Lowering the Floor and Raising the Ceiling: Building With AI
The World Needs More Software Developers, Not a Shrinking Guild of Craftspeople
The Evolution
What it takes to be a builder has evolved. What it means to be a builder hasn’t. Is the craft defined by the journey or the destination?
Generative AI is changing the software engineering profession. It is creating more space for everyone to build. At the floor, it is lowering the entry requirements for people to jump in and build. At the ceiling, it is raising the headroom for senior developers to press further into the frontier of opportunity. It is enabling investment capital to drive the flywheel behind our industry, spinning up software startups and driving the economy forward. Some suggest that generative AI saps the panache of craft for the builder. With every massive disruption since the dawn of high tech, we hear that the current way is the One True Way and that the new thing will ultimately dilute the value of what it means to be a builder.
There is plenty of room under this tent for more people to find joy in building software. As gen-AI-driven productivity increases for each member of the product development pipeline, the business gets more leverage from talent. The return on invested capital – which includes investment in people – will track with productivity gains. More leverage will drive more investment. More investment encourages more people to acquire the skills and experience needed to become builders. And generative AI puts that within reach. The flywheel will continue to spin.
We will need more builders: not a select cadre of the most educated, but a more diverse group brought together with a common love of building with code.
The Great Hand-Wringing
There are many builders who have concerns that deserve airtime. They earnestly believe the craft of the builder is imperiled by generative AI— lower code quality, potential job loss, and a diminishment of the developer’s responsibility to know how things work under the covers. These folks are in good company; many of our software engineering forebears have resisted adopting disruptive technologies – for surprisingly similar reasons. This has been the pattern since the earliest days of our industry back to the 1950s. It is appropriate that developers are skeptical by nature. It’s one of our best attributes. We only want the best solutions to survive and find the thrill of culling ideas part of the fun of our jobs. Besides, change isn’t always for the best. Software engineers, known for their commitment to facts, have often laid out valid arguments for their resistance.
We’ve seen enough hype cycles to furrow our brows whenever an upstart panacea emerges. As owners of our careers, we should understand where things are heading, and speak up if things veer off course. But when our sense of professional identity is defined by the way we build today rather than the timeless and unbounded essence of building, we can get caught in a trap that has ensnared many.
There has always been tension between protecting our craft, boosting productivity, pushing competitive advantage, and driving greater returns on investment. The key question is whether we are protecting the right thing. Each successful disruption in the industry created tensions that were ultimately relieved when significant boosts to developer productivity became apparent. Does generative AI impair our craft and reduce the joy of building or does it do precisely the opposite?
Coming to Grips With the New Reality
The sonic boom of generative AI exploding on the scene has sparked awe, excitement, frustration, and even anxiety—often all at once. It may take time to decide if or how you’ll let AI shape your journey as a builder. As the world shifts around you, finding a place of confident stability can be challenging. Each of us must integrate new information with our own vision of a fulfilling and productive career. Beyond our personal perspectives, we should remember that different skills, goals, and philosophies lead people on unique paths. What works for one builder may not work for another.
Goals
My goal is not to change your mind on whether generative AI is good or not good for you. My goals are to first discuss the enlargement of the pie. A broad spectrum of new and valuable talent gains access to our craft thanks to generative AI. And we need all the help we can get. Second, a look at historically similar challenges our industry has faced, each prodding builders to evolve their craft toward a new – and often untrusted – method.
One: Enlarging the Pie
Lowering the Floor
Generative AI is helping more people participate in, and discover a love of, building software. Given curiosity and persistence to push through the quirks of working with quickly evolving gen AI tools, anyone can become a builder.
Creating a Developer from Thin Air
Let’s start with a story about a guy named Vic. While a deep thinker who is well-read, most of his passions exist in the physical realm. I met him on the mats of a mixed martial arts gym in Seattle in 2017, when he was running a construction business with his brother. He’s an absolute savage on the mats who has travelled the country and trained with the best. At the surface, he could be mistaken as a single dimensional knuckle-dragger. As one of my best friends, I can be forgiven for that superficial description of him.
