A senior cybersecurity product manager who adopted vibe coding says nearly everything she shipped after the initial learning curve started landing — customers adopted it, extracted value, and came back. Her warning: PMs who dismiss vibe coding as a dev buzzword will feel the consequences in about 18 months.

Priyanka Neelakrishnan, an Enterprise Data Security PM leader with over a decade of experience and author of Autonomous Data Security, laid out her argument in a HackerNoon essay. The core claim: vibe coding — describing what you want in natural language and letting AI generate working code from it — changes what a PM can produce, not just how fast they work.

The early experience, she admits, was underwhelming in a specific way. Setup friction is near zero compared with, say, configuring a JDK for a Java "Hello, World!" — but early outputs felt like a search engine repackaging. Ask about a cybersecurity feature and you'd get "a product brochure written by someone who'd read exactly one Wikipedia article about zero trust." The value appeared only after pushing through that phase.

What vibe coding actually gives a PM

PMs have always translated intent — vision into roadmap, CEO revenue targets into product bets, customer pain into prioritization. What's new: a PM can now describe how a customer experience should feel and what outcome a persona should reach, and get back a working, clickable prototype built from plain language. Not a wireframe, not a spec. Something a real user can click through without waiting for engineering bandwidth.

This is not an engineering replacement. Neelakrishnan is explicit: what she prototypes is a thinking tool and validation surface. The moment it heads to production, architects and senior engineers build it properly — robust, scalable, secure. In cybersecurity, she writes, "there is zero room for vibe-coded production code."

Her framing rule: "The quality of your AI output is directly proportional to the quality of your context."

Three separators over the next 24 months

Context. Speed is table stakes; every PM has the same Claude and Copilot. Depth of context is not evenly distributed. Prompt a threat-detection feature brief without knowing the difference between behavioral analytics and signature-based detection, or why a CISO cares about mean time to detect versus mean time to respond, and the output is beautiful-sounding garbage. She points to MCP (Model Context Protocol) servers as an inflection point: connect the AI to your data sources and it stops reasoning in generalities and starts using your product context, customer data, and competitive landscape. The nuance — the PM is one context vector among many. CRM records, support tickets, usage analytics, and competitive intelligence feeds can all be piped into AI workflows. PMs who orchestrate all those streams, rather than typing their mental model into a chat box each morning, will operate at a level that feels unfair to those prompting cold. Context isn't Google-able; it's built through customer calls, domain immersion, and whitepapers nobody reads. Systematize it; don't outsource it.

Speed — the right kind. She pushes back on "AI makes you faster" as commonly practiced: faster at generating documents nobody reads, roadmaps ungrounded in customer signal, backlogs full of pseudo-strategic features. That's noise at scale. The right kind compresses the gap between insight and decision: AI does competitive scans, pattern recognition across interviews, and draft PRDs you then pressure-test. Her model: a senior PM used to spend roughly 60% of time on synthesis and documentation, 40% on judgment. Used well, AI flips that to 70% thinking and deciding, 30% executing. That ratio shift compounds in a way raw output speed doesn't.

Deep domain expertise. Counterintuitively, AI makes it more valuable, not less. If AI can draft a security PRD in three minutes, domain depth still matters because someone must know when the AI is wrong. In healthcare, finance, defense, and security, a bad product decision isn't a missed sprint — it's a breach, a compliance failure, a patient outcome. She notes that agentic AI is creating new attack surfaces and zero trust has shifted from framework to procurement expectation, and no AI tool substitutes for judgment built by living in the domain. The winners combine genuine domain depth with AI fluency — a combination she calls still incredibly rare.

The practice changes

She reports five concrete shifts. Prototyping before engineering: clickable, testable experiences go in front of customers before engineers touch the problem, with same-week feedback loops — sharper than feedback on written specs. Customer research: interview transcripts go into AI with a persona, product area, and hypothesis specified; her job becomes interrogating the output, not first-pass synthesis. Competitive intelligence: AI does first-pass scans of positioning, pricing signals, and capability launches; she validates what matters to customers versus what sounds good in a press release. PRDs: AI drafts serve as thinking surfaces to react against — the version engineering receives always carries her fingerprints. Stakeholder communication: she prompts AI for counterarguments to her own proposals before executive presentations.

Her adoption numbers, she says, are rising significantly — not because she ships more, but because she ships what customers were already reaching for.

The failure mode

The biggest risk isn't irrelevance. It's shallow confidence. AI makes it easy to sound informed and produce fluent content that mimics expertise — a trap that will surface in customer conversations, executive reviews, and moments requiring real judgment. Her test: if you can't defend your AI-assisted output without the AI, something has gone wrong.

She predicts "AI PM" will go the way of "mobile PM" within five years — not a category, just an assumed capability, like writing a user story. The durable skill set for practicing PMs is unambiguous: build context infrastructure, spend AI-bought time on judgment rather than document volume, and treat domain depth as the most defensible asset on the roadmap.