Even before AI code generators emerged, IT leaders sought ways to improve developer productivity. Platforms like 4GL, low-code/no-code, and configurable SaaS helped IT deliver more applications, reduce the developer skill set required to release enhancements, and improve software quality. These tools enabled IT to develop entire classes of applications, analytics, and integrations that couldn’t be built easily or cheaply by coding in Java, .NET, and other programming languages.
“Software has long been treated like infrastructure: built to last, hard to change, and expensive to replace, says Chris Willis, chief design officer and futurist at Domo. “That model is giving way to a future with more applications that are smaller, faster to build, and created to solve a specific job before getting out of the way.”
Code gen, vibe, or write a spec?
GenAI models are the next accelerators for software development. The first tools were copilots for coding assistance, followed by LLMs for generating code snippets. I used code-generation tools to develop regular expressions, extract information from web pages, and categorize data as steps in an app migration. They wrote code that I no longer had the time or skills to develop on my own, but it still required significant work to fix defects and integration issues.
We’re now in a second-generation phase of AI software development, with platforms like Amazon Q Developer, Appian AI-Assisted Development, Bolt, Claude Code, Cline, Cursor, Gemini Code Assist, GitHub Copilot, Kiro, Lovable, OpenAI Codex, Pave, and Replit.