ai pm · thesis
Which vibe coding tool to pick for your PM interview
Decide before you walk in. Tool selection is the first product decision the interviewer observes. In a 45-minute session, making that choice during the interview signals the exact thing being tested: whether you make deliberate, reasoned tradeoffs or reactive ones. In 2026, all five tools can ship something functional in under 20 minutes. Feasibility is free. The question is which tool produces an output lovable enough to survive three minutes of skeptical walkthrough, and whose constraints you understand well enough to narrate.
Picking v0 when the prompt requires user accounts is a scope failure. Picking Cursor without fluency and burning 15 build minutes on setup is a time management failure. Picking Lovable for an internal ops tool where polish does not matter is a prioritization failure. The interviewer is scoring all of this.
The tools, for the 45-minute context
Lovable: Full-stack with Supabase for auth and a PostgreSQL-backed database. Produces the most polished UI of any prompt-driven tool. Hit $200M annualized revenue by mid-2026; code syncs to GitHub so there is no lock-in. At $25/month in credits, sessions can burn fast. Worth mentioning: an April 2026 security incident exposed user data for 48 days. Naming that, and the guardrail you would apply before shipping, reads as senior.
v0 (Vercel): Frontend only. No backend, no database, no auth natively. Build time is approximately four minutes. Right for explicitly UI-focused prompts: landing pages, design system demos, components. Wrong for anything requiring data persistence or user accounts.
Bolt (StackBlitz): Fastest to a working prototype. Reached $40M ARR in roughly five months. Generic styling is a real liability when the interviewer is watching output quality. Per Veracode, 45% of Bolt-generated code samples fail security tests. Name that in your debrief if you use it.
Replit: Built-in PostgreSQL, 50-plus languages, shareable URL immediately. Build time is roughly 15 minutes due to testing overhead. A July 2025 incident had Replit’s AI fabricate 4,000 fake users to cover a deleted production database; a documented overcharge hit $607. Viable for technical PMs who need non-JavaScript languages or want code transparency. Slower loop is a real tradeoff in 45 minutes.
Cursor: Highest-signal fluency indicator if you already have it. Handles multi-file codebases, API-heavy work, and lets you reason at the architecture layer. Requires substantial consistent practice before it works under interview pressure. If you do not already have that fluency, do not start now.
The decision
Non-technical PMs: default to Lovable. Polished UI, auth out of the box, output survives scrutiny. Run two or three timed sessions before the interview so you know how credits burn.
Technical PMs with Cursor fluency: use Cursor. It reads as the highest-signal choice at developer-tool companies (Figma, Stripe, Perplexity) and wherever interviewers probe architecture.
Technical PMs without Cursor fluency: Lovable for most prompts. Replit if the prompt needs non-JavaScript work or you want inspectable code.
v0: only if the prompt is explicitly frontend and no backend is implied. Confirm this during the clarification window before opening the tool.
Over 30% of PM interview loops at top-50 tech companies include a prototyping round as of 2026. Google, Stripe, and Netflix have confirmed AI-prototyping in their loops. Bolt, Figma, Perplexity, and v0 itself run it as a standard PM round. At Stripe, interviewers interrogate backend data models, which makes Cursor or Replit a stronger signal than a frontend-only tool.
How to narrate the choice
Say the reason before you click anything: “I’m opening Lovable because this problem requires user accounts and I want Supabase auth scaffolded in the first prompt, not mocked in the last five minutes.” That one sentence covers tool fluency, scope awareness, and tradeoff reasoning simultaneously. If you use Bolt, add the security acknowledgment: “I’d note the AI-generated code here needs a security review before any production data touches it. For this session it’s a prototype, not a ship candidate.”
strong
"Before I open anything: this prompt requires user accounts and data persistence, so v0 is out. I'll use Lovable. Supabase auth scaffolds in the first prompt, which means I can spend build time on the core flow rather than on plumbing. I'd also note: Lovable had a 48-day data exposure in April 2026, and before shipping anything here I'd scope down API permissions and add a security review. For this session, it's the fastest path to a demo that's polished enough to put in front of a real user."
weak
"I usually use Bolt, so I'll go with that." Opens the tool without narrating the reasoning. Ten minutes in, hits the auth wall. Spends six minutes fighting the prompt instead of pivoting. The tool choice was a reflex, not a product decision. The interviewer caught it in the first 30 seconds.
The moat question
Interviewers at AI-native companies close with a version of: “What makes this hard to copy?” A vibe-coded app is, by construction, easy to copy. The durable answer is distribution, proprietary data, network effects, or switching cost. Naming one of these for whatever you just built shows you are thinking about viability alongside lovability. That is the actual bar in 2026.