - A PostToolUse hook normalized inconsistent date formats from various data sources before the AI processed them. - Architectural components operate independently of the AI's reasoning, providing a robust enforcement mechanism. - The article emphasizes that prompts request actions, while architecture guarantees enforcement. - [ ] https://claudecodefornoncoders.substack.com/p/you-cant-prompt-engineer-your-way - [ ] https://meet.layer3.press/articles/6a20fe1c-8f6b-492f-8645-bae2dbd1570a

New supply chain attack this time for npm axios, the most popular ...

2026-03-31T07:23:32+02:00

New supply chain attack this time for npm axios, the most popular HTTP client library with 300M weekly downloads.

quoting
naddr1qv…xhhs

New supply chain attack this time for npm axios, the most popular HTTP client library with 300M weekly downloads.

Scanning my system I found a use imported from googleworkspace/cli from a few days ago when I was experimenting with gmail/gcal cli. The installed version (luckily) resolved to an unaffected 1.13.5, but the project dependency is not pinned, meaning that if I did this earlier today the code would have resolved to latest and I'd be pwned.

It's possible to personally defend against these to some extent with local settings e.g. release-age constraints, or containers or etc, but I think ultimately the defaults of package management projects (pip, npm etc) have to change so that a single infection (usually luckily fairly temporary in nature due to security scanning) does not spread through users at random and at scale via unpinned dependencies.

More comprehensive article: https://t.co/EJAZbqAPIQ

@feross: 🚨 CRITICAL: Active supply chain attack on axios -- one of npm's most depended-on packages.

The latest axios@1.14.1 now pulls in plain-crypto-js@4.2.1, a package that did not exist before today. This is a live compromise.

This is textbook supply chain installer malware. axios has 100M+ weekly downloads. Every npm install pulling the latest version is potentially compromised right now.

Socket AI analysis confirms this is malware. plain-crypto-js is an obfuscated dropper/loader that:

• Deobfuscates embedded payloads and operational strings at runtime • Dynamically loads fs, os, and execSync to evade static analysis • Executes decoded shell commands • Stages and copies payload files into OS temp and Windows ProgramData directories • Deletes and renames artifacts post-execution to destroy forensic evidence

If you use axios, pin your version immediately and audit your lockfiles. Do not upgrade.

[ ] https://x.com/karpathy/status/2038849654423798197

[ ] https://meet.layer3.press/articles/019efe10-2f3b-224a-7340-02ed0f953b86