The promise of AI in software development gets oversold. You’ve probably seen the demos — someone types a description of a feature and a complete, working application appears. The reality is more nuanced, and for businesses that need real, production-grade software, the nuance matters.
Here’s how AI actually fits into serious custom software development, and why it’s a multiplier rather than a replacement.
The Problem With the AI Hype Narrative
The framing you see in product demos is: “AI writes the code, you just describe what you want.”
For toy projects and demos, this works reasonably well. For production software that businesses depend on, it breaks down quickly:
- AI doesn’t know your business domain
- AI doesn’t know the edge cases your business encounters
- AI doesn’t understand your security requirements
- AI produces code that requires expert review to catch subtle problems
- AI can’t make architectural decisions that will hold up over five years
The output of AI tools needs expert review and modification. That expert is still a developer who understands Rails, knows how to design a database schema for your domain, and can tell the difference between code that looks right and code that is right.
Where AI Delivers Real Value
First drafts. The classic blank page problem — starting from nothing is the hardest part. AI eliminates that. “Write me a Rails model for an inventory item with validations for quantity and sku” gives me a starting point in 10 seconds. I modify it, but starting is the hardest part.
Repetitive patterns. Business software has a lot of repetition. Multiple CRUD features that follow the same structure. Multiple API endpoints that work the same way. Multiple report queries that have the same shape. AI handles the repetition efficiently, freeing me to focus on the pieces that require real judgment.
Exploring approaches. When I’m deciding between two architectural approaches, I can describe both to an AI and get a structured comparison. This isn’t replacing my judgment — it’s organizing my thinking. The final call is still mine.
Code explanation. When I inherit a codebase (this happens in maintenance work constantly), AI helps me understand what existing code does. “What does this query return and why does it use these joins?” is a question AI answers well.
Writing tests for known behavior. Once I know what code should do, AI can generate test cases quickly. I still write the tests for edge cases and business rules, but the basic coverage comes fast.
The Workflow Change
The biggest shift isn’t in any specific tool — it’s in how work gets distributed across a project.
Before AI tools became capable, a developer’s time was split roughly:
- 30% thinking and design
- 70% typing and implementation
With AI assistance:
- 50% thinking and design
- 30% reviewing and refining AI output
- 20% typing and implementation
The ratio of thinking to typing shifts dramatically. This is good — the thinking is where the value is. More thinking time means better architecture, more consideration of edge cases, more time invested in testing the right things.
The risk is that developers skip the thinking phase and let AI make decisions. That’s where projects go wrong.
What Custom Software Needs That AI Can’t Provide
Understanding your business. I spend significant time at the start of every engagement understanding how your business actually works — the workflows, the exceptions, the rules that aren’t in any manual. AI can only know this if I tell it, and telling it requires understanding it myself first.
Long-term architectural thinking. “Does this approach make the next year of development easier or harder?” is a question AI can help analyze but can’t answer with your specific context.
Production experience. Knowing what goes wrong in production, and building defensively against those failure modes, comes from experience. AI tools reflect training data, not lived production incidents.
Accountability. When something breaks at 2am, AI can help me debug but can’t be responsible for the outcome. Responsibility requires human judgment.
The Honest Bottom Line
AI tools make experienced developers faster. They don’t replace experience. They don’t replace judgment. They don’t replace the understanding of your business domain that comes from actually talking to you about what you do.
For clients, this means: working with a developer who uses AI tools well should give you faster delivery. It should not mean lower quality or less-expert work. If an AI tool is being used to shortcut the thinking, that’s going to show up in a year when the shortcuts accumulate into technical debt.
I use AI tools extensively. I’m also directly accountable for every line of code in every system I build. Those two facts coexist because AI is a tool in my hands, not a replacement for my judgment.
Building something that needs to hold up in production? Let’s talk about your project.