The software revolution, part II: AI agents are writing the code
I made a bold prediction 12 years ago. I was writing one of my first posts on the Aha! blog and shared my view that 1980 through 2050 would be recognized as the "software revolution" era. I noted that nascent processes and technologies still needed to mature in order to accelerate the timeline between idea and delivery. Yes, it was a bold prediction in 2013. But perhaps not nearly bold enough for 2025.
Incremental growth is in the rearview — today, AI is powering an exponential software expansion that will fundamentally reshape our world.
The revolution is now happening at supersonic speed. Aided by AI, we are creating software faster than ever before. Last year, Google CEO Sundar Pichai said the company's engineering teams rely heavily on AI to improve coding processes and productivity. And Microsoft CEO Satya Nadella recently talked about AI's performance doubling every six months (although he also said it is not generating any tangible value yet).
The Aha! engineering team has been using GitHub Copilot extensively to accelerate writing code. We recently started using more sophisticated agentic coding tools too. (I would share our current favorite, but the market is moving so quickly it will likely change.) These tools do sophisticated planning and multistep development — enough to write complete libraries and tests autonomously.
Right now, Aha! developers leverage AI for at least 2% of the code they write. We anticipate that will increase to a minimum of 10% next year. What mostly slows us down is that habits are hard to change.
My co-founder Dr. Chris Waters and I recently spent some time digging into the implications of this growth. We came to some profound conclusions, which coalesced into a formal hypothesis that we are (mostly seriously) calling the "Waters theory" of AI-driven software development.
The Waters theory is our observation that from here out, the percentage of code written by AI (not people) will quintuple year over year.
If proved true (and only time will tell), the Waters theory means that the software industry is going to change more in the next five years than the preceding 50. Let's pause to absorb the enormity of that for a moment.
At a broader scale, we believe that AI will be responsible for writing more code than all human developers combined by 2030. And by 2032, the majority of legacy code (the backbone of our digital infrastructure) will be authored by AI.
The implications of this new phase in the software revolution are profound and far-reaching, especially for product builders.
With such an overabundance of nearly free and rapidly generated working code, software products will evolve and multiply at an unprecedented rate. So, what can we anticipate? Following the Waters theory, I foresee transformation in four key areas:
New manufacturing: Emergence of "software manufacturing" as a core infrastructure service, likely dominated by hyperscalers such as Amazon, Microsoft, and Google
New companies: Proliferation of software companies as barriers to entry plummet
New careers: Even more individuals engaged in software companies, with a growing need for people who can orchestrate human and AI agent work
New usage: Explosion in software usage, not just by humans but also by AI agents interacting with other software
As the cost of manufacturing software radically decreases, there will be increased demands on businesses and people. Some will thrive, and others will not be able to make the leap forward.
For businesses, the ability to rapidly gain insight, ideate, prototype, deploy, and expertly support will become the ultimate competitive advantage. For people, duties will become more centered on distilling insights and work and market orchestration than development.
All activities related to manufacturing software, from coding to marketing, will be AI-aided. You can start to understand why we think of this as part II of the software revolution.
What about the cost? AI incurs cost on a utility basis. Each line of code written takes some number of tokens, and large language models are billed by the output token. This creates the disconcerting situation of seeing the dollars tally as code generates line by line, giving the impression of high expense. And there are, of course, associated environmental and societal costs with AI that we will have to reckon with over time.
However, if you had a meter on every human product manager, designer, and software developer today, it would be obvious that the current expense is much, much higher in comparison. The cost of what an AI model can generate in a few minutes — what might have taken a developer many hours — is typically a few dollars.
But simply producing more software does not equate to achievement or value creation. It will no longer work for teams to be organized solely around existing methodologies that were designed for the engineering part of software development.
What will separate the winners from the losers? Winners will spend more time learning and confirming exactly what customers need — because the time to create that functionality will be close to zero.
There will be even less patience than there is today for solutions that are off the mark. And the sheer volume of software (both good and bad) will be staggering. Increased market noise from the glut of new software will make it even harder for truly innovative products to stand out. The key will be to ensure the best ideas have the highest opportunity to emerge.
Differentiated strategy, discovery, and research will be extraordinarily critical for success, as well as new go-to-market approaches and new levels of customer support. It might be time to ponder the value of planning approaches that are more akin to waterfall methods. We will very likely become less "agile" even though we are faster to deliver. Call us heretics, if you must.
The challenges ahead can seem daunting. But we see positive signals too. I believe there is a real opportunity for folks who are dedicated to creating exceptional value. Because as the market noise grows to a cacophony, products that deliver tangible value will become beacons in the churning sea of software.
You could say that we are on the brink of a paradigm shift. But that would assume we have not already toppled over the precipice into the future. We are already moving.
AI is revolutionizing the way software is imagined, created, and deployed right now. Trepidation, excitement, anticipation — it is normal to feel it all. But curiosity will take you the farthest, especially if you can remain informed, discerning, and, well, human in an increasingly artificial world.
We must continue to treat others with kindness and try to build what creates the most value. AI is just our newest partner, and one that is always willing to make us better than we have been. It is up to us whether we are open to the possibilities.
The paradigm has shifted and the code is writing itself. Are you ready to shape what will be written?
