For decades, building software has been a game of telephone. A subject matter expert (SME)—a doctor, a logistician, a financial analyst—knows their field inside and out. To build a tool, they first explain their needs to a business analyst (BA). The BA translates that human expertise into formal requirement documents. A developer then translates those documents into code.
With every translation, vital context is lost. The developer builds what the documents say, not necessarily what the expert meant.
This “translation layer” has always been the single greatest source of friction, cost, and failure in software development. Today, advancements in AI-coding are shattering this model. We are finally entering an era where we can codify human expertise directly into a system.
The New Workflow: SME to AI
Modern AI models and advanced coding assistants act as a new kind of interface. An expert can now state their requirements in natural language: “I need a dashboard that cross-references our supply inventory with real-time shipping delays from our top three carriers and flags any order with less than a 48-hour buffer.”
The AI, trained on vast libraries of code and business logic, generates the initial application. This bypasses the BA’s translation entirely. The SME’s knowledge is mapped directly to functional code. This is revolutionary.
But a dangerous misconception has emerged from this new power: the belief that this first-pass output is a finished product.
The “Zero-Shot” Is Just the Beginning
What the AI produces is what we call a “zero-shot” solution. It is the literal, direct codification of the subject matter expert’s knowledge and instructions. It’s a brilliant, raw artifact—but it is far from a completed application.
The SME is an expert in their domain, not in software architecture.
- They won’t ask for database indexing, but they will be furious when the app freezes with 10,000 users.
- They won’t specify error handling for a dropped API connection, but they will call the system “broken” when it fails.
- They won’t think about security, scalability, logging, or how this new tool integrates with the company’s five other legacy platforms.
The zero-shot solution is a perfect expression of the SME’s “happy path.” It’s a blueprint of the core logic, but it’s not a house. It’s not usable, and it’s certainly not reliable.
The Developer as Solution Engineer
This is where the role of the developer evolves. AI isn’t replacing developers; it’s killing the “coder-as-translator” role and elevating the “developer-as-engineer” role.
The zero-shot output is the new “requirements doc.” A developer with solution engineering experience must now take this raw, AI-generated expertise and turn it into a usable and reliable entity. Their job is no longer to translate, but to architect.
Their questions are:
- Reliability: How do I harden this? What are the edge cases?
- Scalability: How do I make this performant for one user and 100,000 users?
- Integration: How does this plug into our existing authentication, data lakes, and CI/CD pipelines?
- Security: How do I protect this from being a massive vulnerability?
AI lets us capture the what from the expert faster than ever before. But we still need skilled engineers to build the how—the robust, secure, and integrated solution that the business can actually depend on. Moreover, a single application will require fewer software engineers to complete, but enabling this workflow means that we will have many more applications in the backlog than ever before. How many software engineers will be needed will be a factor of how fast or slow a business wants to move.