lee studio

LEE Studio

Agentic product studio. I drive products from concept to release.

Forward deployed product, design and engineering, delivered by one senior operator and a workforce of AI agents. The speed and agility of an AI-native startup, with the seniority enterprise clients expect.

What I do

AI-native speed, enterprise standards.

Twenty years designing, building and leading products: an agency and two B2B SaaS companies I co-founded and sold, product organisations I have led in scale-ups, and enterprise software delivered again and again.

Building software people actually use takes four things at once: the product judgment to frame the problem, the design to make it obvious, the engineering to ship it fast, and the enterprise fluency to get it through the organisation that adopts it.

LEE Studio is where I fold this unique senior skill set in one AI-driven delivery practice.

Testimonial

“Julien understood very quickly our business and was able to propose and deliver a bespoke product that matches perfectly our marketing needs. Despite the complexity of the project, the product was delivered on time and under budget. I will be happy to do many other projects with Julien.”

Stephan LavolleeCIO at JCDecaux

How can I help

Three ways companies work with me

Off-roadmap builds

For B2B platform vendors.

Your biggest customer is asking for something your platform does not do. Building it would bend your architecture or defocus your teams. Saying no puts the renewal at risk. With AI, bandwidth is no longer the constraint. Focus is. I build it alongside your core, not inside it.

Bespoke featureResellable add-onPlatform ecosystem

When a bespoke feature turns out to serve more than one client, I package it as an add-on the next account can buy. Your one-off becomes a product line, and your product starts behaving like a platform.

Pilots into production

For large companies.

You have an AI prototype that impressed everyone in the demo and never reached anyone’s daily work. The model was never the problem. The problem was that nobody framed it, defined what done meant, or built the path to production.

I take one pilot and finish it: reframed, designed, built, instrumented, adopted, handed over.

First versions

For companies of any size.

A new product, a new internal system, a new AI capability inside what you already sell. Too important to postpone, too small to win the prioritisation fight, too expensive through an agency.

I build version one and the system to keep it running, then your team takes it forward.

Testimonial

“Julien is not only reactive and customer oriented, he is very flexible and can adapt easily and quickly to any situations and to the specificities of our projects (quick prototyping, proof of concepts...). Julien’s best quality is something you cannot find easily, even with the best professionals, it’s creativity.”

Nicolas HauvillerInnovation Director at Amadeus

Live projects

Latest builds

An AI-native layer where a team’s ways of working become executable. Methods are codified as flows, agents run the steps asynchronously, humans decide at the gates, and every decision is written as plain markdown into the client’s own repository, so nothing is locked in. Concept, product design, brand, interface, code and deployment: mine.

  • Next.js, React, TypeScript
  • Tailwind, Radix, shadcn/ui
  • Supabase for data and auth, Vercel for hosting
  • Vercel AI SDK across Anthropic, OpenAI and OpenRouter, so the product is never tied to one provider
  • GitHub App integration through Octokit, reading and writing directly into customer repositories
  • MCP server, so external agents can query the memory
  • Trigger.dev for background jobs and scheduled maintenance runs
  • Self-hosted SearXNG for web retrieval, Readability and Turndown for content extraction
  • Zod for schema validation, Vitest for tests, Resend for transactional mail

A cross-border vehicle arbitrage platform for the European market, where the product’s job is to surface where a car is worth more than it costs. Opportunity discovery, pricing data and a four-language interface.

  • React, Tailwind, Lucide
  • Supabase backend, Vercel hosting and serverless functions
  • Four-language internationalisation built without a library
  • Scheduled multi-source crawlers orchestrated through Apify, pulling listings from national marketplaces on a fixed cadence
  • A normalisation engine that reconciles makes, versions, trims, mileage and condition fields across markets into one comparable record
  • AI quality analysis over each listing, scoring condition and disclosure risk and explaining the score in plain language
  • An import cost calculator covering every source and target market pair: transport, import tax and registration, so net profit is computed, not estimated

A training, nutrition and recovery tracker that treats progress as three habits rather than one, where the product’s job is to suggest the right session for today and keep training, food and rest in one place. Any workout you build with an AI imports in one tap, and the whole app can be driven from Claude or ChatGPT in plain language. Concept, product design, brand, interface, code and deployment: mine.

  • React, TypeScript, Vite
  • Hand-written CSS with custom-property theming, no UI framework
  • Installable, offline-first PWA: a Workbox service worker precaches the app shell, and a local write queue reconciles with the server so it keeps working with no connection
  • Supabase for data and auth (Postgres with row-level security, passwordless email one-time-code sign-in), Vercel for hosting and serverless functions
  • An MCP server so Claude and ChatGPT can read and write a user’s own data — OAuth and personal-access-token auth, each call resolved to a short-lived per-user JWT (jose) so Postgres row-level security, not app code, guarantees one user can never touch another’s data
  • Recovery-aware scheduling engine: per-workout rest tags and recovery windows decide what to suggest each day against an ideal week
  • Workout import pipeline: build a session with any AI and import the JSON in one tap, validated with Zod
  • AI macro estimation for food from a photo, a name or ingredients, with per-ingredient recompute
  • Zod for schema validation, Vitest for tests, Resend for transactional mail

A design-led publication for people who love the machines of the 1990s. Editorial thesis, visual identity, photography direction, site and content engine.

  • Hand-written HTML, CSS and JavaScript
  • No framework, no build step
  • One serverless function for contact, on Vercel
  • Custom typography and iconography

Clients

STARTUPS

LentiArbautoNinetiesTrained

ENTERPRISE

AmadeusXeroxFujifilmDellVeoliaJCDecaux

Who am I

Julien Lescure

Portrait of Julien Lescure

Director of product teams and interim CPO across European and US markets. One of 50 coaches trained in person by Marty Cagan on the Product Operating Model. Today, Operating Partner for Bpifrance portfolio companies, AI-Native Product Partner through julienlescure.io, Founder and Principal at LEE Studio.

LEE Studio is where I build. julienlescure.io is where I lead Product, Design and Engineering organisations.

Contact

Let’s talk

Tell me what you want built, or what you are weighing up. I will tell you straight whether I am the right person for it.