Skip to content
Aakash GuptaAakash Gupta

Ankit Interview 3

A new PM interview round is spreading across Google, Meta, OpenAI, and Anthropic, and most candidates have never seen it. It is not product sense. It is not engineering. It is the vibe coding round, where you design a feature and take it live to a prototype while the interviewer watches you use AI. This episode is a full mock interview so you can see exactly what it looks like when you have to ship an existing-product feature under a timer. Full Writeup: [VERIFY - newsletter URL] Transcript: [VERIFY - newsletter URL] --- Timestamps: 00:00 - Intro: the new vibe coding interview round 01:31 - The prompt: build a feature for LinkedIn 02:31 - Defining the goal and the user problem 04:23 - Speed-running the product sense round 06:53 - Prioritizing needs across user groups 12:30 - Mapping friction in the current workflow 17:56 - Writing the PRD and the three non-negotiables 21:30 - Curveball: the design system, strategy, and prior features 25:44 - Running the prompt across Magic Patterns, Lovable, and Claude Code 29:41 - Why you define the design system first 35:57 - Evaluating the five UI concepts 39:45 - Feedback: scoring the interview bucket by bucket 43:41 - The prompt, the tools, and using AI to write the prompt 46:01 - Outro: practice your timing, tools, and crisp explanation --- Key Takeaways: 1. The vibe coding round is a hybrid, not product sense or coding - It sits between the traditional product sense round and an engineering coding round. You design a product, then take it all the way to a working prototype live, while the interviewer watches how you actually use AI. 2. Real interviews ask you to build on existing products, not from scratch - Every tutorial online shows a brand-new product built from zero. The questions you are most likely to face are adding a feature to an existing product like LinkedIn, where constraints and prior features are already baked in. 3. Compensation makes this round worth mastering - AI PM roles at Google, Meta, OpenAI, and Anthropic pay from 300,000 to north of a million dollars. Some AI PM leaders now ask candidates to screen share how they use AI during the interview itself. 4. Speed-run product sense before you touch a tool - Compress the 30-minute product sense discussion into a few minutes. Nail the goal, the user problem, and why this feature over everything else you could build, before opening any prototyping tool. 5. Prioritize by user group, not by feature idea - Map the distinct needs across user segments first, then find the unique needs each group has. This keeps the feature grounded in a real problem instead of a solution looking for a use case. 6. A good PRD has three non-negotiables - Before prototyping, the PRD needs to define the problem, the solution, and the specific context clearly enough that an AI tool can build against it without guessing. 7. Establish the harness before the design system - Set up a context harness inside Claude, then have it learn the target product's design system before you feed it the feature prompt. Getting the harness in place early is what lets the tool pull the right design system in. 8. Use multiple prototyping tools in parallel - Run the same prompt through Magic Patterns, Lovable, and Claude Code at once. Magic Patterns is front-end only with no backend, Lovable publishes to GitHub, and Claude Code pulls from your actual codebase. Each gives you a different read on the solution. 9. Define the design system first, then everything else - Tell the tool to learn the design system of the target product before giving it more information. Concepts that respect the real design system look like features that will actually ship, not generic AI slop. 10. Timing and crisp explanation win the round - The interview is live with curveballs you have not seen. Practice your timing against the clock, practice confidence across different tools, and practice explaining crisply: here are the problems, here are the solutions, here is the PRD. --- 👨‍💻 Where to find the guest: LinkedIn: [VERIFY - guest LinkedIn URL] 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aakashgupta/ Newsletter: https://www.news.aakashg.com #VibeCoding #AIPM --- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications.

AnkitguestAakash Guptahost
Jul 29, 202647mWatch on YouTube ↗

CHAPTERS

  1. 0:00 – 1:19

    Why “vibe coding” is becoming a core AI PM interview round

    Ankit and Aakash set the context for a fast-emerging interview format that blends product sense with hands-on AI prototyping. They contrast it with traditional PM and coding rounds and explain why feature-building on existing products is the most realistic prompt style.

