The Product Builder Certification: From Vibe Coding to Spec Engineering

Gayathri Keerthana (GK)

ex-Amazon Sr. PM, running 3 ventures

PMs and engineers are merging into builders. Vibe coding only gets you halfway.

You're likely already facing this: vibe-coded prototypes break before a real customer can use them. 15,000+ professionals came for my Lenny-featured Lightning Lesson where I showed a primer to fix this.

There's a new role forming in every product team - Product Builder - someone who can ship in short sprints with the customer at the center.

The real question is how to engineer a build that holds past mid-build, without wasting millions of tokens on vague vibe-coding.

Spec Engineering fixes this. Anthropic has been publishing on this but hardly anybody talks about it or shows how it will look like in your world.

In this workshop, I'll show you how to do a spec-driven build on a real problem:

🧩 Plan: Working Backwards into a spec the harness can build from.

🤖 Build: lock "done" as a Bar Raiser commit, build one feature at a time with handoffs that survive context resets.

🔍 Evaluate: a sperate agent that catches what self-grading misses, calibrated to your quality bar

🔁 Strip: remove scaffolding as the model improves

So what:
I show harness design encoded with first principles. I've used these at Amazon scale. They outlast model upgrades.
Scroll down to testimonials for the rest.

Workshop agenda

2:00PM EDT
Working Backwards from a customer brief into a spec your harness can build from
A 1-4 sentence brief expands into a structured spec with a feature list, written in customer language so the cascade hardens around the right vocabulary.

2:45PM EDT
Contract: lock "done" as a Bar Raiser commit before any code is written
Generator and evaluator agree on the testable behavior that defines done. The contract is the rubric the build is judged against, set before the artifact exists.

3:30PM EDT
Build: one agent, one job, with handoff artifacts that survive context resets
Composition discipline: resist the urge to merge the planner and generator. Single-responsibility is what makes the harness durable.

4:30PM EDT
Evaluate: a separate skeptical critic that catches what self-grading misses
Critic > Creator. Calibrate the evaluator with few-shot examples that encode your taste, not the model's defaults.

5:00PM EDT
Strip + clinic: read your traces, remove what's no longer load-bearing
Every component is built to die. On each new model release, walk the harness and strip what the model now does on its own.

6:30PM EDT
Closes with a 1-hour clinic on participant builds.
Breaks will be included at the end of each stage

What you’ll learn

Walk out as a Product Builder, closing the loop from customer problem to shipped agentic build.

Harness design stops feeling like insider jargon. You see the loop, the agents, the handoffs, and how each piece earns its place.
Failure modes like context loss, self-grading bias, and scaffolding decay become things you can spot in your own work.

Written in customer language, expanded from a one-line brief, ready to drive a planner/generator/evaluator loop.

Built against a Bar Raiser sprint contract, with handoff artifacts and a clean state any next agent can pick up.

Skeptical critic agent, calibrated with few-shot examples, files specific findings the generator can act on.

Read traces, ask which scaffolding is still load-bearing, remove the rest. Reusable on every release.

Someone who closes the loop from customer problem to shipped product, regardless of which side you started on.

Learn directly from Gayathri

Gayathri Keerthana (GK)

Contact

AI Consultant, ex-Amazon SrPM (Machine Learning), 3x Founder

Previously at / Worked with

See all products from Level Up With AI

Who this workshop is for

PMs who are becoming builders, coding fluency doesn't matter
Engineers who want to wear the PM hat, build customer centered products
Vibe-coders who want to build products that don't crumble mid-build and make the token burn worth it

Prerequisites

Claude Code, Cursor or equivalent installed
Paid plan: Claude Pro/Max or Cursor Pro
A small tool or app you want to build
bring an idea for an internal tool you'd actually use (e.g., a custom builder pulling from your own data sources)

What's included

Live sessions

Learn directly from Gayathri Keerthana (GK) in a real-time, interactive format.

Lifetime access

Go back to course content and recordings whenever you need to.

Community of peers

Stay accountable and share insights with like-minded professionals.

Certificate of completion

Share your new skills with your employer or on LinkedIn.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Free resource

Spec Engineering: the Unclaimed Role of an AI-native PM

Learn why the old PM spec doesn't translate to AI agents

What changed about the spec and how the gap looks like in your org

Understand why Spec Engineering is a PM skill

The industry named the highest-leverage skill in AI development "specification engineering." Learn how to claim it.

Live demo: building a structured spec vs typical vibe-coding

Take home a structured spec template using a Working Backwards approach that you apply in your next spec engg exercise

Testimonials

GK was excellent to work with. She supported us across AI workshops and AI course tutoring, and consistently delivered high-quality, well-structured materials and outputs. Communication was clear and proactive, and she genuinely cared about achieving strong outcomes, not just “completing tasks.” She also interacted brilliantly with students and workshop participants: engaging, supportive, and very clear in explanations and facilitation. Everything was delivered on time, with strong attention to detail, and she stayed flexible and responsive as priorities and timelines evolved[...]she’s thoughtful, reliable, and simply a great person to collaborate with. I strongly recommend her for training delivery/support and for helping early-stage startups, especially across AI, marketing, and product
Vitaly
Director at Aon STG | AI & Digital Strategy | Venture Builder
I watched all 21 sessions in the Maven/Lenny Rachitsky "AI-Native Product Manager" Lightning Lesson series.
Most "AI for PMs" content sticks to the obvious: one-shot PRDs, prototyping, market research. Useful, but usually surface-level. What I'm after is how AI fits into the mess. The work that looks different at every company and for every PM.

𝟭. 𝗥𝗮𝗶𝘀𝗲 𝗬𝗼𝘂𝗿 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗕𝗮𝗿 𝗮𝘀 𝗮𝗻 𝗔𝗜-𝗡𝗮𝘁𝗶𝘃𝗲 𝗣𝗠 (Gayathri Keerthana Shanmuga Sundaram, Jason P. Yoong)
The best session of all 21. Gayathri goes through fine-tuning a support chatbot agent for Air Canada, focusing on a specific policy area, to keep the evals demo consumable and nuanced. Smart move. Talks about how to use the eval results and how to work with developers on it [...]
Guy Peled
Product Manager
We were seeking a product owner [...] to help us understand how the ML model can be leveraged to solve concrete business problems and surface actionable insights for business users. This is where Gayathri shone. She laid out and executed a strategy for business application… with simple and easy-to-understand language and used relatable examples… Gayathri's contributions have not only advanced the team's understanding of customer behavior but have also set new benchmarks for how we leverage machine learning in business.
Michael
Principal Applied Scientist
I have not come across many such product leaders in my 14 years of experience in tech and product management, including a decade at Amazon… That's when I witnessed that she has the acumen to integrate different methodologies into a cohesive strategy… She consistently pushed the operations teams to look beyond immediate hurdles and focus on broader customer behavior trends… Her entrepreneurial spirit, customer obsession, and innovative problem-solving skills will undoubtedly enable her...
Sarah
Principal PMT (Applied Sciences / ML )