TAXai
Morning AI Digest · Fri, Jun 5 2026

15 new videos worth your attention

last 72h · 38 scanned across 50 channels
💡 Claude Tips & Hacks · 9
Automation & Agents 8
Nate Herk · Thu Jun 4, 17:55 UTC 7/10

Use a 'Grill Me' Prompt to Extract Your Brain Into Claude

  • Instead of dumping your knowledge into Claude all at once (which leaves gaps), this 'Grill Me' technique has Claude interview you with targeted follow-up questions until it fully understands a process or decision
  • After each question, Claude saves your answers to a living document so nothing is lost — even in sessions that run an hour or more
  • At the end of a session, Claude flags anything you couldn't answer well, so you know exactly who on your team to ask next
  • The finished document becomes a reusable reference — you can come back later and say 'grill me again' to update it when things change
  • The core prompt is just 4-5 sentences — the 'skill' is really just a saved prompt you trigger with a slash command inside Claude Code
The idea: Copy the free 'Grill Me' prompt template from the creator's community, paste it into a Claude Code project as a saved skill, then trigger it with a slash command whenever you want to document a business process, service offering, or client workflow.
Why it matters: Replaces 30-60 minutes of scattered brain-dumping with a structured interview that produces a clean, reusable document — useful for onboarding, scoping client projects, or building internal SOPs.
Effort: Low-code (~1h)
Watch → · claude code · prompting · knowledge management · workflow · consulting · sops
Julian Goldie SEO · Thu Jun 4, 17:51 UTC 7/10

NotebookLM Now Turns Any Document Into Podcasts, Videos, and Slides

  • Drop in PDFs, YouTube links, or notes and NotebookLM generates a podcast, narrated video, infographic, slide deck, or quiz — each in under a minute
  • New 'interactive mode' lets you pause the AI-generated podcast and ask questions out loud, getting answers pulled directly from your source material
  • Custom instructions (up to 10,000 characters) let you set a consistent brand voice so every output — audio, slides, video — sounds like it came from your team
  • Data buried in reports or PDFs can be extracted into a clean table and sent straight to Google Sheets, skipping hours of manual copy-paste work
  • Everything stays grounded in sources you provide — NotebookLM won't hallucinate from the internet, so outputs are citable and trustworthy
The idea: Load your existing content — client guides, YouTube scripts, training docs, newsletters — into a NotebookLM notebook with custom instructions set to your brand voice, then use it to auto-generate podcasts, infographics, and slide decks for repurposing across channels.
Why it matters: Turns one piece of content into five distributable formats in minutes instead of a full production day, freeing your team from repetitive content reformatting work.
Effort: No-code (~15 min)
Watch → · notebooklm · content repurposing · no-code · productivity · google · podcasting
Greg Isenberg · Thu Jun 4, 18:11 UTC 6/10

OpenAI Codex Sites Lets You Build Self-Updating Internal Apps With Chat Commands

  • Codex Sites is a tool inside ChatGPT's Codex that lets you build small apps and internal tools — think a kanban board, a tracker, or a dashboard — using plain-English prompts
  • The big differentiator: once the app is built, you can tell Codex in a new chat to update the app ('add this idea to my board') and it does it live, without you touching the app itself
  • You have to prompt it to add memory (so data saves between visits), set up 'safe actions' (approved operations the AI can perform, like add/move/delete), and create 'skills' (reusable instruction sets so future chats know how to operate the app)
  • Current limits: no custom domains yet (you get an ugly URL), no built-in payments or email, and it's more hands-on than tools like Lovable or Replit — but the autonomous-update angle is genuinely new
  • Best for teams that already live in ChatGPT/Codex and want a simple internal tool that an AI agent can keep updated automatically
The idea: Use Codex Sites inside ChatGPT to build a startup idea tracker (or any internal kanban-style board) with persistent storage, then teach it a 'skill' so you can add or update cards just by typing a command in a new Codex chat.
Why it matters: Replaces a manual Notion board that you have to update yourself — instead, you or an AI agent can add, score, and move ideas just by sending a chat message, saving 5–10 minutes of manual data entry per idea.
Effort: Needs a developer
Watch → · codex · app-building · internal-tools · automation · no-code · agents
David Ondrej · Thu Jun 4, 14:14 UTC 6/10

