The Impact of AI on Software Development
Architects not Developers
The first commercially successful bicycle was the Draisine, or "Running Machine," developed by Karl Drais in 1817. These early machines had no pedals and were propelled by riders pushing off the ground with their feet. By the 1860s, the pedal-driven "boneshaker" (or velocipede) gained popularity, with many small manufacturers handcrafting these early bicycles. The 1870s saw the rise of the penny-farthing (or high-wheeler), distinguished by its large front wheel. Prominent manufacturers of the time included Ariel, Coventry Machinists Company, and Singer.
Everything changed with the advent of the Rover safety bicycle, initially conceptualized in the 1870s but fully developed in the 1880s by the British Rover Company. Unlike its predecessors, the Rover featured equal-sized 26-inch wheels, making it far easier to mount and allowing riders to place their feet on the ground when stopping. Major American manufacturers such as Pope, Overman, Western Wheel Works, and Gormully & Jeffery quickly followed suit, producing their own versions of safety bicycles. However, what remained the same was that the value of the product was in the manufacturing and assembly of it. The designs were essentially copied by a multitude of companies.
The American bicycle industry truly took off in the 1890s. In 1890, production stood at 30,000 bicycles. By 1892, the number had doubled to 60,000, but the real boom followed the recovery from the 1893 economic depression. By 1896, over 400,000 bicycles were produced annually, skyrocketing to 900,000 per year in 1898 and 1899. However, as quickly as the industry surged, demand collapsed. By 1904, total U.S. bike sales had dropped 75% to just 250,000 units.
The boom-and-bust cycle of the bicycle industry continued throughout the 20th century. Large retailers exerted pressure to keep prices low, and competition from Asian manufacturers, with significantly lower labor costs, reshaped the market. By the 1980s, even high-end American brand Schwinn had shifted production to Taiwan and China. While the value was still in the manufacturing and assemblage, conducting those at the cheapest possible rates allowed companies to continue to grow their profits.
Recent decades have seen the rise of two dominant American high-end bicycle companies: Trek and Specialized. These brands, now among the world’s largest, focus on cutting-edge design and engineering. Modern bicycles are crafted from lightweight carbon fiber, titanium, and advanced alloys, maximizing strength, durability, agility, and performance. Aerodynamic frames, precision engineering, and sophisticated suspension systems cater to cycling enthusiasts. High-performance components, such as gearboxes, hydraulic disc brakes, electronic shifting systems, and smart connectivity features, are increasingly common. However, manufacturing remains largely concentrated in regions with lower labor costs.
Bicycles have evolved from handcrafted machines where labor was the primary value in the 1800’s to mass produced items of the 1900’s where cost savings was key to today where innovation and design drive success of these highly engineered products. What began as a simple mechanical device has transformed into a sophisticated piece of technology. I tell this story of the history of bicycles from handmade to mass produced to highly engineered as an analogy for what I believe the software industry is going through with AI.
Software, like bicycles, from the start was very handmade. An individual developer (or two if you are pair programming) types away at a keyboard developing software that ultimately serves some purpose and hopefully solves some customer problem. I would argue the push for mass produced software has been around for decades in the form of off-shore development. The key, just like with manufacturing, is that these teams have to produce the same quality software with lower costs, and often they do.
Today with the exponential growth of AI-assisted or AI-driven software development, this push for much cheaper software development has grown significantly. The CEO and founder of Salesforce, Marc Benioff, has stated that Salesforce will not be hiring any more software engineers in 2025. Sam Altman, CEO of OpenAI, recently said that he believed AI will soon be capable of handling many tasks that a junior-level software engineer does today. Altman, on his blog, says about AI agents:
It will not have the biggest new ideas, it will require lots of human supervision and direction, and it will be great at some things but surprisingly bad at others.
Still, imagine it as a real-but-relatively-junior virtual coworker. Now imagine 1,000 of them. Or 1 million of them. Now imagine such agents in every field of knowledge work.
In some ways, AI may turn out to be like the transistor economically—a big scientific discovery that scales well and that seeps into almost every corner of the economy. We don’t think much about transistors, or transistor companies, and the gains are very widely distributed. But we do expect our computers, TVs, cars, toys, and more to perform miracles.
I very much agree with Altman’s vision of the future. I wrote about a year ago in the Impact of AI on Organizations, “One possible scenario is that individual engineers will have a group of AI-bots working for them on small tasks. The engineer will design the feature, while bots develop small parts of it.” I still very much believe this scenario. Teams will likely include a PM, a designer, and an engineer. The engineer will be responsible for guiding hundreds or thousands of agents to build the software. Engineers will still be needed but much for higher level tasks, think more software architects and not coders, and a lot fewer of them per team.
Don’t get me wrong but there will be a lot of fully autonomous AI that can build simple web sites or even mobile apps, like Replit does today, without any engineering support or supervision. Back to our bicycle analogy. If you are buying your six year old child a bike to learn how to ride on, you’re likely going to buy a mass manufactured, cheap one. But if you are going through your mid-life crisis and decide to become a triathlete or cyclists, you’re probably going to go for one like Trek’s Madone SLR 9 AXS Gen 8 that claims it was “built for dominating the roads with equal parts light weight and aerodynamic speed…crafted from our highest-level 900 Series OCLV Carbon…Lightweight IsoFlow frame technology…with a top-level SRAM RED AXS drivetrain with a power meter, carbon wheels, a Trek Aero RSL one-piece carbon bar/stem, and RSL Aero Bottles and Cages for more drag reduction.” Admittedly, I don’t understand most of the words in that last sentence but it sounds highly engineered to me.
Having recently done a little software side-project myself over the course of a couple of weekends, I can attest that while I don’t think AI does a great job designing and architecting a system, it does an excellent job reading and understanding code, and does a surprisingly good job of coding functions and modules. The slow part of the development cycle was between the seat and the keyboard, aka me.
My advice, if you are a software engineer or studying to become one, is to certainly be proficient in coding but you need to possess skills way beyond that. You need to understand how to both architect software properly (scalably and efficiently) as well as how to manage a team of AI-agents. I love hand-made items but most of us purchase a lot of mass produced items and a few items that are highly engineered. For those highly engineered items, we pay a lot because the value comes from the expertise required to design it…not produce it. Software engineers of tomorrow, likely renamed to something like software agent leaders, will need to be even more highly educated and skilled.
Another industry to draw parallels from is farming. If you still think of farmers as luddites you are very much mistaken. They are often incredibly adept at understanding and using technology. Are there fewer farmers than in the early 1900’s, yes of course. But the roughly 2M that remain know how to manage technology, not just crops and livestock.
The evolution of bicycles from handcrafted machines to mass-produced commodities and finally to highly engineered marvels mirrors the transformation happening in software development through AI. Just as modern bicycles derive their value from sophisticated design and engineering rather than manual assembly, the future of software development will prioritize architectural expertise and the orchestration of AI agents over traditional coding. Engineers will become software architects and managers of AI-driven development, much like how premium bicycle brands thrive on innovation rather than manufacturing costs. As the software industry rides this wave of change, those who adapt by embracing advanced design, scalable architecture, and AI management will find themselves in high demand, just as cyclists seek the finest engineered bicycles for peak performance.