    • Vibe coding = hybrid of product thinking + rapid prototype creation
    • AI PM interview loops increasingly assess tool fluency via screen share
    • Most real prompts are feature additions to mature products (not greenfield)
    • High compensation and high bar for AI-native execution
  2. 1:19 – 2:35

    The prompt: ship an AI-first LinkedIn feature that helps people keep in touch

    Ankit lays out the mock interview: design and prototype a LinkedIn feature that reduces barriers to networking and increases engagement. The constraints are explicit: it must be AI-first, and the candidate must use AI tools live to prototype.

    • Assume PM role at LinkedIn; focus on staying in touch with network
    • AI should reduce the cognitive/awkwardness barriers to outreach
    • Interview structure: ~30 minutes prototyping inside a 45-minute session
    • Outcome expectation: concept + tangible prototype within LinkedIn constraints
  3. 2:35 – 4:20

    Clarifying the business goal: move users from passive consumption to active networking

    Aakash probes for the “why now” and success criteria, anchoring to LinkedIn’s mission and platform engagement. Ankit frames the objective as lowering cognitive load to drive retention and active participation in the network.

    • Goal: convert passive content consumers into active network engagers
    • Core lever: reduce cognitive load / barriers to initiating contact
    • Tie to mission: economic opportunity via engagement and retention
    • Early agreement on what success should influence (engagement → retention)
  4. 4:20 – 9:37

    Speed-running product sense: segmenting users and picking a focus

    Aakash quickly applies a traditional PM segmentation approach to identify who to build for. He compares job seekers vs. ‘professional profile maintainers,’ using rough sizing to prioritize which cohorts are both large and addressable.

    • Segmentation with estimated sizing to guide prioritization
    • Job seekers: referrals, target-company connections, learning from seniors
    • Professional-profile maintainers: large cohort with overlapping needs
    • Decides to focus on soft job searchers / profile maintainers as core
  5. 9:37 – 11:06

    Shifting from user buckets to the real hurdle: why outreach feels awkward

    Ankit redirects from segmentation to the underlying friction that prevents outreach. Aakash articulates the key blockers: forgetting relationship context, lacking a reason to reach out, and difficulty writing a non-awkward message.

    • Primary hurdle: no genuine/non-awkward reason to reach out
    • Weak ties matter most; context is missing or stale
    • Users forget how they know someone and what to say
    • Problem framing: ‘cold outreach is hard’ + ‘notifications aren’t enough’
  6. 11:06 – 13:02

    Live LinkedIn workflow audit: discovering friction in the current product

    Aakash opens LinkedIn and audits the existing ‘My Network’ and ‘Catch Up’ surfaces to find concrete issues. He surfaces broken/empty states, confusing notification coupling, and missing relationship context when viewing connections.

    • Catch Up empty state and unclear triggers for reconnecting
    • Job changes/birthdays tied oddly to notification settings
    • Broken navigation and poor discoverability of who to contact
    • Profile view lacks “why/how we’re connected” and recent interaction cues
  7. 13:02 – 17:13

    From audit to solution direction: recommend weak ties + provide context + draft outreach

    They converge on a solution direction: proactively surface “warm” signals for weak ties and generate context-aware draft messages. Ankit emphasizes that weak ties are the value and the product must supply genuine reasons plus better contextual writing than today’s AI message helper.

    • Two-part product need: (1) who to contact (2) what context to use
    • AI messaging exists but feels robotic and lacks relationship context
    • Feature should turn weak-tie prompts into authentic outreach moments
    • Focus on warmth + contextuality as core design principles
  8. 17:13 – 21:02

    Lightweight PRD in minutes: goals, non-goals, metrics, and definitions

    Aakash sketches a minimal PRD optimized for prototyping speed: define weak connections, outline non-goals to prevent spam, and set leading/lagging metrics. Ankit adds nuance: be explicit about what context is available on LinkedIn vs. external relationship history.

