If your agents are doing the work, you need to know whether the market is pulling that work out of your hands or whether you're still pushing. The signal hasn't changed; the surface has.
Tren Griffin's guide, published through a16z, does something useful: it refuses to let founders treat product-market fit as a milestone you declare. It's a condition you feel, most reliably as demand that starts to outpace your ability to supply. Griffin is careful to separate the before-PMF job, which is learning as fast as possible and killing attachment to your current hypothesis, from the after-PMF job, which is operational discipline so you don't spill the demand you've earned. Retention and referral patterns are the honest scorekeepers.
The translation is this: when agents execute the core workflow, founders tend to mistake automation efficiency for market fit. Your agents completing tasks faster is an internal metric. The external signal Griffin points to still applies — are users returning unprompted, expanding usage, complaining loudly when the service is down? An agent-native product has PMF when the human in the loop stops being you and starts being your customer, pulling the output into their actual decisions. Before that moment, your only job is shortening the loop between assumption and evidence.
- retention and referral patterns are still the honest scorekeepers, even when agents do the executing
- before PMF, your optimization target is learning velocity, not agent performance benchmarks
- the danger zone is confusing workflow automation with genuine market pull.