Last year, Vic and a business partner decided to start a tech company. Its mission is deeply connected to his love of fighting sports. On a shoestring budget, they cobbled together a barebones MVP using a small team of contractors. Lacking senior technical leadership, their first version was… rough. The codebase was fragile, required significant human intervention to configure, and wasn't built to last. I could see Vic’s frustration. He lacked agency. The system design, infrastructure, data and code was a black box to him. Like every startup at this stage, they were burning through a limited pile of cash and had to acquire and retain customers to survive.
Over the next several months, Vic took control by digging in with the help of ChatGPT to both learn what the system was and discover what needed to be done. He now floats through the AWS console with ease, from S3 to EventBridge to CloudWatch, describing what's under the hood. He popped open a lambda and showed me some python he had written with ChatGPT to replace the shell scripts written by the first gen [human] developers. He had been writing code with gen AI for 10 hours the day before and had finally gotten a critical problem fixed. He was beaming. He had taken possession of the system and was thrilled to understand the path forward.
I saw before me a man transformed. With a couple more years in this grind, I expect Vic will both catapult his skillset and the enterprise value of his company.
Supercharging a Product Manager
Tim is an experienced product manager with an ability to see where things are going and imagine solutions that resonate. This makes him a firehose of ideas, filling up roadmaps – only to wait for developers to come up for air and build his vision. Tim discovered Repl.it’s generative AI capabilities allowed him to prototype and quickly iterate. He is now selling his ideas with code and not just slide decks. By removing the lag time to design and code, he can refine ideas more quickly and render them without things getting lost in translation.
When only developers get tools that accelerate their work, it creates more problems than it solves. The rest of the organization doesn't keep pace. More features succeed by speeding up product managers at the beginning of the feature production pipeline, enabling ideas to be shaped, tested, and evaluated earlier, before development gets started. By helping product managers directly convert their ideas into on-screen proof, they are empowered and get more joy by taking a more direct part in building.
With generative AI, Tim took a leap to become more than a dreamer. In addition to that, he’s now a builder.
A 2024 McKinsey study showed gen AI can boost product-manager productivity by 40 percent in prototyping and early validation tasks,[1] but it doesn’t scratch the surface of their potential when they can prototype with code.
Raising the Ceiling
While generative AI democratizes entry into software development, it simultaneously expands what's achievable at the top end of the skill spectrum. For experienced developers, the benefits go beyond coding faster. The more profound impact is how AI removes cognitive overhead and eliminates implementation friction, freeing senior developers to operate at a higher level of abstraction and tackle problems that were previously beyond reach. The ceiling rises because the fundamental constraints change.
This benefits in numerous ways. Experienced developers can prototype across a broader range of technologies with a shorter learning curve. They can explore architectural approaches that would have been too time-intensive to validate through manual trial and error. Like shooting hoops from the foul line, repetitions combined with error correction help people learn. This feedback loop doesn’t go away with generative AI, it happens faster. By being able to quickly iterate on tough frontier issues or new technologies, senior developers can solve more and improve the value of their technical contributions (make mistakes, fix, learn) more efficiently. By definition of their role, they must push boundaries. By pursuing innovative ideas that exist at the intersection of multiple domains or taking risks at the frontier of organizational knowledge, they can create opportunities for the business that would have previously been impractical.
When senior developers find flow – however they incorporate gen AI into their work streams – they can unlock so much potential in their skills development and for their teams. Let’s look at examples of what raising the ceiling looks like.
A Busy Manager Finds Time to Build
Cindy is a Senior Manager who has the hybrid instincts of a software engineer and product manager. When a product challenge surfaces, she thinks “I can build that.” But her time is severely constrained because the job of a manager is orchestration, driving decision making, and supporting her team. She has closely tracked AI advances and understands the gaps between its capabilities and what builders really need. Her mantra for her team is to “evolve with the tools” and she practices what she preaches. She’s foundation-model agnostic, switching fluidly between Claude, ChatGPT, Cursor, and Gemini to deepen her technical knowledge and write code. Even with limited time, she can take an idea from concept to MVP in hours, not weeks, without getting in the way of her other obligations.
Without generative AI, launching a new project would be too time-consuming for someone balancing leadership and life. In a recent conversation, Cindy told me “I don’t have to be the strongest engineer in a given technology to be able to build." That sense of agency—the confidence to take risks without needing to be an expert—is what every developer, no matter how senior, deserves.
