When AI Writes the Code: Will Developers Think More and Type Less?

"I think we will be there in three to six months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing essentially all of the code." — Dario Amodei, CEO and Co-founder of Anthropic (developer of Claude) Source | Full Discussion

Imagine describing an app idea to a friend—and it instantly turns into working code. That’s the future Dario Amodei sees coming fast. Sure, tech loves a flashy prediction—self driving cars, anyone? We’re still waiting on those.

Whether AI writes 90% of our code in six months or 50% in a year, is not important: it’s already here. AI is already writing significant amounts of code, and its role in software development is only growing. At the rate AI models and agent technology are improving, in 12months it will be writing most of the code.

At the same time, as software integrates into everything, we’ll be writing more code and pushing the boundaries of complexity of the systems we build—making AI an indispensable co-creator.

Its not a “what if”—it’s a “get ready now.” So, what does this mean for software development? Here are some thoughts.

Coding’s Evolution: From Machine Code to Prompts

Programming languages have always evolved and moved to higher levels of abstraction. Making the complex, simple. 

Machine Code → Assembly → High-level Languages

 Each shift made programming more accessible, intuitive, and faster. Learning Python or JavaScript is easier (and enjoyable) than learning Assembly :-). You may call that subjective and yes, there are cases where low level languages are the right choice. But objectively, we know the demand for JavaScript and Python developers are much higher. Why? More modern software are written with those languages than Assembly or COBOL.

We also rely on sophisticated Integrated Development Environments (IDEs), community resources like Stack Overflow and GitHub to speed up the development process.

 AI-generated code is essentially the next logical progression that combines all this and more.

Lowering the entry barrier

High-level languages like Python and JavaScript lowered the barrier by abstracting underlying operations like memory management, pointers, and registers.

Similarly, AI coding assistants like GitHub Copilot and LLM’s in general like Claude or Gemini can abstract away syntax, boilerplate, design patterns, and even some architectural considerations required when using today’s high-level languages.

 People who previously couldn't or wouldn't learn detailed programming constructs can now describe requirements in plain language and see immediate results, significantly widening access.

Democratizing software development

AI will make software creation more about ideas and less about specific coding skills. People can now describe what they want in plain language and get working code. That’s not just productivity—it’s democratization.

More people, including non-technical domain experts (scientists, educators, artists, entrepreneurs), will participate directly in software development. For example, they could easily build custom software tools for their work, to perform repetitive tasks.

This is a positive outcome and one that could turbocharge speed to market, delivering products faster than ever before.

 But this requires rethinking how we hire and upskill tech talent.

Rethinking the role of software developers

The software developer role is not going away, it is evolving from being the "core builders" of software—writing line after line of code—to becoming "thinkers," "teachers," and "reviewers." These aren’t entirely new hats; engineers already wear them to some extent. But as AI handles more of the grunt work, these skills will take center stage. Imagine guiding AI tools to produce better outputs, teaching them context, or reviewing their code for quality and intent—tasks that demand not just technical know-how.

New skills & responsibilities

Engineers will evolve to supervise, refine, and ensure the quality of AI-generated software. It’s important to remember that while AI might write the code, most software will ultimately serve humans or create value for humans. Hence AI will require input from humans to set the goals or use cases of the software.

Some of the skills that will become more important include

• Deep domain knowledge (understanding user requirements)
• Critical Thinking (ensuring alignment with goals)
• Working with AI (understanding effective prompting)
• Innovation & creativity (generating ideas, novel use-cases)
• Clear communication (defining problems, goals etc of systems)

All this does not mean programming skills will become obsolete; a teacher or reviewer still needs to understand code deeply to do their job well. More abstraction can lead to a risk of becoming detached from core principles. Engineers should still maintain a strong conceptual understanding of areas such as UX design or how things work at lower levels (for debugging and optimization), even if they may not be doing the hands-on work.

Rethinking hiring: Beyond the coding test

Organizations will need to change their hiring practices to assess for the right skills when hiring software developers. Coding tests will remain relevant, but their importance will diminish significantly. Instead, hiring should shift toward evaluating higher-order skills and capabilities such as the ones listed above. For example, more emphasis should be on system design, scenario-based problem-solving, innovative thinking, AI interaction skills (e.g. prompt engineering, AI coding tools) and strong communication/collaboration skills. Adaptability becomes key. Engineers who are eager to learn, experiment, and adapt to evolving technologies will become more valuable than others.

The main takeaway from Dario’s message is that we all need to continuously upskill in understanding AI, not just as a coding tool but as a core partner in software delivery. Use AI as a thought partner, learn their strengths and weaknesses.

P.S. I teamed up with Grok 3, ChatGPT 4.5, and Claude 3.7 to write this. By using different models you get different insights, writing styles and thinking patterns. Its a good way to get to know this new "alien intelligence" in town. Whether its coding, writing or learning, its always more enjoyable to collaborate with a diverse team of thinkers :-)