Home
May 1, 2026
View All

Cursor Ships Security Agents, Wispr Flow Gets a Floating Notepad

The Short Version#

Two shipped product changes worth knowing about today: Cursor added always-on security agents to its Teams and Enterprise tiers (a meaningful step toward making AI coding tools enterprise-safe), and Wispr Flow launched a floating scratchpad that lives on top of every app. Plus Simon Willison is asking a genuinely useful infrastructure question about vibe-coded apps that PMs building AI tools should think about.

Cursor — Security Review Beta (Teams & Enterprise)#

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

What happened: Cursor launched Security Review in beta for Teams and Enterprise plans. There are two distinct agents: a Security Reviewer that checks every PR for vulnerabilities, auth regressions, and security anti-patterns, and a Vulnerability Scanner that runs continuously across the codebase. Both are always-on — they don't require developers to remember to invoke them.

Key capabilities:

  • Security Reviewer: Runs on every PR automatically; flags vulnerabilities, auth issues, and insecure patterns before merge
  • Vulnerability Scanner: Scans the entire codebase on a continuous basis, not just new code
  • Both agents operate as background processes — no manual triggering required
  • Available on Teams and Enterprise plans (not free or Pro tiers)

Why it matters for PMs: This is Cursor crossing a threshold that enterprise buyers have been waiting for. Individual developer productivity was never the blocker for enterprise adoption — security review, compliance, and audit trails were. Shipping always-on security agents turns Cursor from "the editor your engineers want" into "the editor your security team will sign off on." That's a different sales conversation. It also signals what the next wave of AI coding tool differentiation looks like: not more autocomplete, but agentic oversight baked into the workflow.

The plan-gating is deliberate. Security Review being Teams/Enterprise-only creates a hard upgrade incentive — if your team is using Cursor on individual Pro plans, this is a reason to consolidate.

Critical questions:

  • How does Cursor handle false positives at scale? An always-on agent that flags too aggressively will get disabled by developers, which defeats the purpose.
  • What's the human-in-the-loop model? Can security findings block merges, or are they advisory only?
  • How does this compare to dedicated security tooling (Snyk, Semgrep)? Is this complementary or a replacement play?
  • Does the Vulnerability Scanner run on historical code at onboarding, or only from the point of activation forward?

Action you could take today: If your team uses Cursor and is on individual Pro plans, this is the moment to run an upgrade-to-Teams calculation. Pull your current license spend and compare it to a Teams plan — Security Review alone may justify the delta if your org has any enterprise sales ambitions or security compliance requirements.

Wispr Flow — Scratchpad (Floating Notepad, Beta)#

Source: https://wisprflow.ai/whats-new / https://docs.wisprflow.ai/articles/9618237082-using-the-scratchpad-to-save-and-edit-notes Credibility: High (first-party roadmap/changelog, shipped today)

What happened: Wispr Flow shipped Scratchpad, a lightweight notepad that floats on top of whatever app you're in. It replaces Flow Notes. You pull it up with Option + S on Mac. The core idea: a voice-dictated notepad you can access without context-switching, that persists above your other windows like a sticky note. The update also includes "Transforms" — a beta feature that lets you rewrite any text with AI directly from the scratchpad.

Key capabilities:

  • Keyboard shortcut access (Option + S on Mac) — no app switch required
  • Floats on top of all other applications
  • Replaces the previous "Flow Notes" feature
  • Transforms (beta): AI rewriting directly on captured text
  • Available in v1.1.34

Why it matters for PMs: Wispr Flow keeps expanding what "voice-first AI" means. They started as dictation-to-text. Now they're building ambient capture infrastructure — a layer that sits above your entire desktop and intercepts your workflow at the moment of a thought, not after you've switched to a note app. Scratchpad is small but directionally important: it's the difference between "AI helps you after you decide to use it" and "AI is available wherever you are."

The Transforms feature is worth watching. If you can capture a rough voice note and immediately rewrite it into a formatted Slack message or email draft without leaving the scratchpad, that's a meaningfully different workflow than what any current note tool offers.

Critical questions:

  • How does Scratchpad sync or export? A floating note that doesn't connect to your existing systems (Notion, Linear, etc.) has limited stickiness.
  • What's the differentiation from Apple's new system-level dictation features? As OS-level voice capture improves, Wispr Flow needs reasons to exist outside the OS.
  • Does Transforms work offline or is it dependent on a model call? Latency will determine whether it feels instant or awkward.

Action you could take today: If you already use Wispr Flow, try Scratchpad for one full work session — specifically during meetings or async video reviews where you currently have to pause and switch apps to capture a thought. That's the workflow it's designed to replace.

