Why your goals are lying to you
Ravi Mehta, product advisor and former EIR at Reforge, opens with a provocation most PMs won't expect: your goals aren't a strategy. They're a measurement system — and plugging them in before you have a strategy is precisely how teams end up optimizing a KPI that moves sideways while the business doesn't move at all.
The five-layer stack
Mehta lays out a system of five interdependent layers: company mission → company strategy → product strategy → roadmap → goals. The chain runs in both directions. When you're building, work top-down: mission informs strategy, strategy shapes the roadmap, roadmap generates the goals you'll use to measure whether your bets paid off. When something isn't working — a goal you can't hit, a roadmap that feels scattered — work bottom-up and trace the break. Missed goals? Check the roadmap. Broken roadmap? Audit the product strategy. Misaligned product strategy? Surface the company-strategy gap.
The key move: goals belong at the bottom of the chain, not the top. Their job is to falsify the strategy — to tell you whether your bets worked — not to motivate a team or define direction.
Latency vs. velocity
Mehta draws a distinction that reframes how PMs think about productivity. Velocity is the quantity of work shipped. Latency is the time from idea to testable learning. Most large organizations have engineered themselves to maximize velocity — lots of PRDs, lots of launches, quarterly roadmaps — at the direct expense of latency. Startups, by contrast, have tight turning radii: less volume, but fast loops.
For 0-to-1 work, Mehta argues, latency is the variable that matters. A team that ships twelve features per quarter but waits six weeks to learn whether any of them worked is flying blind. A team that ships three features but gets a learning signal within days is actually moving faster in the dimension that matters.
AI has already changed the denominator
Mehta's most concrete claim: the fastest teams he works with now measure idea-to-test latency in hours, not weeks, because they're using AI to collapse the legwork — competitive teardowns, draft specs, low-fi mockups — that used to consume associate-PM bandwidth for days.
But this creates a new problem. Once AI removes the time cost of learning, the binding constraint shifts. It's no longer do we have time to study this. It's do we have the taste to interpret what we just saw. Product judgment — the ability to look at a competitor's onboarding flow, a pile of user feedback, or a draft spec and know what's good — is now the scarce resource. Speed without taste just produces fast wrong answers.