I spent 48 hours in San Francisco meeting billion‑dollar AI founders and data‑lab leaders, and I distilled a set of observations that shaped my product priorities for the “agentic age”.
Data brokers are reportedly earning $2 M a week selling company data to AI labs.
The new bar is $1 B in revenue in 12 months.
Someone showed me about 10 apps worth $5B+ each that he’d cloned.
The AI that remembers the most about you wins.
Every company wants to plug its internal data into agents.
The hire everyone wants: someone with taste who can write evals.
Distribution is the bottleneck. Founders are sponsoring buses, buying billboards and throwing parties with famous DJs.
These insights led me to prioritize swappable model harnesses, robust data‑cleaning pipelines, memory‑centric agent designs, hiring evaluators with strong judgment, and high‑visibility distribution tactics.
- Build runtime‑swappable model harnesses.
- Deploy automated data‑quality assessment before ingestion.
- Engineer recall benchmarks to measure memory performance.
- Recruit evaluators who have “taste” for idea selection.
- Allocate budget to attention‑capture channels (buses, billboards, DJ‑hosted events).
I reflected on how software constraints have shifted from pure engineering to imagination, and I decided to double‑down on surprising designs.
I’d spend a lot more time on the version of my idea that makes people say "wait, what is that?"
Most software looks the same, so anything that looks different gets noticed, shared and remembered.
I note that the ChatGPT plugin ecosystem has grown over 2000% since the announcement.
My step‑by‑step plan is to identify a high‑frequency manual workflow, build it as a ChatGPT plugin, craft the description using exact user phrases, ship five variants, and double‑down on the one that gains the most traction.
I claim that 1.2 billion people use ChatGPT every week.
I argue that building a ChatGPT plugin is a high‑risk/high‑reward opportunity because of the massive user base and the new UI feature that recommends plugins directly in conversations. Early entrants can capture outsized distribution, but the window is limited—"In 12 months these slots will be crowded." I therefore decide to prioritize building a ChatGPT plugin now rather than later to leverage the early‑move advantage.
- Identify a tool or workflow people already search for weekly (e.g., PDF‑to‑Excel, invoice generation).
I link this strategic outlook to my broader professional narrative on my LinkedIn profile.
- Build it as a ChatGPT plugin.
- Write the description using the exact phrases users type into ChatGPT.
- Ship five distinct plugins and monitor which gains traction.
- Double down on the most successful plugin.