Artificial Intelligence has transformed software development. Today we can generate code, analyse documents, automate workflows and build assistants that would have been difficult a few years ago. But one thing hasn’t changed: businesses still need software that is reliable, secure, maintainable and aligned with real goals. That’s where engineering matters. I see AI as a powerful tool — not a replacement for thoughtful software engineering.
AI should solve problems, not create demos
Adding AI to a product because it’s popular rarely creates value. Before recommending it, I ask: does this genuinely improve the experience? Could a traditional solution be more reliable? What happens when the AI is wrong, and how will users verify the output? Sometimes AI is the right answer. Sometimes it isn’t. The goal is never to build “an AI product” — it’s to build a better product.
Where AI adds real value
AI helps most where people spend time processing information: intelligent search using natural language, knowledge assistants trained on internal documentation, customer support that escalates the hard cases, workflow automation, content assistance, and document analysis that extracts structured data from contracts, invoices and reports.
Where AI should be used carefully
AI is powerful and imperfect. For financial calculations, medical or legal workflows, security decisions and critical operations, careful engineering matters more than an impressive demo. In those cases AI should assist humans, with clear boundaries and oversight — not replace them.
AI engineering is more than calling an API
Many AI applications begin and end with sending a prompt to a model. Production systems need more: prompt and context management, retrieval strategies, data privacy, model selection, caching, latency, monitoring, evaluation, fallback behaviour and user feedback. Without these, AI quickly becomes unreliable.
My hands-on experience
I’ve spent the last couple of years using modern language models daily in real development work — and building with retrieval-augmented generation, function calling, embeddings and the Model Context Protocol. AI accelerates many tasks; engineering judgement decides which suggestions are actually used.
What AI cannot replace
Understanding business requirements. Designing architecture. Making long-term technical decisions. Balancing trade-offs. Protecting security and privacy. Building maintainable systems. AI is becoming an extraordinary engineering partner — not a substitute for experience.
If you’d like help putting AI to work in your product, see my AI Engineering & Automation service or start a conversation.