Wiki
Tags
About
FAQ
RSS
AIPOST

AI in Practice, page 2

How to put AI tools to work, on the job and in everyday life

Earlier articles, page 2

Earlier articles in AI in Practice, newest first.

Four Hooks That Let TypeScript Mods Rewire Claude's Coding Workflow

Claude Mods let TypeScript code change Claude Code's interface and tool behavior. Mods hook in before, instead of, after, or around an action. Prompt cache cuts input cost about 95 percent but expires after an idle hour. Have Claude audit your last 30 sessions to suggest mods that fit your habits.

AI in Practice

Headlight Bulb Directory Built by an AI Agent: Data First, Pages Second

Claude Code built a roughly 1,140-page headlight bulb directory from public data. Five days of scraping covered 8,462 vehicles and 97.2% of target search volume. Government data and owner manuals replaced a paid data license. Fill in the vehicles people search for first and audit samples for errors. Wait one to two weeks for indexing before building links.

AI in Practice

OpenAI's Dot: How Sam Altman Sorts Urgent Work and Tests Ideas

Dot filters urgent work so Altman can protect time for creative thinking. Voice notes produced five or six working feature versions overnight. Cross-app context helped his agent recover information hidden in an image. Use agents for triage and prototypes while keeping key judgments human.

AI in Practice

Hermes Agent Setup: Connecting Tools and Coordinating Roles

Get one Hermes Agent assistant working before adding specialized profiles. Claude connects through an Anthropic API key, WhatsApp and Gmail separately. Kanban holds drafting until research is ready, then routes a review. Keep a person approving outbound drafts and test scheduled jobs manually.

AI in Practice

Meta Muse Handles Chores but Hits Checkout and Login Barriers

Meta Muse completed calls and canceled subscriptions in a several-week trial. Duplicate subscription cancellations saved one user over $40 a month. Bot checks and sign-ins needed human help; Amazon blocked checkout. Keep sensitive work email separate from accounts connected to an AI agent.

AI in Practice

Meta Muse: Shopping, Audio, Business Calls and Remote Files

Meta Muse combines shopping, audio briefings, calls and remote file edits. Daily research can become a podcast on a recurring schedule. US business calls require approval and return transcripts, not live audio. Remote file edits warrant a separate computer with restricted access. Referrals award 1 billion tokens to each person, up to 30 redemptions.

AI in Practice

Smart Glasses Bring Voice Input to Existing Claude Code Sessions

A self-built setup, not a built-in feature, mirrors Claude Code to smart glasses. Whisper transcribes speech locally on the Mac, with no audio sent to the cloud. A controller button sends the reviewed text to the running Claude Code session. Use the glasses for text, a phone for color images and a desktop for complex coding.

AI in Practice

AI Agent Code: Why Green Tests Need Independent Checks

Passing agent-written tests may only confirm the agent's own assumptions. In one case, acceptance tests caught defects missed by over 12,000 green unit tests. Those acceptance tests flagged defects in 80 of 2,000 runs. One comparison favored Playwright CLI for suite runs and MCP for exploration. Add independent checks and use mutation testing to see if tests catch faults.

AI in Practice

Claude Code Builds a Video App Through Planning and Testing

Claude Code helped assemble upload, playback, and creator tools in one app. An eight-milestone plan defined routes, data rules, and media work. The creator dashboard displayed metrics from mock data. Check access rules, authentication, and real metrics before deployment.

AI in Practice

AI Coding Agent Projects: Choosing From 29 Tools by Site and App Type

An AI coding agent writes the app; the services around it still have to fit. 29 tools in 14 categories form a menu, not a list to install in full. Pick tools that start cheap, scale to production and let an agent operate them. Content sites: Astro on Cloudflare Pages; interactive web apps: Next.js on Vercel. Add databases, login, email and payments only when the app needs them.

AI in Practice

HarnessRouter Unifies AI Agent Workflows on a Local Console

Free, open-source HarnessRouter runs AI agent harnesses from one local console. Project benchmark: same model, four harnesses, 77% to 85% task success. Gemini CLI and Pi both caught all seven planted issues in a 34-entry expense test. Requires Docker Desktop and your own model provider API key. Start with a file whose answers you know, then compare harnesses and audit trails.

AI in Practice

Get new articles by RSS or podcast

Add the RSS URL to your RSS reader or the podcast URL to your podcast app to get every new article as soon as it is published. The podcast reads out the headline and key points.

How to subscribe
AIPOST
© 2026 AIPOST. All rights reserved.

AIPOST is an AI publication covering practical AI, AI security, performance, startups, health, ethics and industry news. No account is needed, and our privacy policy explains how we handle personal information.