What does it actually cost to use AI as a second brain? Stephanie Sy ran the numbers. In one month, she used Codex across 188 sessions and 295M+ tokens to turn meetings, notes, scripts, reports, and trackers into reusable work artifacts. Her key insight: AI’s real value is not just faster answers. It is durable operating memory. For teams, that means moving beyond experimentation and measuring what matters: tokens, tool calls, artifacts created, human review needed, and what actually gets reused. This is the kind of grounded AI work we care about at Thinking Machines: helping organizations turn AI from isolated experiments into systems that compound knowledge and drive real business impact. Read Stef’s breakdown: https://lnkd.in/gjcg9_xs #AI #GenAI #SecondBrain #EnterpriseAI #ThinkingMachines
My team's about to do a coding-agent budgeting exercise, so I audited myself first across every Codex session I ran in May. Here's the tokenomics of my second brain. One month, 188 sessions, 295 million tokens, $315 if billed via API. Then I remembered I'm paying $100 for a ChatGPT Pro plan. Phew! 😅 What surprised me: → One daily workflow (auto-processing my meeting transcripts) ate 25% of the spend. Worth every cent — but I should look at using a cheaper or local model. → You can see that I'm more of a knowledge worker than a coder. My most productive sessions are workbenches: reading my notes, running scripts, building the deck/site/spreadsheet, filing the artifact back for next time. → 3,053 durable pieces of output built in a month. A bit of an overcount given how many files Codex generates as part of a single coding or deck-making task, but I'm reasonably 2x more productive than a year ago. → 4 day gap when my colleague and I did a series of face to face workshops exactly when my face to face meeting capture pipeline had a bug See my writeup for the full tokenomics of running an AI second brain 👇