You don’t need to work eight hours a day to get eight hours of work done if you have the right intelligent infrastructure in place. I recently came across a breakdown of a daily workflow that completely redefines what personal productivity looks like in the age of large language models. This incredible deep dive comes from Matt Wolfe, a leading voice in the AI space who just pulled back the curtain on his entire software stack.
The premise here is simple but powerful: stop using AI just for novelty and start using it to eliminate friction. The expert explains that by chaining specific tools together, he has reduced his necessary active working time to just a couple of hours a day while maintaining a high output of videos, newsletters, and code. This isn’t about using one chatbot for everything, it’s about selecting specialized agents for research, coding, content creation, and administrative grunt work.
📌 The Research and Browsing Engine
The foundation of this workflow is a complete shift away from traditional search engines. The creator reveals that Perplexity has effectively replaced Google for him. Rather than sifting through blue links, he uses it to get direct answers for everything from camera settings to sports scores. While he uses the Pro plan to toggle between models like Claude 3.5 and GPT-4o, he notes that the free version provides nearly the same utility for general search.
However, the real power move here is the browser itself. He utilizes Comet, a browser that integrates Perplexity directly into the interface. This allows for a feature he calls “Slash Commands” within the browser’s sidebar assistant. Instead of typing out long, repetitive prompts every time he reads a news article, he simply types /newsreview.
- How it works: He pre-programmed a prompt that asks, “Break this article down, is there anything newsworthy, and what are the implications for non-technical consumers?”
- The benefit: This instant context switching saves massive amounts of mental energy. He also uses a /commentreview command to filter through YouTube comments, ignoring trolls and highlighting constructive feedback automatically.
To keep the information pipeline full, he relies on Feedly, but with a twist. He uses their “AI Web Alerts” feature. This isn’t just an RSS reader; it’s an AI agent that actively scans the web for keywords in publications he does not subscribe to. It learns over time, allowing him to train it by saying “show me less like this,” effectively building a personalized news curator that runs 24/7.
💡 The Creative Brain and Custom Software
When it comes to actual creation, the expert has a clear preference for Claude over ChatGPT. He argues that Claude’s tone is less sycophantic and more direct, avoiding the fluffy “That’s a great question!” filler that plagues other models. He utilizes Claude’s “Projects” feature to create a digital “YouTube Producer.”
This is a fascinating application of data analysis. He exports his real YouTube analytics (views, retention, subscriber growth) as CSV files and uploads them into a Claude Project. This gives the AI a persistent memory of what performs well on his channel. When he brainstorms video ideas, the AI references his actual historical data to suggest titles and angles that are statistically likely to succeed.
But the most impressive part of this stack is the use of Cursor. He isn’t just writing code; he is building bespoke micro-tools to solve his own problems.
- The Command Center: He built a personal dashboard that pulls API data from his newsletter, social media, and revenue sources into a single view.
- The Thumbnail Generator: He fed an AI his headshots and successful thumbnail examples to create a tool that generates on-brand images automatically.
- The Video Titler: A custom script where he drops in an MP4, and the tool transcribes it and suggests titles based on the content.
He pairs this coding workflow with Wisper Flow, a voice-to-text tool that runs in the background. Unlike standard dictation, this tool cleans up the audio in real-time, removing “ums,” “ahs,” and repeated words. This allows him to ramble his thoughts into code editors or chatbots and have perfectly formatted text appear on the screen.
✅ Specialized Agents for Admin and Assets
The final piece of the puzzle involves offloading the tasks that usually drag a creator down: meetings and repetitive assets. For meetings, he uses Granola. Unlike other tools that awkwardly join the call as a bot, Granola runs natively on the desktop, listening to the system audio. It generates notes and, crucially, identifies action items for all participants. If he scribbles a vague note during the call, the AI uses the context of the conversation to flesh that note out into a full detail later.
For visuals and audio, he relies on two specific heavy hitters:
- Nano Banana (via Gemini): He uses this specifically for generating the cartoony, stylized elements of his thumbnails. It allows for rapid iteration, he can ask for a character, request shading changes, or adjust the composition in seconds.
- ElevenLabs: This handles the ad reads for his podcast. He has cloned his own voice so effectively that his producers can generate the sponsorship segments without him needing to step into the recording booth. He reviews the scripts to ensure safety, but the actual recording time is reduced to zero.
Even YouTube Studio has become part of the AI stack. He uses the “Ask Studio” feature to query his channel naturally, asking questions like “Which videos drove the most subscribers in the last 14 days?” This surfaces insights that would otherwise be buried in complex analytics tables, such as realizing a year-old video is suddenly driving new growth.
If you want to see the exact prompts he uses or how that custom dashboard looks in action, you definitely need to watch the full breakdown.
Watch the full video here: