Product-First, Add-AI-Later
Design the end-to-end product experience first, then add AI to solve specific problems — never retrofit AI into a broken flow.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 3
- Confidence
- 85%
A product principle for building with AI: start from how the whole product works end to end and what the user experience should be, then add AI to solve specific problems inside it. It warns against the reverse — building cool AI tech and trying to retrofit it into someone else's product. Use it whenever excitement about AI capability tempts you to lead with the tech.
Origin
Anton's biggest career lesson, from being first employee at Stockholm education-AI startup Sana Labs. They built a powerful personalized-learning AI API but the model required customers to 'switch out the engine' of an existing product — retrofitting that worked poorly. The company pivoted; Sana now does great, but not on that product.
Core principles
- 01Start with the end-to-end product and user experience
- 02Add AI to solve specific, identified problems — not as the centrepiece
- 03Retrofitting AI into an existing product built around something else is very hard
- 04Cool tech is not a reason anyone will use or keep the product
How to run it
- 1
Map the whole user journey
Define how the product should work end to end and what the ideal user experience is, independent of any AI.
Pro tip Ask what the big picture of the user is before deciding where intelligence belongs.
- 2
Locate where AI actually helps
Identify the specific problems within that journey where AI meaningfully improves the experience, and add it there.
Pro tip Adding a well-scoped AI feature into a working product beats an AI-first product bolted onto nothing.
Watch out An AI capability that forces customers to re-architect their existing product will struggle to get adopted.
- 3
Validate real demand, not novelty
Check that the AI is solving a real, recurring problem people care about, not just a novelty they try once and forget.
Pro tip Anton floated a 'Lenny mode' product-coach that interrogates the builder: what problem, how many people have it, how much it matters, what's the experiment plan.
Watch out Novelty products get used briefly then abandoned when they don't solve a real problem.
In the wild
Sana built an advanced AI API for personalized learning that required existing edtech products (e.g. a Duolingo-like app) to swap in the AI engine. The retrofit was too hard and the product wasn't successful; the company later succeeded on a different product.
→ Lesson: start with the end-to-end product, then add AI to specific problems — don't retrofit.
Common mistakes
Leading with the tech
Being so excited by cool AI that you assume everyone will love it, forgetting that people only keep tools that solve a real problem for them.
Retrofitting AI into an existing engine
Asking customers to swap out the core of a working product to insert your AI creates painful, low-adoption integrations.
Is it for you?
Best for
Founders and PMs building AI products or adding AI features to existing products
Not ideal for
Pure research or capability demos where adoption isn't yet the goal
From the transcript
“you have to start with like how is this product working end to end and then add AI or think where should we add AI”
“it's a very hard like retrofitting like oh you have to switch out the engine and put in this Ai”
From the episode
Building Lovable: $10M ARR in 60 days with 15 people
Anton Osika (co-founder and CEO)