When we talk about AI and software development, the focus is usually on productivity. Faster pull requests, fewer routine chores, auto-generated tests, psychedelic autocomplete, and more. But that’s the shallow end of the change curve, according to Idan Gazit, who leads GitHub Next, a team that supports Copilot and GitHub’s long-term research and development.
deeper changes are occurring in front One line of code will be written.
“AI doesn’t just change the way we write code,” Gazit says. “We’re starting to choose what we build with in the first place.”
That shift is already evident in this year’s Octoverse report. In 2025, TypeScript will overtake JavaScript and Python to become the most used language on GitHub. This is a 66% year-over-year jump and the largest language movement in more than a decade.
But this story isn’t about TypeScript beating Python. That means AI is starting to shape language trends from the inside out.
The changes in the last generation are where Running code: cloud, containers, CI/CD, open source ecosystem. Next is about that. What is the code made of?and why those choices suddenly have different stakes.
TypeScript passed Python. But the real story is why.
Developers usually don’t switch languages just for philosophical reasons. Switch languages when something significantly speeds up, simplifies, or reduces risk. And what feels “easy” increasingly has to do with how well AI tools support working with that language.
“Statically typed languages provide guardrails,” Gazit says. “When an AI tool generates code, it needs a way to quickly know if that code is correct. Explicit types provide that safety net.”
The input language reduces the surface area of the hallucination. It also gives the model more structure to reason about during model generation. It’s not a theoretical advantage. This became a behavioral signal in the data.
- AI models tend to perform better in languages that expose information about correctness, such as type systems.
- Developers who use AI tools are more likely to adopt typed languages for new projects
- The more teams rely on AI assistance, the more language choices they have. Determining AI compatibilityIt’s not just a personal preference
This change establishes a feedback loop.
AI assistance is a new consideration for developers when choosing languages and frameworks
AI models are best at writing code in popular languages like TypeScript, Python, Java, and Go, to name a few.
“If a model has seen a trillion examples of TypeScript, but only a few thousand examples of Haskell, that model is going to be better at TypeScript,” Gazit says. “That changes your motivation before you even start coding.”
When an AI tool generates code, it needs a way to quickly know if the code is correct. Explicit types provide that safety net.
Idan Gazit, Head of GitHub Next
Before AI, language choice was a trade-off between runtime, library ecosystem, and personal fluency. After AI, new constraints emerge. How effective will the model be if we choose this language?
“Python is the dominant language for machine learning, data science, and model training,” Gazit says. “Why not go with the one that already has the most robust frameworks and libraries? Those are the wheels you don’t have to reinvent.” So TypeScript can’t win. against Python; each wins in situations where it is the right tool for the job. and AI adds value.
The surprising winner of the AI era: the “duct tape” language
One of the most unexpected signals in the Octoverse data wasn’t about TypeScript or Python, but about Bash.
shell script saw AI-generated projects grew +206% year over year. So what does that give?
Because AI creates painful language tolerable.
“Very few developers like writing Bash,” says Gazit. “But everyone needs it. It’s the duct tape of software. And now you can have agents write the parts you’re uncomfortable with, so you can use the right tool for the job without having to consider the tradeoffs.” The problem is solved when AI automates the tedious layers of programming. “Do you enjoy this language?” And become “Should I consider using it if I don’t have to write the code myself?”
Few developers like writing Bash. But everyone needs it. Software duct tape. And now that you can have your agent write out the parts you’re uncomfortable with, you can use the right tool for the job without having to consider the tradeoffs.
Companies aren’t asking themselves, “Should I implement AI?” already. They’re asking, “What happens after we do that?”
“A lot of companies are sitting on the sidelines, waiting for it to get warm enough to jump in,” Gazit said. “Now they realize the value. Junior developers grow faster and senior developers work less and spend more time on architecture.”
This has the following secondary effect:
| Before AI | After AI |
| Skills measured in lines of code | Skills assessed through verification, architecture, and debugging |
| Junior shipping is slow | Juniors ship faster than seniors review them. |
| Senior developers write the most difficult code | Current senior developer judge most difficult code |
| The tools were mostly a matter of preference: IDE, linter, build setup. | The tool is currently surface area What AI can work on: The wrong stack can block or limit agent assistance |
Typed languages accelerate this change. The stronger the safety rails, the more work can be delegated to automation.
The next horizon: When language is no longer a constraint
Currently, runtimes are still fragmented, so language choice is important. JavaScript is required in your browser. The model requires Python. The firmware is assumed to be C.
But it is already falling apart.
“WebAssembly is starting to change the rules,” Gazit says. “If you can target Wasm in any language and run it anywhere, that removes one important consideration when choosing a stack.”
Combine this with AI-generated code and you have a likely future.
- Developers write in Rust (or Go, or Python)
- AI generates code in that language
- Compiler target Wasm
- The same code runs on the web, edge, cloud, and local sandbox
That’s not a future where TypeScript wins. it is A triumph of portability This is a natural extension of the rise of containerization over the past decade as a way to package and run software.
Competition between languages may end up being less about syntax and more about leveraging ecosystems such as package depth, tooling maturity, model proficiency, and debugging ergonomics. We’re not quite there yet, but early signs of portability from AI-driven tools to Wasm suggest it’s coming sooner than most teams expect.
What should developers actually take away from this?
This isn’t a “Learn TypeScript Now” blog (though there are certainly plenty of them out there).
Important signals are:
| shift | what it really means |
| Increase in input languages | AI benefits from structure |
| Python remains dominant in AI | Ecosystems outlast language and framework fads |
| +206% increase in shell scripts | AI not only removes productivity barriers, but also pain barriers |
| Companies rapidly adopting AI | The definition of “senior engineer” will change next. |
| WebAssembly Maturity | Language loyalty is replaced by language interoperability |
The point is not about stack switching. it’s about Optimize leverage, not loyalty.
The languages and tools that survive the next decade will not be what developers love most, but what they offer to developers. and The machine’s advantages are the most shared.
Want to stay ahead of the curve?
Read the latest Octoverse report and consider trying Copilot CLI.
Other resources:
author