Simon Willison — "We Need RSS for Vibe-Coded Apps"#

Source: https://simonwillison.net/2026/Apr/30/rss-vibe-coded-apps/#atom-everything Credibility: High (direct analysis from a recognized expert in developer tooling and AI workflows)

What happened: Simon Willison published a short piece arguing that the explosion of vibe-coded apps — lightweight, single-purpose tools built quickly with AI assistance — creates a discoverability problem that the web hasn't solved yet. His argument: RSS existed to share abundant content (blog posts, podcasts) without requiring centralized platforms. The same infrastructure gap now exists for apps. There's no lightweight, decentralized way to share and discover vibe-coded tools the way you'd subscribe to a blog feed.

Key patterns:

  • The abundance problem is real: as app-building gets easier, the number of small tools being created will vastly outpace discovery mechanisms
  • Centralized platforms (app stores, Product Hunt) introduce curation bottlenecks and incentive misalignment
  • RSS-like federation would let creators share apps directly, without going through a gatekeeper
  • The implication is that the tooling layer around vibe-coded apps is still missing — not just building, but sharing, versioning, and discovering

Why it matters for PMs: This is a legitimate open infrastructure question, and Willison is raising it at exactly the right moment. If you're building any kind of platform that touches developer tools, no-code builders, or AI app creation, the discoverability layer is a real product gap. Tools like Lovable, Replit, and v0 make building trivially easy — but none of them have solved "how does someone who isn't the creator find this app?" Whoever builds that layer first will have significant distribution leverage.

There's also a more immediate PM takeaway: if your team is creating internal vibe-coded tools (and many teams are), you probably have no system for cataloging, versioning, or sharing them. That's a concrete problem worth solving now.

Critical questions:

  • Is the right model RSS-style federation, or is it something more like npm for apps — a registry with versioning and dependency tracking?
  • What's the incentive for platforms like Lovable or Replit to support open discovery versus keeping users inside their own ecosystem?
  • How do you handle quality signals (ratings, usage metrics) in a decentralized discovery system?

Action you could take today: Audit whether your team has any informal catalog of AI-built internal tools. If you've had even one team member vibe-code a useful script or app in the last three months, you have a discoverability problem. A simple shared Notion page with links and descriptions is a better starting point than nothing.

Quick Hits#

  • Stripe: "Everything we announced at Sessions 2026" — a full product roundup from Stripe's annual conference on April 29, including signals on agentic commerce and programmability changes. Worth reading if you track fintech product strategy: https://stripe.com/blog/product

  • Microsoft Word: Launched a "Legal Agent" in Word on May 1, rolling out to Frontier (enterprise) customers. AI agents embedded in document workflows is the pattern; legal is the first vertical. https://techcommunity.microsoft.com/blog/microsoft365copilotblog/word-legal-agent-in-frontier/4516218

  • Harrison Chase (LangChain): Published "Your harness, your memory" — an article arguing the agent harness (not the model) is the dominant architectural pattern for building agents, and therefore where memory, learning, and customization should live. Relevant to anyone making build-vs-buy decisions on agent infrastructure: https://x.com/hwchase17/status/2042978845347745871

  • Vercel AI Gateway: Grok 4.3 is now available on Vercel's AI Gateway — 1M token context window, December 2025 knowledge cutoff, improvements in tool calling and instruction following. If you're routing models through Vercel's gateway, it's available now: https://vercel.com/changelog/grok-4-3-on-ai-gateway

  • Karri Saarinen (Linear): Posted on LinkedIn noting "we are nearing the six-month mark from the last large jump in model coding capability" — a quiet but pointed observation that the improvement curve may be flattening, at least for now. Worth sitting with if you're making bets on capability timelines: https://linkedin.com/in/karrisaarinen/

The Thread#

The enterprise unlock problem. This week's pattern: AI tools that started as individual productivity tools are now shipping the specific features that unlock enterprise adoption — security review (Cursor), legal agents (Microsoft Word), data residency (Notion). None of these features make the core product better for individual users. All of them make the product safe enough for procurement teams to approve. The build order matters: nail individual stickiness first, then add the enterprise wrapper. Cursor's security agents are the clearest example yet of a developer tool executing that playbook in real time.

Sit With This#

Cursor's Security Review agents are gated to Teams and Enterprise plans — not because the feature requires it technically, but as a deliberate upgrade incentive. Individual Pro users who want always-on security review have to consolidate to a team plan.

For your product: Think about the last feature your team shipped. Was it available to all users, or did you gate it by plan? If you gated it: what was the actual rationale — revenue, infrastructure cost, or a forcing function for upgrades? And how would you know if the gate is working as intended versus just annoying users who don't convert?