A Senior Manager Gets His Coding Mojo Back
David and I have worked together at a couple of startups. As he worked up the career ladder into management, the pull of code was never far from his mind. Last year, he rediscovered his love of building with code thanks to the gen AI features of Repl.it and Cursor. He shared with me recently that he achieved a better state of flow when building with generative AI assistance. Because David can prototype and iterate quickly, he’s now focused on projects that would have required a full dev team in the past—proof that his professional ceiling has moved upward. Thanks to the productivity he has enjoyed, he now has a small portfolio of apps generating passive income for him.
David doesn’t sweat the details of generative AI code quality. He lets the AI do almost all the coding. He focuses his technical design attention on the data. His rationale for this seemingly incongruent imbalance of technical attention is fascinating. He suggests that if quality becomes an issue he can simply “have the AI regenerate the whole project down the road.”
David is all-in on AI. He presents this provocative challenge to his peers in software leadership:
“Would you trade a team of free developers for a year in exchange for giving up your ability to use generative AI?” You can guess which way he goes on that trade.
Research and Development
Most small to midsize software companies have no distinction between technology research and development. Large companies have the luxury of dedicated R&D teams looking at emerging technologies. Their purpose is to create new opportunities with a much longer time horizon that most companies can’t afford. Meanwhile, developers at companies with limited resources maintain outdated code stacks because the cost of rebuild and replacement is just too high. Or is it?
Think about how consumer search features have evolved. The emergence of cost effective embedding models and managed vector search should transform the way that software products implement search. The user experience, quality, cost, and speed is unparalleled compared to the incumbent technology. A single senior developer who plows into a greenfield project like this can help cue up a well-formed architecture and prototype to dramatically reduce the risk and cost of evolving foundational features. One senior developer, leveraging gen AI to help with research, documentation, and code, can become an ultra-productive technology researcher. Several can evaluate options in parallel and help their companies take advantage of disruptive changes, improving competitive differentiation.
The Economy, the Investor Class, and the Builder
Another ceiling that is lifting is economic growth potential at the macro-level. GDP grows when more companies develop products faster and more efficiently.[2] There are significant factors at play with generative AI helping strengthen the economy at large and the ability for investors to deploy more capital to keep the flywheel moving.
Funding levels are on the rise again. After the dip in 2022–2023, venture capital is rebounding.[3] In Q1 2025, the IT sector (which includes software) accounted for 74 percent of all VC investment,[4] with AI-related deals dominating.[5] As capital flows back into the market, more startups will get funded—especially those leveraging generative AI—fueling the cycle of innovation, talent demand, and economic growth.
Startups, fueled by investor capital, will both succeed and fail faster, with lower overhead. This turnover at industrial scale is necessary to find the best ideas and companies with the capabilities to scale them. Companies typically pivot several times before succeeding or failing. Accelerating this cycle, and the supporting investment cycle, helps move the economy forward more robustly. Generative AI helps by allowing a company to build more with less, getting early test products out the door with lean staffing. Less capital deployed per early stage startup allows investors to make more bets. Faster turnover helps the entire ecosystem learn faster.
AI-savvy developers will be needed to help power this economy. Builders with the greatest resiliency to adapt and excel using generative AI will be among the many beneficiaries of the rising market ceiling. Given the tighter market conditions in the post-pandemic trough, developers must differentiate to ensure stable employment.
We know the direction but we don’t know the destination. More changes are coming that will shift the software labor market. As AI coding agents mature, they will become a talent competitor to human builders. Building relevant skills and being able to tell the story is an investment that can defend builders’ ability to continue to participate in the market.
Data compiled from Crunchbase, EY, Bain & Company, Statista, US Bureau of Labor Statistics.[6][7][8][9][10][11]
Two: Historical Upheavals
We are living in a loop which evolves as it repeats itself. Each iteration learns from the previous, moving us up further from the metal, pushing our productivity further, yielding a bounty of new inventions, and attracting more opportunity and talent to our field. Layer upon layer, we have managed to keep our sense of connection with past generations of developers whose workdays and approaches to building didn’t remotely resemble our own. That we even call ourselves software engineers today would be laughable to many of our predecessors.
