How I use AI
every day

For anything that involves writing — PRDs, stakeholder emails, exec decks, interview prep — I describe what I need and get a first draft in under a minute. I then spend my time making it better rather than staring at an empty doc.

I think better out loud than in writing, so I've built my workflow around voice. I hit record in Granola and just talk — about a product problem, a customer situation, whatever's on my mind. The transcript goes into Claude or Cursor and comes out as a structured document.

I create one context file per product I work on. It stores everything relevant — requirements, past decisions, JIRA status, Confluence links, key stakeholders. At the start of any Claude session I load this file and immediately have full context without rebuilding it from memory.

When I need competitive context or market data, I ask Perplexity rather than opening five browser tabs. For deeper synthesis — like loading an analyst report or a batch of customer call transcripts — I use NotebookLM so I can ask specific questions across all of it at once.

Before any high-stakes conversation — exec alignment, a difficult customer call, a job interview — I describe the context to Claude and ask it to argue against my position as hard as it can. The goal is to find the holes in my thinking before someone else does.

I use Claude Code to build complete web apps by myself. I act as the PM and designer — I describe what I want, review what gets built, and redirect. Side projects I used to shelve because I couldn't build them fast enough now actually ship.

When I want to prototype something that builds on an existing product, I use Figma MCP to pull the real screen designs into Claude Code. That way I'm iterating on top of the actual design system rather than building something that looks completely different from the product.

When I have a rough idea for a new screen or flow, I describe it in Google Stitch and get a full UI design back in seconds. I use it to explore 2–3 directions before deciding which one to prototype or take to engineering — much faster than doing wireframes by hand.

I set up OpenClaw on a VM and connected it to Ollama, which runs open-source LLMs locally. The result is a personal assistant I can message on Telegram from anywhere — and since everything runs on my own machine, no data leaves my environment. Useful for anything I don't want going to a cloud API.

Before a big presentation I open Grok voice and speak through my talk as if I'm in the room. Grok asks me questions, pushes back on vague points, and tells me where I lost the thread. It's closer to real rehearsal than reading slides to myself.

My watchlist lives in Google Sheets. A Make.com scenario iterates through each ticker, sends it to Perplexity AI with my fundamental analysis criteria — things like P/E ratio thresholds, revenue growth, debt levels — and writes a pass or fail against each criterion back into the sheet. I run it before deciding whether to research a stock further or ignore it.

I built a pipeline in n8n that takes documents, chunks them, generates vector embeddings, and stores them in Supabase. A chatbot sits on top — when someone asks a question, it retrieves the most relevant chunks via vector similarity search and passes them to an LLM to answer from that context. No backend code written by hand.