TypeScript, Python, and the AI feedback loop changing software development (github.blog)

GitHub Next lead Idan Gazit argues the biggest AI-driven change in software development isn’t just productivity boosts — it’s shifting what developers choose to build with. GitHub’s 2025 Octoverse shows TypeScript surged 66% year‑over‑year and overtook both JavaScript and Python as the most-used language on GitHub. Gazit attributes this to a feedback loop: models perform best in popular languages (TypeScript, Python, Java, Go), and statically typed languages give AI-generated code clearer guardrails, reducing hallucination surface area and making generated code easier to validate. That feedback loop is reshaping tool and language selection: teams now factor in how well AI will support a stack. Unexpected winners like Bash (+206% growth in AI-generated projects) illustrate that AI removes the pain of unpleasant languages, letting developers use “duct tape” tools without manual drudgery. Enterprises are seeing juniors ramp faster and seniors shifting from writing code to validating architecture and debugging, while tooling and type systems define the surface area for agentic assistance. Looking ahead, WebAssembly and cross-target runtimes could decouple language from deployment, making portability and ecosystem leverage — package depth, tooling maturity, model familiarity, debugging ergonomics — the primary battlegrounds. The practical takeaway: optimize stacks for shared human–machine leverage (types, tooling, model support), not language loyalty.