Lenny Just Dropped an AI Interview Coach — And It Closes the Feedback Loop We've Always Been Missing

Lenny's Newsletter just dropped one of the most useful AI-for-real-work pieces I've seen this year: a Claude Code-based AI Interview Coach built by Noam Segal, based on workflows pulled from 30+ tech professionals.

The headline story: Logan, a senior engineer who hadn't interviewed in eight years, used the system to land an Anthropic offer in two weeks.

The unlock wasn't "AI helped me practice." The unlock was the feedback loop.

Here's why that matters — and how to use it whether you're job hunting or not.

The Problem With Traditional Interview Prep

Lenny's article identifies three failures in the traditional prep model:

1. Impostor syndrome — candidates undervalue their own experience
2. Blind preparation — no way to know if your resume is landing, your stories are working, or your weak spots are improving
3. Insufficient practice — you can only "interview" so many real companies before you burn out

All three failures share the same root cause: no feedback loop. You go in, you do the thing, you walk out, and you guess at what went well. Every coaching field on Earth has solved this problem except interview prep.

This is exactly the gap AI is built to fill.

What the Interview Coach Actually Does

The system runs across three phases of the interview process.

Pre-Interview Phase

• Company research and culture mapping
• A "story bank" of your past experiences — surfacing things you'd forgotten you did
• Predicted question generation, mapped to your stories
• LinkedIn-based interviewer intelligence (their background, their likely focus areas)

This alone is more prep than most candidates do.

During Preparation

• Mock interviews — 4-6 question arcs with full feedback
• Progressive drills — 30-second answers ladder up to 3-minute responses
• Pushback simulations — what happens when they challenge you mid-answer
• Rapid-fire story retrieval — can you grab the right story under pressure

Post-Interview Phase

• Transcript analysis (recorded via Granola, Zoom, or Google Meet) scoring substance, structure, relevance, credibility, and differentiation
• Gap identification with root-cause diagnosis
• Side-by-side rewrites of your weaker answers
• Rejection debriefs and offer-negotiation scripts

Every output feeds the next phase. That's the feedback loop.

Why This Generalizes Beyond Job Interviews

Here's where I'd add my own spin to what Lenny wrote. This pattern — pre / during / post with feedback at every stage — is the template for using AI in any high-stakes verbal performance:

• Sales calls — pre-call research, mid-call coaching prompts, post-call transcript analysis
• Client pitches — practice runs, real-time framework review, debrief and refinement
• Webinars — pre-show prep, live producer support, post-show analytics on what landed
• Internal presentations — same architecture, different content

The interview coach is the demo. The real product is the "AI-coached repetition with feedback" model itself.

What Makes This Different From ChatGPT-Style Prep

The critical detail in Lenny's piece: it's built in Claude Code, not as a chat interface. That distinction matters.

A chat interface forgets. You have to re-explain context every conversation. The story you told in chat #3 is gone by chat #12.

Claude Code with the right project structure persists. Your story bank is a file. Your interviewer notes are a file. Your transcript analyses are files. The system gets smarter about you every time you run it.

That persistence is what closes the feedback loop. Without it, you're back to guessing.

Quick Setup if You Want to Try It

From the article:

1. Install the Claude desktop app
2. Download Noam's GitHub project (linked in Lenny's piece)
3. Run the kickoff command to initialize coaching state
4. Use prep [company] for company-specific preparation
5. Use analyze for post-interview transcript feedback
6. Type help anytime to see available commands

If you're not job hunting, I'd encourage you to fork the architecture for your own use case. The structure of pre-prep, in-the-moment coaching, post-analysis with feedback is genuinely transferable.

The Bigger Lesson

What I love about this build isn't the interview coaching itself. It's that someone finally made AI useful for a thing humans were already trying to do — and did it in a way that fits how the work actually happens.

This is the AI tool I keep telling people to build for themselves. Not a chat with an LLM. Not a one-shot prompt. A persistent, structured, feedback-driven system that remembers you and learns about you across sessions. Once you've built one for any high-stakes recurring task, you stop using ChatGPT for that task forever.

The Bottom Line

Logan's two-week timeline to an Anthropic offer is the headline. The real story is that AI tools are finally getting good at closing the feedback loops humans have been quietly suffering from for decades. If you're interviewing this year, install the Claude desktop app and grab Noam's project. If you're not interviewing, build the same architecture for whatever your version of "high-stakes recurring verbal performance" is. The competitive advantage from a working feedback loop is real, and most of your peers will never build one.