Ex-Anthropic founder demos how he builds entire products using Codex daily

  • Pedro (founder of MagicPath, ex-Anthropic) switched from Claude Code to Codex 5 months ago — says Codex uses fewer tokens for the same work and has a tighter agentic loop
  • His biggest productivity trick: text expansion shortcuts on Mac/iPhone that paste long, detailed prompts into Codex instantly — so he never types the same instruction twice
  • He uses a 'plan with the smart model, build with the cheap model' workflow: get the expensive model to write a plan, then switch to a cheaper model to execute it
  • His startup advice: build for agents, not users — meaning your product should be accessible via API or SDK so AI agents can use it, not just humans clicking through a UI
  • Demo-first culture: he says a 40-second Screen Studio video of a working prototype on Twitter/X consistently outperforms long launches — half a million views on one demo
The idea: Set up Mac text expansion shortcuts (in System Settings → Keyboard → Text Replacements) with your most-used Codex or Claude prompts — for example, a shortcut that expands into a full 'review this code for bugs and edge cases' prompt — so you can trigger detailed instructions with 2-3 keystrokes in any AI tool.
Why it matters: Cuts the time spent retyping the same detailed prompts every day to near zero, and since shortcuts sync to iPhone, you can fire off complex agent instructions from your phone while away from your desk.
Effort: No-code (~15 min)
Watch → · codex · ai agents · productivity · workflow · startups · demo strategy · transcript truncated
Julian Goldie SEO · Fri Jun 5, 01:00 UTC 5/10

Talk to Your AI Out Loud and Get Real Work Done Hands-Free

  • Hermes is a free, locally-run AI agent you can talk to out loud — it listens, responds with a voice, and actually completes tasks rather than just answering questions
  • You can swap the AI 'brain' behind it — the presenter uses MiniMax M3, a new model that handles images, video, and massive amounts of context
  • It connects to Telegram, Slack, WhatsApp, and Discord — send a voice note, it does the work, talks back
  • Hundreds of free voice options including accents and languages, with the option to run everything locally so your files never leave your machine
  • Demonstrated use cases: brainstorming content ideas while walking, planning onboarding flows, drafting a week of tutorials — all by speaking out loud
The idea: Set up the Hermes voice agent on your computer (no subscription required) with a free AI model like MiniMax M3, then connect it to a team chat tool like Slack or WhatsApp so you and your team can delegate tasks by voice note.
Why it matters: Lets you capture and act on ideas while away from your desk — useful for busy consultants who think on the move and want to avoid losing ideas or spending time typing up briefs.
Effort: Needs a developer
Watch → · voice-ai · productivity · local-ai · agent · no-typing · content-workflow
Dan Martell · Thu Jun 4, 13:00 UTC 5/10

Map Your Business Bottleneck First, Then Point AI at It

  • Most people use AI randomly — the real move is to identify the one stage (leads, sales, delivery, retention) choking your revenue, then automate that first
  • The '10-80-10 rule': you define the goal (10%), AI does the heavy lifting (80%), you add your human touch and review (10%)
  • AI tools like Your Atlas and Revieo can qualify leads, book calls, and follow up automatically — so you only talk to ready buyers
  • The 'Camcorder Method': record yourself doing a task on Zoom, feed the transcript and screenshots to AI, and let it learn to replicate your process
  • The pitch is directionally correct but this video is mostly a framework — specific tool recommendations are light and the playbook is gated behind a DM
The idea: Use Make.com or Zapier to connect a lead form to an AI qualifier (like Your Atlas or a Claude-powered chatbot), so new leads are automatically scored and only the best ones land in your calendar as booked calls.
Why it matters: Stops your team from wasting hours on tire-kicker calls by filtering leads before any human gets involved — potentially saving 5-10 hours of sales calls per week.
Effort: Low-code (~1h)
Watch → · sales automation · lead qualification · business operations · ai workflow · no-code · bottleneck analysis
Julian Goldie SEO · Thu Jun 4, 07:44 UTC 5/10

Run Claude Code's Full Power Free Using Google's API

  • Claude Code is Anthropic's most powerful AI coding agent — it reads your whole project, plans changes, writes code, and fixes errors from a plain-English instruction
  • The trick: install a free 'router' tool that redirects Claude Code away from Anthropic's paid API and points it at Google's free Gemini API instead — no credit card needed
  • A CLAUDE.md file in your project folder acts like a permanent briefing document — Claude reads it every session so you never re-explain your setup, naming rules, or workflows
  • Useful slash commands inside Claude Code: /init sets up your CLAUDE.md, /compact summarizes long sessions to save your free quota, and Plan Mode lets you review what Claude will do before it does anything
  • The video is partly a sales pitch for a paid community called AI Profit Boardroom — the free setup instructions are real, but expect upsells throughout
The idea: Have a developer set up Claude Code with the Claude Code Router pointed at Google AI Studio's free Gemini API, then create a CLAUDE.md file in each client project folder describing your team's workflows, tools, and conventions so every Claude session starts fully briefed.
Why it matters: Your team stops re-explaining project context to AI every session, and gets access to a powerful coding agent at zero ongoing cost — useful if you're delegating any build or automation work.
Effort: Needs a developer
Watch → · claude code · free tier · coding agent · gemini api · claude.md · no subscription
Julian Goldie SEO · Thu Jun 4, 16:44 UTC 4/10

