Your Next User Might Not Be Human: A PM's Framework for Agent-Ready Products
Software is quietly picking up a second audience — AI agents acting on a human's behalf. Here's a framework for auditing whether your product actually works for both.
For twenty years, "who is the user" had one answer: a person, looking at a screen, making a decision. That assumption is starting to crack.
Across product-led SaaS this year, a growing share of usage isn't a human clicking through your UI — it's an agent, acting on a human's behalf, calling your API, reading your docs, filling in your forms. The PM discipline that grew up entirely around human attention, human comprehension, and human trust now has a second audience it wasn't designed for. Most roadmaps haven't caught up.
The shift is already showing up in the data
This isn't a thought experiment. Three signals converged this week that are worth naming plainly:
- Teams building agent-native tooling are shipping in public right now — background agents that record how you work and turn it into automation, inboxes wired directly into durable agent workflows, sandboxed VMs built specifically so an agent can safely act unsupervised. These aren't demos; they're shipped products with real usage.
- Analysts tracking product-led growth are now describing a "headless" trajectory for software — where the agent, not the human, is the one actually operating the tool, and the human's role shifts from operator to approver.
- On the hiring side, "API PM" and "AI platform PM" are now showing up as distinct functional specializations, separate from the generalist PM role — which only happens when a company has decided a product surface needs dedicated ownership.
None of this means humans are disappearing from your product. It means most products now have to satisfy two different readers of the same information, often on the same screen, and most product decisions are still being made as if there's only one.
Why this breaks quietly, not loudly
The dangerous part of this shift is that it doesn't show up as an outage. It shows up as a slow erosion of trust in your product's own outputs.
A feature built entirely for human judgment optimizes for things agents don't parse well: ambiguous copy that reads fine to a person but has no single correct interpretation, state that lives only in visual layout rather than in the underlying data, confirmations that assume a human will notice if something looks wrong. An agent acting against that same surface doesn't pause to sanity-check the way a person does — it does exactly what the interface told it to do, at whatever speed it's capable of, and it does it whether or not the interface was actually clear.
Keep reading
Spec Hallucination: The 15-Minute Claim Trace for AI-Written Specs
AI writes specs that sound sourced but often aren't. The Claim Trace is a 15-minute audit that labels every claim Sourced, Inferred, Assumed or Invented before engineering builds from it.
What Counts as Evidence? The Risk Receipt for Value, Usability, Feasibility, and Viability
AI made feasibility cheap to prove. Value, usability, and viability still need their own evidence — and a great prototype demo counts for less than it feels like. Here's the Risk Receipt framework for checking what you've actually retired.
What Product Managers Should Never Delegate to AI
AI can summarize evidence and propose options, but some product decisions should remain human-owned. Here are the judgments PMs should never delegate without review.