What you'll learn in this episode
In this conversation, Amir Elion joins Pavitra on The AI Momentum Podcast to explore why speed is the most dangerous drug in AI building right now. Based in Stockholm, Sweden, Amir is a product strategist and founder who previously led AWS Innovation Programs in the Nordics. He shares his live workflow across Claude Code, Base44, and Claude, walks through Castifai, and makes the case for why product leaders must become architect-builders to avoid shipping vibe debt at vibe speed. These are topics Amir regularly speaks about at conferences and corporate events.
From months of frustration to hours of shipping
Amir opens with the contrast that shaped his view of AI. A decade ago as a product director, turning a good customer insight into a testable product took months of engineering cycles. His first magic moment with generative AI was building an LLM-powered Chrome extension for LinkedIn commenting - in his non-native Swedish - in a few days, despite not being a software engineer. The realization dropped: he could now solve problems for himself and for others without waiting in a backlog queue.
Vibe debt - when ease of execution hides strategic misalignment
Amir coins a distinction the show's host frames as vibe debt: decisions made simply because AI tools made them too easy to apply. A Sunday quick win calcifies into a Tuesday permanent blocker when founders confuse momentum decisions with technical decisions. Amir's defense is to think twice or three times before committing to a big new feature, explore three to five parallel versions before choosing, and rely on design principle files that Claude reads at the start of every session.
Design principles and modular architecture as AI guardrails
To keep AI models aligned with business goals, Amir maintains .md files covering design principles and modular architecture rules. A signature rule: prompts are never hardcoded - they are stored as data entities with an admin screen for testing different versions. Iterating on a prompt never requires touching code. Whenever the model drifts into a recurring mistake, the rule is added to the principles and carried into every future session.
Three hours instead of three months
Amir walks through a recent example: preparing for a meeting with a social impact startup, he asked Claude Code on a Saturday to build five different dashboard versions. The team spent 20 minutes picking and mixing during the meeting, and within 15-20 more minutes had a working demo. He then asked Claude to publish to GitHub and Vercel - one instruction, a live link for stakeholders minutes later. The old wireframe-mockup-prototype sequence collapses when creation cost approaches zero.
The hybrid testing loop - recorded humans plus AI analysis
AI is strong on details. It is blind to human emotion. When UX expert Martin hesitated for 1-2 seconds before granting Castifai permission to post on his LinkedIn, that pause revealed a trust gap no AI report would flag. Amir's workflow: record real users walking through the app for 20-30 minutes, transcribe the session, then feed the transcript to Claude to cluster UX issues into prioritized fixes. Don't give up on humans - help them express what they really need and push it forward with AI.
Castifai - a live walkthrough of where vibe meets reality
Amir demonstrates Castifai, his tool that turns YouTube, Spotify, TikTok, and other sources into LinkedIn posts, articles, and infographics in your own voice. The stack:
- Base44 handles batteries-included backend - database, storage, auth, image generation, WhatsApp, automations, analytics
- Claude Code handles the increasingly complex architecture locally via a GitHub sync that pushes back to Base44 for publishing
- Nano Banana Pro generates the infographic images
- A cloneable support app template ships with every new product - release notes, feature requests, bug reports, consistent from day one
The core principle visible throughout: eliminate, don't add. A good product in a saturated market often wins by taking features away, not adding them.
The architect-builder identity shift
Product leaders are being forced to become architect-builders. The distance between thought and shipped product has essentially vanished, and the role is less about writing requirements for engineers and more about designing AI workflows, maintaining principle files, and making high-quality decisions quickly - because the tools will happily execute low-quality ones at full speed.
The real constraints when technology stops being the bottleneck
When building is cheap, three problems remain:
- Working on the right problems - sustainable competitive positioning, not just novelty
- Getting attention - go-to-market at scale, the hardest remaining problem for most founders
- Redesigning the operating model - how roles, teams, and AI agents work together with shared organizational memory
Amir argues most founders underestimate problems two and three.
Learn and be curious - plus resilience
Amir's favorite Amazon leadership principle is Learn and Be Curious, essential when the landscape shifts monthly. Ask juniors, poke at AI features in every app you use, read books, play with new tools even if you discard them. He argues resilience should be added as a principle given the pace of AI-driven disruption - founders need the conviction to keep building through constant change.
Book Amir as a speaker
Amir regularly delivers keynotes and workshops on the topics covered in this episode - vibe coding and vibe debt, AI workflows for product leaders, Amazon's Working Backwards methodology, and leadership principles in the age of AI. He speaks at conferences, corporate leadership offsites, and founder events across Sweden, the Nordics, and Europe. Learn more about Amir's speaking topics and availability.