    • Define weak connection (e.g., no interaction in last ~3 months)
    • Non-goals: spam, out-of-context outreach, messaging strangers
    • Leading metrics: opens/reads, responses; lagging: retention/engagement
    • Context sources: shared employers/schools/locations + engagement signals
  9. 21:02 – 22:35

    Curveball: building a prototyping ‘harness’—design system, strategy, prior features

    Ankit challenges Aakash to explain what he’d pre-load to prototype faster. Aakash proposes a reusable harness: LinkedIn design system, strategy/feature context, and an AI prototyping skill pack, then starts assembling it in Claude Code using screenshots as grounding.

    • Top accelerants: design system, product strategy, prior feature context
    • Create a context library and reusable prompt scaffolding
    • Use screenshots to anchor UI fidelity and reduce hallucination
    • Treat AI as ‘middle 60%’: humans do upfront thinking + final refinement
  10. 22:35 – 26:30

    Parallel tool workflow: Claude Chat → Magic Patterns/Lovable/Claude Code

    Aakash demonstrates a multi-tool pipeline: use Claude Chat to generate a strong prototyping prompt, then run it across multiple prototyping tools in parallel. He explains why he separates ideation (more trusted model) from UI generation (faster prototyping tools).

    • One prompt, many tools: compare outputs quickly (front-end speed)
    • Magic Patterns for fast front-end; Lovable for build + GitHub flow
    • Claude Code for harness + code-level control with design system
    • Parallelization offsets latency and increases iteration speed
  11. 26:30 – 30:28

    What makes a strong prototyping prompt (and why design system comes first)

    Ankit asks for a quick rubric on prompt quality. Aakash outlines four ingredients—clear steps, rich context, explicit output format, and explicit exclusions—and reinforces the value of feeding design-system guidance early, especially for tools without a harness.

    • Prompt quality rubric: steps, context, outputs, and don’ts
    • Iterate prompts in-flight; stop and refine when fidelity is lacking
    • Define/learn the design system before requesting full UI generation
    • Use stronger models for concept generation; tools for rendering
  12. 30:28 – 34:18

    Evaluating five UI concepts and selecting the most promising surfaces

    They generate multiple UI concepts (home feed surfaces, network inbox-style tab, constellation map, profile context layers, etc.). Aakash triages them using innovation vs. impact vs. surface area, deciding the best approach combines multiple surfaces rather than a single standalone view.

    • Concept set includes: warm-signal feed, reconnect queue, constellation map
    • Evaluation lens: innovation, success impact, and available surface volume
    • Conclusion: combine concepts (surfaces + context layer) for best outcome
    • Prioritize pragmatic high-volume surfaces over flashy but niche UI
  13. 34:18 – 37:11

    Prototype review: Warm Threads, Reconnect Queue, Drafts-first, and Constellation

    Aakash reviews the prototypes produced in Claude Code, assessing fidelity to LinkedIn’s design language and strength of contextual outreach. He unexpectedly finds the practical surfaces (Warm Threads / Reconnect Queue) outperform the ‘killer’ constellation idea once rendered.

    • Warm Threads: strong design-system adherence and high usability
    • Reconnect Queue: clear weekly prioritization with context for outreach
    • Drafts-first: leverages context to reduce message-writing friction
    • Constellation map: intriguing but less compelling in real UI form
  14. 37:11 – 47:19

    Scoring and feedback: bucket-by-bucket evaluation + timing advice

    Ankit scores Aakash across the interview framework (user, problem, solution, PRD, prompt, tools, backend) and explains what mattered most: fast-but-deep thinking, real product auditing, and tool fluency. Aakash closes with a meta framework (UPS PPPB) and guidance to practice timing, workflows, and crisp communication.

    • Strongest signals: live audit, harness setup, multi-tool prototyping workflow
    • Improvement: compress early product-sense time to allocate more prototyping time
    • Prompting: clear structure and ability to explain why it’s good
    • Final advice: practice timing, tool confidence, and crisp explanation

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.