On the ACCENSEO accounting SaaS in 2025-2026, I was the only human developer facing a product no early-stage team usually ships alone: French accounting, taxation, banking (Open Banking PSD2), invoicing, 2026-2027 e-invoicing compliance, AI assistant. The cadence was unforgiving: if I dropped below 8K lines a month for 14 months, the product would not be ready before the e-invoicing deadline.

I ran a complete full-stack cycle every day: Next.js 16 + React 19 + TypeScript on the UI side, Prisma + PostgreSQL on the data side, Better Auth with MFA (email, TOTP, SMS), AI SDK for multi-provider orchestration (OpenAI, Claude, Gemini), Stripe for SaaS subscriptions, and a dedicated Node.js microservice for automated invoice retrieval. I personally handled the Open Banking PSD2 integration across 3 providers in parallel (GoCardless/Nordigen, Bridge, Qonto), the EDI Teledec submissions for VAT / IS / CFE / DAS2 / PAS, and a Chrome extension for automated form-filling on impots.gouv.fr. The pair-programming with Claude Code carried 80-85% of the implementation, but every file was reviewed, every tax computation validated by script, every feature put through non-regression tests.

234K lines shipped solo, 42 autonomous features, 382 API routes, 91 Prisma models, 6 differentiated roles, and the platform was running production-ready by the end of the cycle with 2026-2027 e-invoicing compliance reached ahead of the regulatory deadline.

What struck me on this project is that staying an operator as CTO changes everything: without the daily fullstack practice, I would never have framed a regulated domain in months. That posture - a CTO who still codes - is the one I want to keep on the next scale-up role, because it earns me the right to challenge senior technical decisions instead of delegating them.

Late 2025 I needed a Tailwind v4 obfuscation tool for several ACCENSEO engagements, and the only existing alternative was broken. Rather than crafting a workaround per customer, I decided to turn it into an OSS product - because a well-built npm package costs less to maintain than a dozen custom patches. The window was tight: 6 weeks before the community fell back to hacks.

I set up a TurboRepo monorepo with pnpm workspaces to drive 25 internal modules, 5 bundler plugins (Vite, Webpack, Rollup, esbuild, Nuxt module) sharing the same core engine, and a Commander.js CLI for framework-less usage. The stack: TypeScript 5.7, Babel AST to parse JSX/TSX/Vue/Svelte/Astro/Qwik, PostCSS to transform compiled CSS, magic-string to preserve sourcemaps, Vitest for the 295 unit and e2e tests, and VitePress for the docs. Strict typing across the codebase, builds via tsup with full sourcemaps.

82K lines of TypeScript published to npm in 6 weeks, 10 frameworks supported, automatic detection of Tailwind v3 vs v4, first Tailwind v4-compatible tool on the market. Picked up by external teams within the first weeks, organic mentions in tech newsletters.

That package is what kept me sharp on modern TypeScript outside customer time. I have made it a rule to ship at least one OSS per year for the same reason: a CTO who stops publishing code eventually loses the daily reading reflex. And it is also a concrete credibility signal when I discuss a scale-up role - people ask for the OSS, not the resume.

On the Fleurance Nature refit in 2017, I came in as Senior Software Engineer full-stack at Smile on a Magento Enterprise Edition 1.10 platform: 60 accumulated custom modules, 1,040 modified PHP files, 3 stores (Fleurance Nature France, International, Mincifine) to refit simultaneously. The complexity came from both backend (EAV pricing matrix, 4 customer groups, bidirectional ERP) and frontend (mobile responsive, 3 distinct themes, internationalisation).

I owned backend through frontend without a hand-off: server-side PHP 5.3 + Zend Framework + Magento EE 1.10, MySQL with EAV, Varnish as reverse proxy, and the ElasticSearch backport on Magento 1.10 nobody had done before (autocomplete, faceted navigation, virtual categories). Front-side I produced the responsive wireframes (mobile, tablet, desktop), wrote the graphic specifications for the 3 themes, integrated the WordPress blog via RSS parsing (512 articles migrated lossless), and shipped the responsive front-end theme on the 3 stores. I documented the functional specs across 7 versions (from 30 to 50 pages) as the business cases unfolded.

Production rollout across 8 environments with zero major regression, the 58-day post-launch warranty handled in person, and the refitted platform held for more than 5 years in production without any further rewrite.

That assignment locked in a reflex I now replay: legacy PHP depth + modern stacks are not 2 separate worlds. Being able to read a Magento 1.x or Symfony 2 codebase today gives me access to CTO advisory audits that strictly modern-TypeScript profiles cannot run. It is exactly what let me chain Symfony 3-7 then the modern TypeScript stack with no breaks.