Yes, we still need engineers

Despite advances in AI-assisted prototyping, production-ready software still requires extensive engineering work including backend support, testing, security, scalability, and compliance. A recent piece argues AI has not replaced software engineers, and companies are hiring more engineers even after improvements in AI-generated code. The author emphasizes that prototypes ease communication but are far from production-ready products.

Yes, we still need engineers Prototyping has done wonders for speeding up software delivery. I start with a clone of our UI repository and open Claude Code. I'll prompt it to help me add a button or whatever , set up some test data, and record a demo of a full workflow. It looks great in a Loom video But it's far from production-ready. We still need: - Backend support - Regression testing - Functional testing - Security testing - Performance testing - Scalability considerations - Edge case considerations - Architectural considerations - Accessibility requirements - Design system considerations - Maintainability considerations - User permissions considerations That's just off the top of my head What about: Lots more stuff - Logging, metrics, and tracing - Monitoring and alerting how do you find out it broke before users do? - Error handling and graceful degradation - CI/CD pipeline, deployment, and rollback strategy - Feature flags - Infrastructure provisioning, config management, and secrets handling - Rate limiting and abuse prevention - Retry logic and idempotency - Concurrency and race conditions - Backups and disaster recovery - Database schema design and migrations - Data validation and integrity constraints - Caching strategy and invalidation - PII handling, data retention, and deletion - Authentication - Authorization, roles, and permissions - Session management - Multi-tenancy and data isolation - Regulatory compliance GDPR, HIPAA, SOC 2, etc. - Audit trails - Dependency licensing - Data residency requirements - Localization and translation - Timezone, currency, and date/number formatting - Right-to-left support - Empty, loading, and error states - Offline and slow-network behavior - Responsive layout, mobile, and cross-browser quirks - Analytics and instrumentation - A/B testing hooks - Cloud cost / budget impact at scale - Code review - Unit, integration, and end-to-end tests and coverage - Technical and user-facing documentation - Runbooks, on-call, and support - Versioning, backward compatibility, and deprecation paths - I used Claude to generate the rest of this list My prototypes have eased the communication from idea to delivery better than any other tool, but it's still far from a production-ready product. I enjoy making them, and engineering enjoys referencing them, but anybody that thinks their vibe-coded prototype is ready for production is fooling themselves. A recent piece https://www.normaltech.ai/p/why-ai-hasnt-replaced-software-engineers discussed how AI is not behind mass layoffs, and may likely never be. After working with a team https://mattsayar.com/getting-a-job-in-2026/ rocketing forward with AI adoption in a space that needs protection from AI more than ever, I couldn't agree more. Indeed, it seems we're starting to hire more engineers even after the noticeable boost in quality in AI generated code in November's inflection point https://simonwillison.net/2026/Jan/4/inflection/ . Software is an industry intent on automating as much as possible. It usually automates the boring stuff https://automatetheboringstuff.com/ , which lets us focus on the fun stuff. And the fun stuff is solving real problems, not typing code.