Free AI Agent Combo Claims to Run Your Content Business Automatically

  • Minimax M3 is a free Chinese AI model built for multi-step autonomous tasks — not just chat — claiming up to 12 hours of unsupervised operation
  • Hermes Agent is a free tool that lets the AI actually control your computer: open apps, create files, browse the web, and schedule recurring tasks
  • The pitch: set up one workflow and the agent handles weekly content planning, community monitoring, and lead research on autopilot
  • A cloud-hosted version (Max Hermes) means the agent runs 24/7 without needing anyone's laptop to be on — useful for team operations
  • Both tools are free and open-source, but the setup still requires terminal commands and local software installs — not a click-and-go experience
The idea: Ask a developer to set up Hermes Agent connected to Minimax M3 on a cloud server, then configure recurring tasks — like pulling top community questions each morning and dropping a summary doc into a shared Google Drive or Notion page — so your team arrives to a ready-made daily brief.
Why it matters: Replaces the daily manual work of monitoring community discussions and researching content topics, potentially saving 1-2 hours of prep work per team member each morning.
Effort: Needs a developer
Watch → · ai agents · content automation · no-code workflow · community management · lead research · open source
Coding with AI 2
Nuno Tavares · Thu Jun 4, 16:00 UTC ★ 8/10

GoHighLevel Now Lets an AI Agent Run Your Entire Follow-Up Workflow

  • GoHighLevel added a built-in AI agent step you can drop into any automation — it handles tagging, creating opportunities, sending emails and texts, and generating booking links, all from a single prompt
  • Pre-built prompt templates cover the most common scenarios: new form leads, Facebook leads, no-show appointments, Instagram DMs, and more — so you're not starting from scratch
  • The presenter ran a live test: someone filled out his website contact form, and the agent automatically sent a personalized SMS and email with a booking link for about 2 cents total
  • You can copy the default prompt, paste it into Claude or ChatGPT (which already knows your business details), ask it to personalize the prompt for your brand, then paste the result back into GoHighLevel
  • The feature is still in beta and costs a small premium per action run, but at 1–2 cents per execution it's cheap enough to experiment with immediately
The idea: In your GoHighLevel account, create a new workflow, add your trigger (form submitted, appointment no-show, or Instagram DM), then add the new AI Agent step — pick a pre-built template, paste in a Claude-personalized prompt, and publish.
Why it matters: Replaces manual multi-step follow-up sequences with a single AI step that tags the lead, creates the opportunity, and sends a personalized message automatically — turning a 30-minute automation build into a 15-minute setup.
Effort: Low-code (~1h)
Watch → · gohighlevel · automation · ai-agents · crm · lead-follow-up · no-code
All About AI · Thu Jun 4, 17:00 UTC 4/10

AI agent trades crypto autonomously and switches strategy mid-session

  • The creator used an AI coding agent (Codex) to collect market data, propose three trading strategies, pick the best one, and then execute trades on a crypto platform — all with minimal manual input
  • The 'agentic' part: a monitoring agent checked the trade every 2 minutes, noticed the market shifted, scrapped the original short strategy, and switched to a long position — earning ~$7 in under an hour
  • Tools used: Hyperliquid (crypto exchange), Codex (OpenAI's coding agent), and a beginner instruction file the creator is sharing publicly
  • Claude Code is mentioned as a direct alternative to Codex for the same setup — same idea, different agent
  • Heavy caveat: this is a demo with $10 at risk, not a proven money-making system — the creator is clear it's a hobby/learning project, not a business
The idea: Hire a developer to set up a similar agentic trading loop using Claude Code connected to a crypto exchange API, where the agent monitors positions on a timer and adjusts strategy automatically based on live market data.
Why it matters: For someone curious about AI automation, this is one of the clearest real-world examples of an AI agent making decisions and acting autonomously over time — the same pattern could apply to monitoring dashboards, alerts, or other business data loops, not just trading.
Effort: Needs a developer
Watch → · ai agents · crypto · trading · automation · claude code · autonomous agents
News & Capabilities 5
OpenAI · Thu Jun 4, 18:06 UTC 7/10

OpenAI Codex Turns Design Ideas Into Clickable Prototypes in Minutes

  • Describe a feature idea, upload a visual reference, and the AI generates three different design directions to choose from — no design tool needed to start
  • Once you pick a direction, it builds a fully clickable, interactive prototype you can test in your browser, toggling features and scrolling around like a real app
  • You can annotate the prototype with feedback and ask for changes conversationally, without redoing anything from scratch
  • One click sends the finished design into Figma — including the prototype, user story context, and critique notes, not just a screenshot
  • The 'Sites' feature publishes the prototype as a live URL so your whole team can click through it without needing any accounts or installs
The idea: Use the OpenAI Product Design plugin inside Codex to go from a written brief (and optional reference image) to a shareable, interactive prototype — then push it to Figma or publish it as a live link for your team to review.
Why it matters: Cuts the time from 'I have an idea' to 'the team can click through it' from days of back-and-forth with a designer down to under an hour, ideal for presenting new service offerings or internal tool concepts to clients.
Effort: Low-code (~1h)
Watch → · prototyping · design · openai · codex · figma · no-code
The AI Daily Brief · Thu Jun 4, 12:30 UTC 7/10

