Will AI Really Write 90% of All New Code Within 6 Months? Probably
Dario Amodei, CEO of Anthropic, recently claimed in an interview that "AI will be writing 90% of the code" within 3–6 months. This has sparked intense discussion about software development's future, t
If you take away only one thing from this post, take away this: AI coding is just another type of abstraction. And as with every other major abstraction-driven shift in programming over the past 50 years, it will probably lead to more employment overall.
Abstraction Evolution in Programming History
We all take high-level programming languages for granted today, but in the ‘70s abstractions were looked at by assembly programmers like senior devs are looking at “vibe coders” today.
The skepticism followed familiar patterns: "You can't abstract away the machine." "Real programmers write assembly." "High-level languages will make code slow and bloated." These concerns echo today's debates about AI-assisted coding, where experienced developers worry about losing touch with fundamental programming concepts. Yet history shows these transitions ultimately expand capabilities rather than diminish them.
From Assembly to High-Level Languages
The transition from assembly language to C in the 1970s represents one of the most significant shifts in programming abstraction. To put it simply: C is easier to program in, compared to Assembly. Being easier to use, C allowed you to write programs faster. This transition didn't eliminate programmers but dramatically increased their productivity by abstracting away machine-specific details.
The productivity gains were substantial—while assembly required meticulous instruction-by-instruction coding, C offered constructs like control flow statements that expressed complex logic concisely. This "translates to faster development cycles compared to writing everything in Assembly"7.
Did this lead to the death of software engineering as a career field? Of course not, instead it redefined their focus toward higher-level problem-solving rather than low-level implementation details.
Object-Oriented Programming
Then came the rise of object-oriented programming, and in particular Sun’s “Write once, run anywhere” (very successful) campaign to push Java onto basically everything for almost every use case. There are four pillars to object-oriented programming, and abstraction is number one.
This transition also didn't replace programmers but transformed how they conceptualized problems. By modeling real-world entities and hiding implementation details, OOP enabled larger, more complex systems to be built with improved productivity.
Boilerplate Automation: The Precursor to AI-Generated Code
Modern development has already embraced significant automation of repetitive code through the use of boilerplates and other templates. Low- and no-code platforms began to pop up, and containerization and infrastructure-as-code dramatically increased the ability for smaller teams to build huge products.
And yet again, software engineering employment increased.
AI as Abstraction
Historical data shows that each major abstraction shift has substantially increased programming productivity and code volume. Increased efficiency, scalability, and accessibility expanded access to software as a solution for business and market problems, driving employment in the sector.
Just as C allowed for more portable, reusable code compared to assembly, AI-assisted coding could enable another quantum leap in productivity and code volume.
Let’s apply our observations of historical changes in code abstraction to AI and see if our assumptions hold. What dynamics are at play here?
Lowered Barriers Increase Participation: AI coding tools make it easier for non-experts or hobbyists to start building software. This means more people can produce code—even if it's just combining auto-generated pieces or tweaking them. While the fraction of hand-written code produced might drop, the absolute amount of code could well increase.
Productivity Amplification: Developers using AI tools can rapidly produce boilerplate or repetitive code. This lets them focus more on the creative or complex parts of software. With more projects being started and maintained, the overall volume of code—both artisanal and auto-generated—grows exponentially.
Complementary Roles: Even if 90% of the code comes from AI, human developers will remain essential for architecture, debugging, and integration. These critical tasks stay highly artisanal. AI-generated code handles the heavy lifting of repetitive work, but human insight remains crucial for making sense of and improving that output.
So what am I predicting? Way more code. Sure, 90% of code will be written by AI, but if the sheer volume of code increases by an order of magnitude, or if AI is embraced as a new form of abstraction, or a combination of both, hitting that number won’t be all that surprising.
Is Amodei's Timeline Realistic?
Here’s the full quote from Amodei:
"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"1.
This aggressive timeline differs significantly from historical abstraction transitions, which typically took years or decades to fully materialize. Additionally, Amodei suggests that "eventually all those little islands will get picked off by AI systems", potentially indicating he has a more complete replacement than this analysis suggests.
Industry expert Vineet Vashishta offers a counterpoint: "We'll be using AI driven tools for 90% of our programming tasks. But AI will not be writing 90% of the code". He argues that AI's limitations in following complex instructions and producing error-free integration will preserve human developers' essential role.
A New Way of Working, Again
The historical evidence largely supports the hypothesis that AI represents another step in programming abstraction rather than human replacement. However, the pace and ultimate extent of this transformation may differ from Amodei's specific predictions.
What emerges is a picture of AI as a powerful new abstraction layer that will likely:
Generate substantial portions of routine and boilerplate code
Dramatically increase total code production volume
Shift human focus to higher-level design, integration, and quality assurance
Transform the nature of programming rather than eliminate the programmer
The Assembly-to-C transition provides the most relevant historical analogy. Just as C compilers automated machine code generation while requiring humans to provide design and logic, AI seems poised to automate many coding tasks while still requiring human creativity and oversight for the foreseeable future.
I’m sure Amodei knew exactly what he was doing when he said what he did—but I’m not as sure that everyone parroting what he said all over social media for attention understands the nuance. Human software engineers aren’t going anywhere. They certainly aren’t going to be replaced en masse by AI code generators, but rather will continue the long tradition of evolution into more effective architects of increasingly complex software systems—directing AI to handle implementation details while focusing on the uniquely human aspects of problem-solving and design.