“Real programmers would not stoop to” Assembler
Before the age of the assembler, machine code was written in octal codes in pencil on sheets of paper. Humans were the assemblers. Crucial to the performance of their craft was knowing opcodes and memory addresses. They were just a hair’s width removed from the metal, having evolved beyond the “patch cord” developers of the previous era. Their forebears might have muttered that “you would never really know what the machine was doing if you couldn’t follow the cable.” But the attraction of higher productivity pushed the adoption of machine code.
When machine code gave way to assembly language, allowing developers to use mnemonic commands and references to memory instead of directly controlling the machine, they again moved one more step away from the metal. And the grumbling started.
“People argued that it was wrong to program in symbolic code. I was trying to explain to programmers why they should use symbolic code. “No“, they said, “We’re used to octal… It’s better. You’re closer to the machine.[12]” Ed Fredkin, MIT computer scientist
Assemblers did the work formerly central to the value created by the builder. The translation process done by assemblers allowed developers to get work done much faster,[13] with fewer defects. But there were holdouts.
“A real programmer would not stoop to wasting machine capacity to do the assembly.”[14] - Richard Hamming, Turing Award winner and Bell Labs mathematician
High-Level Languages and Structured Programming introduced “that damn C compiler”
Between the late 1950s and the early 1970s, high-level languages like FORTRAN, Pascal, and C became the standard approach for software development. They further abstracted the work of the builder, removing them almost completely from connection to the machine. Their craft shifted from the low-level focus to building code that could be easily moved across systems and shared. But for many builders of that era, it was the connection to the machine that made them builders.
“At the Users Group, someone barked: ‘I’ll never use that damn C compiler—it’s a black box! I don’t want some translator hiding my loops in registers. Give me my own mov, add, and jump instructions!’”[15] Dennis Ritchie, Turing Award winner, Co-Inventor of C and Unix
In March 1968, Edsger Dijkstra’s landmark letter ‘Go To Statement Considered Harmful’ began a movement to reshape standards of readability, maintainability, and portability— valuing human productivity more than raw CPU performance.[16] GOTO statements, which veteran programmers used to milk every bit of oomph from every machine cycle, were gone. Early compilers ran 10–20 percent slower than hand-tuned assembly, giving die-hard programmers reason to believe the new way was a disaster.[17] Given the slow performance of computers of their time, this created a combined sense of loss of intimate control and performance. But the need for productivity won.
The GUI, Event-Driven Programming, and IDEs “make developers lazy”
The late 1980s and 1990s brought about the graphical user interface era. The event loop and high-level constructs supported by Microsoft’s strongly evangelized developer APIs required developers to completely rethink programming to continue to be relevant. When GUI building development environments came out, developers gained an automated coding assistant that hid the details inside an opaque box. By clicking a few buttons, all the windowing code was generated under the covers. Some of the earliest attempts at this, like Visual Basic, were fiercely objected to by the old guard.
“Visual Basic is for script kiddies. Any serious Windows program deserves handwritten Win32 code.” — Andy Reid, Microsoft MVP and Win32 expert
I love the irony of this quote. Win32 itself was a massive abstraction layer on top of the underlying operating system. It’s hard to argue with success. The combined power of these innovations ushered in The Age of the Desktop App, abstracting developers completely away from the machine, changing the game forever.
The Dot.Coms, Cloud Managed Services, and “the defenseless”
Another significant abstraction pulling software developers further from the machine was the emergence of two back-to-back disruptors: web application development and managed services in cloud computing. Managed services wrap services, which in turn wrap services, never providing access to the real machine. The challenges mitigated by managed cloud services were previously the domains of deep builder expertise. With these differences, some skills were no longer relevant while new ones became eagerly sought after. For the naysayers, the fear of loss of control over the direct access to the system was summed up well in this quote:
“One reason you should not use web applications to do your computing is that you lose control… Do your own computing on your own computer...If you use somebody else’s web server, you’re defenseless.”[18] - Richard Stallman, Free Software Foundation, 2008
In this environment, builders typically write a thin veneer of code on top of a layer cake of other services that they had no part in developing. Did the craft lose something or did it gain something? The ease with which teams could build more efficient and scalable software created massive value for the industry. Builders shifted their craft with the times.
Gen AI makes developers “lemmings”
We finally arrive at today and the current upheaval in software development. I’ll include one quote here, which could have just as likely been attributed to any of the previously covered generations of developers. This critic warns of the “lemmings” developers who follow AI-generated code “off a cliff.”
