Software-as-a-Prompt: How AI is enabling on-demand software (www.siddharthbharath.com)

Last week’s anecdote — a speaking coach’s custom video-analysis app built, deployed, and live in under ten minutes using Lovable and Google’s Gemini — illustrates the rise of “software-as-a-prompt”: AI models that generate working applications on demand from natural-language descriptions. Models like Gemini 2.5 Pro, GPT-4o and Claude can scaffold projects, write multi-language code, create front ends, database schemas and basic integrations; platforms such as Replit AI, Lovable.dev, Bolt.new and Cursor wrap that capability in a chat/agent workflow that handles project files, iterative feedback and deployment. Low-cost real-world examples include a $0.x app for video feedback and a $75 vehicle-document system built with Cursor, Claude, Next.js, Supabase and Stripe — no prior engineering required. That capability matters because it dramatically lowers time and cost to build niche software, threatening feature-bloated, expensive SaaS for point-solution use cases and opening opportunities for AI-first startups and internal tooling. Important caveats remain: AI-generated code can have quality, maintainability and security gaps (e.g., injection or auth flaws), and debugging or scaling often needs human engineers. Enterprises are responding with hybrid approaches (Salesforce Einstein GPT, Microsoft Copilot) and the market will likely see rapid verticalization: quick wins for small, single-purpose tools now, with enterprise-grade displacement taking 3–5 years as reliability, security and maintenance patterns mature.