Agency Over Skills, and Cursor's Security Agents

Lenny's Newsletter / Max Schoening (Notion) — Why Agency Beats Skills in the AI Era#

Source: https://www.lennysnewsletter.com/p/why-cultivating-agency-matters-more Credibility: High (first-party interview with Notion's Head of Product, published in Lenny's Newsletter)

What happened: Max Schoening — Notion's Head of Product and Design — joined Lenny's podcast to argue that the AI era has fundamentally changed what PMs and product teams should optimize for. His core claim: agency (the disposition to go figure things out, make decisions, take ownership) matters more than skills (specific technical or domain knowledge) because skills are increasingly offloadable to AI tools. He also drew a sharp line on vibe coding: it made more software, not better software.

Key PM craft patterns:

• Agency as the new compound interest: Skills depreciate faster when AI can replicate them. The ability to take initiative, define the problem, and drive toward an answer becomes the durable differentiator.
• "Drive it like it's stolen": Schoening's phrase for how Notion ships — moving fast with high ownership rather than waiting for permission or perfect information.
• Vibe coding critique: More software shipped doesn't mean more user value. The judgment about what to build and why doesn't get automated — and that judgment is what PMs own.
• Design and product converging: Schoening runs both design and product at Notion, and his thesis is that this convergence is accelerating with AI — the separation between "what" and "how it looks/feels" is narrowing.

Why it matters for PMs: The "skills vs. agency" framing is a useful lens for evaluating your own development and your team's. If you're spending PM development time on certifications, frameworks, or tool fluency — and not on expanding your judgment and initiative muscles — you may be optimizing for the wrong thing. The more AI handles execution, the more the PM's irreplaceable contribution is the willingness and ability to own outcomes, not just tasks.

The vibe coding critique is also worth taking seriously as a PM. The "build more faster" promise of AI-assisted development is real. But the product quality problem isn't about code — it's about what gets built. More surface area without more user insight is a liability, not an asset.

Critical questions:

• How do you actually hire for agency versus skills? What does an interview or work sample look like that surfaces this?
• Does the agency argument hold across seniority levels, or is it primarily a senior PM/lead differentiator?
• If vibe coding creates more bad products, does that create a market opportunity for teams with high judgment, or does it just create more noise?
• Notion runs design and product under one leader — how much of Schoening's framework is specific to that org structure versus generalizable?

Action you could take today: In your next 1:1 or team retro, ask: "What did we decide in the last two weeks, versus what did we execute?" If the answer is mostly execution, that's a signal about where agency might be atrophying.

Cursor — Security Review Agents, Now in Beta#

Source: https://cursor.com/changelog/04-30-26 Credibility: High (first-party changelog)

What happened: Cursor shipped Security Review as a beta feature for Teams and Enterprise plans. Two always-on agents run continuously against your codebase:

• Security Reviewer: Checks every PR for vulnerabilities, auth issues, injection risks, and insecure configurations. Leaves inline comments on the PR — same as a human code reviewer, but running on every commit.
• Vulnerability Scanner: Scans your full codebase on a schedule for existing vulnerabilities, not just new changes.

This ships two days after the Team Marketplace update (May 1), which lets admins configure plugins and MCP servers at the org level without requiring a connected repository first.

Key capabilities:

• Runs on every PR automatically — no manual trigger required
• Inline PR comments for security findings (integrates into existing review workflow)
• Scheduled full-codebase scans separate from PR checks
• Teams and Enterprise only (not available on individual plans)
• Still in beta — expect iteration on false positive rates and coverage

Why it matters for PMs: This is Cursor stacking enterprise-grade infrastructure on top of what started as a developer productivity tool. Security review is a real procurement objection for enterprise deals — "does your AI editor introduce vulnerabilities?" is a question every enterprise IT and security team asks. By building the answer into the product itself, Cursor removes a blocker and creates a new reason to upgrade to Teams/Enterprise.

For PMs evaluating developer tooling: when a tool starts shipping security features as part of the core product (not a third-party integration), that's a signal it's competing for the enterprise budget line, not just the individual developer credit card. The build-vs-buy question shifts — does your security tooling need to be separate, or does it belong in the editor?

Critical questions:

• What's the false positive rate on the Security Reviewer? Too noisy and it becomes background noise developers learn to ignore.
• Does the Vulnerability Scanner produce remediation suggestions, or just identification? Identification without actionable fixes is only half the value.
• How does this interact with existing SAST tools (Snyk, Semgrep, SonarQube)? Redundancy or replacement?
• Is the security data used to improve Cursor's models, and if so, what are the data handling implications for enterprise customers?

Action you could take today: If you're on a Cursor Teams or Enterprise plan, enable the Security Reviewer beta on one repo and track false positive rate for a week. That's the single most important signal for whether this is a workflow addition or a workflow tax.

Cursor — Team Marketplace: Plugins Without a Repo Requirement#

Source: https://cursor.com/changelog/05-01-26 Credibility: High (first-party changelog)

What happened: Cursor's Team Marketplace now lets admins create and configure team-level plugin bundles without first connecting a repository. Previously, setting up the marketplace required a repo connection as a prerequisite — a real friction point for IT/ops teams trying to standardize tooling before projects spin up. Plugins bundle MCP servers, skills, and subagents, and admins can now control install behavior (auto-install vs. optional) at the org level.

Key capabilities:

• Create team marketplace without repo connection
• Add/remove/configure first-party plugins directly in settings
• Control install behavior per plugin (mandatory vs. optional for team members)
• Bundles can include MCP servers, skills, and subagents in a single installable unit

Why it matters for PMs: The repo requirement was a classic adoption bottleneck — the people who want to standardize tooling (IT, DevOps, security) often aren't the same people who create repos (individual engineers). Removing this prerequisite decouples the admin workflow from the engineering workflow, which is how enterprise tools get deployed at scale. Cursor is building the organizational plumbing that makes company-wide standardization possible, not just individual adoption.

This pairs with the security agents announcement in an important way: you can now set security review as a mandatory plugin across your org, without requiring teams to configure it repo by repo. That's a fundamentally different product motion than "here's a feature, go turn it on."

Critical questions:

• What's the permission model for who can add plugins to the team marketplace? Does it require admin, or can team leads self-serve?
• Are there audit/logging capabilities so security teams can see which plugins are active across the org?
• How does mandatory install behavior interact with individual developer preferences? Can engineers opt out of org-level plugins?

Action you could take today: If your team uses Cursor and you've been waiting for a cleaner way to standardize MCP servers or subagent configs, this is the feature that unblocks that. Sketch out a team plugin bundle for your most common AI workflows — even one shared MCP server config reduces per-engineer setup time significantly.