Module 3 · Card 1 of 8

Agent tools for PMs in 2026

Not a tool roundup. The question this module answers: when is an agent the right interface, and when does it cost you more than it gives you? The frame is the economy of trust — agents earn it, lose it, and you decide who gets to commit changes to your stack.

Module 3 · Card 2 of 8

Three categories live right now

Drafting agents (Cursor, Notion AI) — you read every output. Triage agents (Linear AI, GitHub Copilot for PMs) — they propose, you approve. Execution agents (Claude with computer use, Anthropic agent SDK builds) — they act, with guardrails. Each category has a different trust contract.

Module 3 · Card 3 of 8

What's actually shipping (Q2 2026)

Cursor: PRD-drafting flows referenced by 38% of enterprise PMs (Cursor product analytics, Q1 2026). Linear AI: auto-categorization of incoming bugs against your existing labels — 22% adoption in the same survey. Anthropic's agent SDK: in production at scale for 4 of the 10 largest SaaS companies.

Module 3 · Card 4 of 8

The trust contract

Read-output: low trust required, low cost when wrong. Approve-or-reject: medium trust, you absorb the review burden. Auto-commit: high trust, the agent can ship something measurable while you're asleep. The cost of being wrong is what changes — choose the tier based on the action's reversibility.

Module 3 · Card 5 of 8

Where to delegate first

Start with reversible, low-stakes work: ticket triage, meeting notes summary, draft pre-reads, competitive scan synthesis. Build the workflow + your trust calibration before delegating anything that touches production scope or stakeholder communication.

Module 3 · Card 6 of 8

Where to keep driving yourself

Anything that compounds. Roadmap decisions, prioritization debates, strategic positioning — these create your moat. The agent can prepare the inputs (data pulls, doc drafts) but the decision stays yours. The agent that decides for you is the agent that erodes the part of the job that's yours to keep.

Module 3 · Card 7 of 8

Calibrating trust over time

Every agent action is a data point. Track the disagreement rate — how often you override the agent's suggestion. Below 5% means you're under-using it. Above 30% means it doesn't have the context yet. The middle band is where the leverage lives.

Module 3 · Card 8 of 8

End of module

Next: five questions. Two minutes. Then look at your last sprint and ask which 60-minute task could have been a 5-minute review.