“You want real connection to code? You earn that. You dig in. You wrestle with segfaults at 3 in the morning. You pace your apartment muttering about pointer arithmetic.”[19] - Jj, UK-based developer, 2025
Russ, a senior developer at my current company, noted that the key to success is the builder retaining agency and ownership through the collaborative process with gen AI. One-shot prompting doesn’t bear great results. Building great code with AI requires a high-context, interactive exchange of information. Like meandering down an intriguing rabbit hole in Wikipedia, the flow of questions and answers can be truly generative, and can result in solid code, if the humans pay attention.
Building Forward
The great hand-wringing that accompanies every major shift in our industry is familiar: each advance that seemed to threaten our craft ultimately created more builders, more opportunity, and greater economic value. Generative AI isn't replacing the joy of building—it's redistributing it more broadly and encouraging skills growth. The construction business owner who can now debug AWS Lambda functions, the product manager who can prototype without waiting for dev availability, the senior managers who rediscovered their love of coding—these aren't threats to our craft.
Yes, the nature of our craft is changing. The bare metal knowledge of memory addresses gave way to understanding algorithms, which gave way to architecting systems, which now gives way to orchestrating AI-assisted development. Each transition required adaptation, and each time, the builders who embraced change found their impact multiplied rather than diminished. Many challenges lie ahead. We’ll be blaming AI agents for system outages when we prematurely hand the reins over without adequate oversight. But like all mistakes, we’ll tweak, iterate, and learn. The technology is ours to harness.
We builders have a choice: join the frustrated chorus suggesting that gen AI spells doom for software development. Or we can recognize what's actually happening: the tent is getting bigger, the ceiling is getting higher, and there's never been a more exciting time to be a builder. With generative AI, we aren’t replacing craftspeople. We’re enabling millions more to discover the joy of bringing ideas to life through code.
References
[1] McKinsey & Company. How Generative AI Could Accelerate Software-Product Time-to-Market. February 6 2024.
[2] The Economic Potential of Generative AI: The Next Productivity Frontier. June 14 2023.
[3] Q4 2023 Venture-Capital Outlook: “Market to Seek New Floor in 2024.”. December 2023.
[4] National Venture Capital Association and PitchBook. PitchBook-NVCA Venture Monitor, Q1 2025. April 2025.
[5] CB Insights. State of AI Q1’25 Report. April 2025.
[6] Bain & Company. Global Private Equity and Venture Capital Report 2024. January 2024.
[7] Statista. Value of Venture-Capital Investment in the United States from 2006 to 2023 (in Billion US Dollars). Accessed May 29 2025.
[8] Bureau of Labor Statistics. Employment Projections: Software-Developer Workforce, 2015-2025. 2024 release.
[9] Occupational Employment and Wage Statistics: 15-1252 Software Developers. May 2023.
[10] Crunchbase News. Q1 Global Startup Funding Posts Strongest Quarter Since Q2 2022—With a Third Going to Massive OpenAI Deal. May 2025.
[11] National Venture Capital Association and PitchBook
[12] Fredkin, Ed. Oral History Interview. Interview by Gardner Hendrie. Computer History Museum, July 5 2006.
[13] Hamming, Richard W. The Art of Doing Science and Engineering: Learning to Learn. New York: Gordon & Breach, 1997.
[14] Hamming
[15] Ritchie, Dennis M. “The Development of the C Language.” In History of Programming Languages II, edited by Thomas J. Bergin Jr. and Richard G. Gibson Jr., 671-98. New York: ACM Press/Addison-Wesley, 1996.
[16] Dijkstra, Edsger W. “Go To Statement Considered Harmful.” Communications of the ACM 11, no. 3 (March 1968): 147-48. (Digital copy available via ACM Digital Library).
[17] Hennessy, John L., and David A. Patterson. Computer Architecture: A Quantitative Approach. 6th ed. San Francisco: Morgan Kaufmann, 2019.
[18] Stallman, Richard. “Cloud Computing Is a Trap, Warns GNU Founder Richard Stallman.” Interview by Bobbie Johnson. The Guardian, September 29 2008.
[19] Jj. The Copilot Delusion. Blog post, May 2025.