Building with AI
After harness engineering we are now in the beginning of the IA Factory era. Take a look at why Software Factories Fail (or in video).
A forceful critique of lights-off coding factories, arguing that harnesses cannot replace accountability and system understanding : safe agentic coding still needs product review, architecture, program design, and vertical slices.
Reactions:
Can software factories actually work? → A grounded map of software factories, where agent loops help and where human review still wins
Software Factories, Light and Dark → A useful vocabulary for agent loops, harnesses, light factories, and the risks of lights-out software delivery
Lean, GenAI et Delivery Tech - entretien avec Soufiane Keli : L’IA implique que l’on soit en mode Lean sur chacune des étapes → Pourquoi l’IA impose des boucles Lean plus courtes, du cadrage produit jusqu’au delivery et à la modernisation legacy
Agentic Engineering at Zalando: a snapshot → Zalando’s playbook for scaling LLM access, cost controls, prompt caching, and agent adoption across 250 teams
Building an automated software development lifecycle with AI agents → A concrete ADLC pipeline with specs, planner and builder agents, independent verification, and human release control
How we built a software factory to drive Astro’s GitHub issue count to zero → Cloudflare’s agent triage pipeline for Astro: reproduce bugs, diagnose fixes, ship previews, and cut issue backlog
Spec-driven development: When the spec becomes the product → Why AI-native development shifts value from code generation to intent, specs, coordination, and traceable product decisions
Why Software Factories Fail → A forceful critique of lights-off coding factories, arguing that harnesses cannot replace accountability and system understanding
AI & People & Process
- “Code was never the hard part” is an insult to all programmers → A sharp rebuttal to the idea that coding is trivial, and why craft still matters with LLMs
- AI is removing the middle class of software engineering → Why AI accelerates weak engineering cultures into chaos, and raises the bar for technical judgment
- On a mis dix ans à enterrer le mythe du 10x engineer. L’IA est en train de le déterrer. → Une réponse nuancée à la polarisation du métier par l’IA, sans retomber dans le mythe du héros
- Understanding is the new bottleneck → Techniques for keeping up with agent-written code: explainer docs, quizzes, micro-worlds, and shared review spaces
Learning From the Field
- How we raised mobile end-to-end test stability to 98% (2026) → Shopify’s stricter mobile E2E API, using visual matching and guarded test steps to fight flakiness
- We replaced Redis with MySQL for inventory reservations—and it scaled (2026) → Shopify’s MySQL reservation design with SKIP LOCKED, composite keys, and hard-won observability lessons at peak traffic
- You should never be angry at work → A practical argument for turning engineering anger into care, credibility, and better decision-making at work
Architecture
- Git à toutes les échelles → Les compromis d’infrastructure derrière l’hébergement Git à grande échelle, des packfiles aux architectures distribuées
- PostgreSQL for Everything → The case for simplifying stacks by pushing queues, search, analytics, and caching further into PostgreSQL
Tooling & Ecosystem
- TanStack - Hotkeys → A headless hotkey utility for React apps, with scopes, sequences, and accessible keyboard interaction patterns
- Hunk → Terminal diffs for humans & agents
Security
- AI assistant hacks gym website in first known Australian autonomous cyber attack → Broken Object Level Authorization is the first of the Top 10 OWASP Api Security, we talk about it on [Culture API][/culture-api/]
Trends
- How Notion hit 100M+ users by hiring fans instead of marketers → Notion’s early growth machine: fan communities, templates, ambassadors, and patient product-market fit before scale