1: Outside of research, TypeScript becomes more popular than Python for AI related work.

As of August 2025, Github's analytics show that TypeScript has overtaken both Python and JavaScript as the most commonly used language. Reasons given included the fact that many project builders now bootstrap new projects with TS as the default language.

I think TypeScript has another advantage over Python in that its type system is more mature, which plays well with LLMs. Python has optional type annotations, but the ecosystem is still working through the growing pains of maturing from a dynamically typed language into a statically typed one.

To be fair, I'm less sure of this statement than with many of the others in this list. Python was already hugely popular, even before the mass adoption of LLMs. Still, Github's analytics are nothing to sneeze at. Especially considering that they're analytics and not just survey data.

2: Go continues to grow faster than Rust.

I've been keeping an eye on Go and Rust for a while. While I enjoy working in both languages, Go's focus on simplicity and productivity while still running on a performance tier above C# and Java makes it the best of both worlds in my view. Rust has its place obviously, but I think Go brings more value to the software engineering market as a whole.

In addition, its simplicity means LLMs have an easier time generating Go code. Where the main drawback of Go was once that it required more boilerplate than more sophisticated languages, I think LLMs have largely negated that issue.

3: "Agent Optimization" becomes a thing.

There's a joke that SEO killed SEO. I don't know if there's a term for it yet, but I predict that we'll soon optimize web content for LLM's with the hope of being cited as a reference. Professionally, there will be as much reading of tea leaves and selling of snake oil as with SEO, but the emergence of some best practices seems likely.

I see the irony in predicting the above in a written blog post. I think providing "proof of life" over the internet is also going to become more important.

4: Innovation continues in the compiler space.

In 2026, Angular and Svelte will officially turn ten years old. This means that React, Vue, Angular, and Svelte will have been around for a decade or more. The notoriously turbulent reputation of the JavaScript ecosystem no longer holds water.

In March of 2025, the TypeScript team announced that they were re-writing the TypeScript compiler in Go. The React team announced "React Compiler", a tool for improving developer ergonomics in ways that a runtime library can't. Svelte, which has long regarded itself as a compiler over a library, remains one of the most ergonomic tools for writing richly interactive web apps to this day.

For developers, updates to compilers means there should be less cognitive load when working with these libraries. It also means that training data for LLMs will remain relevant for longer. I think that this progression is a net win for everyone.

5: Best practices for coding with AI will solidify.

At the time of this writing, working with AI and LLMs is still the wild west. Agents, MCP servers, and auto-complete have all changed the way we write code.

Generated code was a big part of my life in 2025, largely from teammates. In many cases, I had no idea that I had reviewed generated code until well after it had been merged. I can also think of half a dozen PRs I've rejected due to the inclusion of inline comments with emojis.

My end of year takeaway with regard to generated code is that, as humans, we need to make submitting poorly written code more expensive than not. At the end of the day, a human still needs to roll up their sleeve and debug a program. We need to estimate tasks, predict the impact of our changes, and communicate the intent of the code we author with a professional level of precision. I think we're still figuring out how to do that.

Final Thoughts

I genuinely didn't expect AI to be mentioned in all five of my predictions. I think it illustrates a common theme: AI's role in our workflows needs consideration moving forward. (Not exactly a hot take.)

I've read a lot of posts recently that seem to focus on quantity. "I wrote more apps!", "I shipped more features!" These read as noise.

2026 is going to be very different for me professionally. As we head steadily toward it, I aim to use the bandwidth afforded to me by AI for quality, not quantity. Better documentation, quality of life scripts that count, and especially code reviews. As workflows crystalize and the novelty of these new tools wears off, my hope is that we'll develop a greater sense of signal to noise ratios, and ultimately produce better software.

Happy 2026!