Here's how I use AI day to day as a founder who lives mostly on the non-technical side
Too much of this conversation still assumes the best AI workflows are only for coders. They're not. What changed for me was using AI as a system I delegate to, review, and steer.
Tl;dr: start operating in "departments" or "jobs," run your day in parallel tasks, build runbooks, keep your context portable.
Disclaimer: I'm optimizing for operating leverage and faster decision making that takes in a significant amount of data I otherwise would struggle to find time to analyze.
What that looks like:
- I run multiple streams at once in Claude Code and/or Deck AI. The key is to treat them as departments or jobs. For example, one job researches a market, another drafts messaging, another pressure-tests a product decision. So I assign, redirect, and review when I'm ready vs. trying to do multiple things in a single session.
- I ask for multiple iterations against one objective, framed as a day of work. For instance "run through 10 iterations of {ask}." I've found instead of stopping at the first answer, it drafts, critiques its own work, and tries again. The end result is far stronger than version one. /loops and /goals is a dramatically more powerful way of doing this once you've got the hang of it.
- Combined with the above: I dispatch a team of agents at a single goal, each with a different job. To be sure on this one, request that Claude Code "dispatch a team of agents" to do {x}. One works the go-to-market angle, another acts as strategist, another as technical expert, and so on. Then I combine what's strong and discard the rest. It feels closer to managing a team than prompting a chatbot. In practice this means "dispatch a team of agents with agent a acting as gtm lead... each agent should run 10 iterations of {ask} which is the equivalent of one workday"
- I memorialize what works into runbooks my agents build by looking back on the steps we took. Next time I want to repeat something, I point an agent at the runbook and it has everything it needs. No starting from scratch.
- I religiously dictate vs type. You're wasting valuable insights by trying to consolidate all of your thoughts into text. The model will understand.
- I use Granola for call notes because it integrates cleanly with Deck. That turns conversations into usable follow-up and context instead of notes I never read again. It's also super lightweight and I appreciate the templates and the new primer feature they released.
- I set scheduled tasks in Deck to review product data, monitor Slack and HubSpot, and send me summaries of what actually needs my attention. It creates a recurring layer of review so I'm not manually checking every system for what changed. For what doesn't fit Deck, I run local cron jobs that ship data to my assistant by email.
- I kick off one long research task before bed almost every night. We often say there aren't enough hours in the day. Now I wake up to real progress, because it spent hours structuring and refining while I was offline.
- I maintain my corpus of context: family, product, GTM, voice. Better context is what makes the output consistently useful instead of randomly impressive. And I can carry this (I save mine locally) to conversations across models. This is a main benefit of claude code and codex vs. the web apps because the context is portable - just point the terminal at it.
- I use plan mode religiously. A lot of bad AI usage is just bad task definition. When the plan is clear, the output gets better. For non-technical users this can improve output significantly. Planning improves writing and analysis as much as it improves code.