If you're letting agents handle workflow and wondering why retention still feels mushy, the answer probably isn't your stack. It's that you haven't found the thing people genuinely can't stop using.
Tren Griffin's guide, written for the a16z audience but sharper than most of that genre, makes the case that product-market fit is not a milestone you declare — it's a force you feel. Retention curves that flatten, organic inbound that outpaces anything you did deliberately, support tickets that reveal love rather than frustration. Griffin is careful to distinguish the feeling of pull from the delusion of push: founders spending on acquisition before the product earns it are just buying noise. Before fit, the only currency worth accumulating is learning speed.
The wrinkle for AI-native builders is that agents compress the iteration loop so dramatically that you can fool yourself into thinking speed equals signal. Griffin's framework reminds you that faster cycles only matter if you're measuring the right thing, and the right thing is still whether real people return without being prompted. When agents are doing the work, the human who holds judgment becomes your unit of retention analysis — does that person keep trusting the system, keep expanding what they delegate to it, keep pulling more of their workflow in? That pull is your PMF signal now.
- Retention among the humans holding judgment is your PMF metric, not agent task volume
- before fit, use your faster iteration cycles to kill bad hypotheses ruthlessly, not to polish output
- after fit, the main risk is that your agentic infrastructure can't scale to the demand you've earned.
