How LennyRPG Got Built In Weeks Instead Of Months — The AI Workflow Worth Stealing

Lenny Rachitsky published a full breakdown of how Ben Shih (a product designer at Miro) built LennyRPG — a browser-based role-playing game that turns 300+ episodes of Lenny's podcast into a Pokémon-style world where you collect and battle podcast guests.

I read it twice. Not because I'm building a game, but because the workflow Ben used is exactly the workflow every operator should be running for their own business.

Here's the breakdown.

What LennyRPG Actually Is

For context: Ben took the entirety of Lenny's podcast back catalog and turned it into a game where you:

• Walk around a pixel-art world
• Encounter podcast guests as characters you can capture (like Pokémon)
• Answer product-knowledge questions in battles
• Compete on a leaderboard tracking your progression across 275+ collectible characters

It's a real game. It launched on Vercel. People are playing it.

Now, here's why this matters for non-game-builders.

The Six-Phase Workflow He Used

Ben's process is the part to steal. The article walks through how he structured the build:

1. Ideation — Visual sketching in Miro to communicate the concept clearly
2. PRD Creation — Using AI to interview him until requirements were nailed down
3. Proof of Concept — Built with Phaser 3 (after testing RPG-JS and pivoting)
4. Feature Scaling — Automated systems for processing 300+ transcripts, generating quiz questions, creating avatars
5. Polish — Code reviews and UI refinement
6. Deployment — Vercel + Supabase backend

The tools he used: Claude Code, Codex, Cursor, GPT-4o, OpenAI image generation, Supabase MCP.

The Lesson Most People Will Miss

The headline lesson Ben gives is the one I'd put on a sticker:

"Nailing the core idea and PRD determines 80% of how smooth the rest of the build will be."

In other words: the time you spend at the front end defining what you're building pays back 5-10x in the execution phase.

This is the lesson that translates everywhere. It's not a game-building lesson. It's a project lesson. Whatever you're building — a funnel, a course, a product, an offer — the time you spend on the spec is the highest-leverage time you'll ever spend.

If you skip the PRD step, you'll spend 3 weeks building the wrong thing and have to start over. If you spend a Saturday on the PRD, the build phase is half as long and ten times less painful.

The Automation Multiplier

The second lesson hidden in the article is the CLI automation Ben built.

Instead of:

• Manually processing 300+ podcast transcripts → AI built a CLI tool that did it
• Manually creating 250+ pixel-art avatars → AI built a generation pipeline
• Manually finding audio assets → AI did asset hunting

Each one of those tasks would have taken weeks of human time. With AI-built CLI tools, they took hours.

This is the part that translates to every business:

Manual task	AI-CLI version
Writing 50 individual blog posts from a template	One CLI that runs the template against 50 inputs
Building 30 personalized welcome emails	One CLI with a prompt + a list of customers
Generating 100 social posts from a content library	One CLI that pulls + reformats + outputs
Creating 200 product variants for a catalog	One CLI that reads spec + generates copy/images

Whatever you're doing manually right now in batches of 10+, AI can build you the CLI to do it in batches of 200.

When To Pivot (Lesson 3)

Ben initially built the proof of concept in RPG-JS before realizing it didn't fit the quiz-based combat he wanted. He switched to Phaser 3 and the project sped up.

This is the lesson the AI YouTubers don't talk about enough: first tool choice is rarely the right tool choice. The skill isn't picking the right framework on day one. It's recognizing within the first week when the framework is fighting you, and pivoting fast.

For business owners, this same lesson:

• The first email tool you pick will probably be wrong
• The first ad platform you test won't be your forever channel
• The first offer structure you launch will need to be revised
• The first CRM you implement will get replaced

The fast-learners don't get this right on attempt one. They get good at knowing when to switch before they're 6 weeks deep in the wrong stack.

Hybrid Decision-Making (Lesson 4)

The fourth thing Ben names — and this is the hot take — is that AI is great at execution and sanity-checking, but humans should retain creative and strategic control.

Specifically:

• AI shouldn't pick which features matter most
• AI shouldn't decide game difficulty balance
• AI shouldn't make the brand voice calls

Those are human decisions. AI is the build engine, not the brain.

I've been saying this on every live for the last six months. The operators who try to "let AI decide" what to build are going to ship a lot of mediocre output. The operators who use AI to build their decisions faster are going to ship better stuff than they could have alone.

What Every Operator Should Steal

If you're not building a game, here's how to translate Ben's process to your business:

1. Sketch your idea visually first — even a paper doodle works
2. Write a real PRD before you start building — even for a 1-page lead magnet
3. Identify the repeating tasks early — anything you'd do more than 10 times manually is a candidate for an AI CLI
4. Pivot fast when a tool doesn't fit — don't sunk-cost-fallacy your way through the wrong stack
5. Stay in charge of the creative and strategic calls — let AI do the build, not the thinking

That's it. That's the whole article in five bullets.

The Bottom Line

Ben's full write-up is worth the 15 minutes whether you build software or not. The process he used is the new operator's playbook for shipping ambitious projects on a small team.

The compression is real. What used to take a 3-person team three months can now take a 1-person team three weeks. That's the era we're in.

The question isn't whether AI changes your build process. It's whether you adopt the workflow before your competitors do.