Enterprise AI shifts from hype to cost control and smarter workflows

  • OpenAI's Codex app now has 5 million weekly users — and the fastest-growing group is non-technical knowledge workers, not developers
  • New Codex features let you highlight parts of a document to discuss, access pre-built role bundles for sales, analytics, and more, and turn any output into a shareable website — no downloads required
  • Microsoft released a family of in-house AI models optimized for cost, claiming 10x cheaper than GPT-5 for some enterprise tasks when tuned for a specific company
  • Uber capped employee AI spending at $1,500/month — a sign that 'token costs' are becoming a real budget line for companies using AI at scale
  • A US executive order now encourages (but doesn't require) AI labs to share advanced models with the government 30 days before public release — mostly formalizing what was already happening voluntarily
The idea: Try OpenAI Codex's new 'Sites' feature to turn a report, dashboard, or budget tracker you've built in Codex into a shareable web link your team can interact with — no file attachments or downloads needed.
Why it matters: Replaces static PDFs and slide decks with live, updatable pages you can share instantly, cutting the back-and-forth of version control and file sharing for internal team deliverables.
Effort: No-code (~15 min)
Watch → · enterprise ai · codex · openai · microsoft · cost management · knowledge work
OpenAI · Thu Jun 4, 17:00 UTC 6/10

OpenAI's reasoning model solved an 80-year-old unsolved math problem

  • An OpenAI reasoning model disproved a famous geometry conjecture posed by mathematician Paul Erdős in the 1940s — something mathematicians have tried and failed to do for 80 years.
  • The model wasn't trained on math specifically — it's the same general-purpose model used for coding and chat, just given more time to 'think' before answering.
  • The key insight: giving the model longer to reason (called 'test time compute') dramatically improves accuracy — more thinking time = more correct answers, even on problems at the edge of human knowledge.
  • Mathematicians reviewing the proof then used the AI's ideas to knock down additional open problems, showing AI as a research catalyst, not just a solver.
  • Practical advice from the team: get a ChatGPT Pro subscription, ask the boldest question you can, and don't break problems into smaller pieces first — the model often surprises you with a better path than you'd assume.
Watch → · reasoning · openai · math · research · ai capabilities · science
OpenAI · Thu Jun 4, 20:25 UTC 5/10

Zapier engineers use Codex to write full project tickets in hours, not weeks

  • A Zapier engineer shows how OpenAI Codex acts as a research-and-writing layer across Slack, Google Docs, Coda, and Jira simultaneously
  • Tasks that used to take weeks of gathering context from multiple tools now produce a full project brief in hours
  • The workflow pulls scattered information together automatically and outputs structured Jira tickets with detailed instructions
  • This is a professional services / knowledge-work use case, not just a coding trick — relevant for any team juggling multiple project-management tools
  • Video is a short promotional testimonial — no step-by-step setup shown
The idea: Use Make.com or Zapier with an OpenAI module to draft detailed project briefs or client scope documents by pulling notes from Google Docs, Slack messages, and Notion pages into one structured output.
Why it matters: Cuts the time to write a full client project scope from several hours of copy-pasting and summarising down to a single automated run — useful before every new engagement.
Effort: Low-code (~1h)
Watch → · project management · automation · jira · zapier · codex · scope writing
Wes Roth · Thu Jun 4, 03:20 UTC 5/10

Microsoft Launches Its Own AI Models and Bets on Custom Business Agents

  • Microsoft quietly built its own family of AI models (called MAI) in just 6 months — deliberately designed to trail the bleeding edge by a few months to keep costs way down
  • Their big idea: instead of renting a generic AI, businesses train a custom version of a Microsoft model on their own workflows, making it cheaper and harder to replace than switching to the latest ChatGPT
  • Microsoft is embedding AI agents directly into Windows so they can run tasks autonomously in the background — think a 24/7 assistant inside Teams, Outlook, OneDrive, and Calendar
  • The new 'Scout' product is an always-on autopilot agent built on OpenAI's open-source agent technology and integrated into Microsoft 365
  • OpenAI and other major AI agent platforms are now launch partners on Windows, meaning the tools your team already uses are about to get a lot more autonomous
Watch → · microsoft · ai models · enterprise ai · ai agents · windows · strategy