(Prison Captain in “Cool Hand Luke.”)
Is Gen AI Coding a Developer Skill?
Recently, Adrian Cockroft posted on LinkedIn about the skill set issue for gen AI coding:
“The skill set needed to use AI agent swarm tooling well is a mixture of product manager and dev team manager. So developers who don’t have a product management mindset or have never managed a dev team will fail by trying to micromanage the output of the tool rather than specifying the outcome of the product and managing the agent team to deliver that outcome.”
This is very interesting. Back when I was doing technical writing at Digital I noticed that it was easier to teach someone about technology than it was to teach someone how to write.
Either you could write or you couldn’t. And if you could, you could learn to write about software. But if you couldn’t write, you wouldn’t be able to write about software, even if you understood it.
It’s about communication – some people understand that communication exists in the mind of the listener, or processor of the human or programming language, if you will. You have to be able to imagine how your audience is going to interpret what you say or write.
If that audience is a computer, you have to understand how the computer is going to interpret your program or prompt.
In the case of communicating with an LLM, this means understanding how the LLM is going to interpret your prompt. I’m thinking a developer might have a better understanding of this than a product manager. I could be wrong, but I think there’s more to this. I think this issue goes to the heart of the computer programming skill set.
“I used to say that at Netflix it was faster to teach the developers how to operate with cloud, than to teach the operators how to develop. Now I think it’s going to be faster to teach product managers how to develop with AI agents, than to teach developers how to be product managers,” Cockroft adds.
I followed Adrian’s progress closely when he was working for Netflix. He was the public face of one of the early pioneers of moving to the cloud. He is great at communication and explaining things clearly. I’ve attended several of his talks and have spoken with him personally many times.
In fact I invited him to visit us at Citi when I was leading cloud migration for the institutional bank to give us some guidance. We were following the Netflix model.
I clearly remember him explaining that DevOps is a culture change, not just a different way of building and deploying applications. Devs had to take on ops roles, which previously were assigned to a separate department of an organization’s IT shop.
Clearly gen AI coding represents another culture change, but it’s also rooted in the question of what kind of activity programming is. It may be too soon for a clear answer.
Add if we replace junior devs with engineering managers and product managers, or simply eliminate junior devs from the organization, who will learn to become the new senior devs?
WIll it really be easier to teach product managers and dev managers to work with gen AI?
And finally, Adrian is specifically talking about agentic swarm coding here, not plain old “vibe coding.”
It’s not clear yet which approach to gen AI coding will dominate. It could be agent swarms – that approach seems very promising.
But we are also starting to see recruiting ads for vibe code cleanup specialists, which appear to be developer roles associated with plain old “vibe coding.”
Ultimately the debate appears to be more about skills than roles. Which brings up the question, what kind of skills are required to be a successful software developer?
Is Programming a Craft or a Science?
This is a question that’s been around as long as computers have: Is programming a craft skill or a scientific skill?
In other words, does programming rely on the inherent skills of an individual (that they either have or don’t have), or can it be accomplished by anyone who can successfully learn and apply scientific methods and practices?
Is programming like throwing a unique hand crafted pot?
Or is programming more like fitting standard parts together in an assembly line, which produces a mass of the same items produced the same way?
Are you creating unique pottery or are you just helping to assemble cars? (I suppose it can be some of each.)
This is of course an old debate, and a significant technology advance such as gen AI changes the parameters of the discussion.
Ultimately though it’s about the ability of humans to communicate with a computer. Is this innate or can it be learned? And if it’s innate, is it a skill that is present in someone regardless of their role in an organization?
A software program is more like a book than a mechanical object, since it’s something that’s written once and copied multiple times. With a mechanical good you have to create each item separately. This perhaps makes it more like art than science.
The cost of a software program can be measured in the number of hours it takes a person to create it. The cost of a manufactured product is determined by adding together the cost of each part together with the time it takes to assemble them. This perhaps also argues in favor of programming as a craft activity.
What’s Different about Gen AI?
In the case of gen AI, the big change is interacting with computers using human language (rather than programming language).
Programming language was itself a huge step forward as a replacement for machine language. You had a kind of intermediary form that was easier for humans to work with. Less “cognitive load” as we might say today than translating everything into and out of binary in your mind. You just had to understand the semantic meaning of the syntax – i.e. how the computer was going to interpret (or compile) it.
We still have to do that, of course, since human language is not as precise as computer language, and now we have to figure out when an LLM is telling us the truth or blowing smoke.
The evolution of computer programming can be seen as a long, slow march to a point still in the distance at which computers will completely understand humans.
Until then, it will remain up to humans to understand how computers work, and adapt themselves to the computer. While it’s true that computers have come a long way in adapting to humans, there’s still a long way to go.
It remains important for humans to understand how gen AI works, and adapt themselves to it.
An Historical Example
In late 1989, Nippon Telegraph and Telephone (NTT) announced their Multivendor Integration Architecture (MIA) initiative.
This summary from the Japanese computer history museum describes the results of the project.
I was among the original authors of the STDL specification, which was originally developed for MIA to standardize transaction processing programming.
STDL was then endorsed globally by NMF/SPIRIT, and eventually standardized by X/Open (The Open Group), as described in the SPIRIT Platform Blueprint.
At the time we were thinking that the key to software productivity and reducing software licensing and development cost were enterprise computing standards adopted by all vendors.
An article we contributed to the MIT Press volume on ”The Future of Software” drew the correlation between threading standards for nuts, bolts, screws, and other fasteners and the emergence of standard sized parts as the key enabler of mass production. Not, as is commonly believed, the assembly line.
Without standard parts and fasteners, building automobiles was a complete individual (craft) activity that only wealthy people could afford. The Model T changed all of that and it was ultimately because of Ford’s insistence on procuring standard parts.
We thought NTT’s insistence, and later on the telco industry’s insistence, on procuring standard software would have a similar impact on the computer industry. But the industry was still evolving and changing – for example even though MIA achieved portability and interoperability as demonstrated publicly at Telecom ‘95 in Geneva, it was all procedure oriented.
During this time we heard informally from our Japanese colleagues over beers that they were interested in competing in the software industry. Japan in the late ‘80s and early ‘90s was doing well in the automobile, electronics, and photography industries, among others. But they were having trouble cracking the software industry.
They told us that successful computer programming was more likely to emerge from an individualistic culture, such as America, and than from a communal culture, such as Japan, to develop software.
The MIA project was going to close the gap by setting standards that all developers could use, including the big Japanese “software factories” that developed applications for Japanese corporations, which didn’t have IT departments.
Looking at it that way, MIA/SPIRIT was essentially an effort to close the computer programming skills gap by defining scientific and industrial methods.
The Intellyx Take
It’s likely to take some time to settle this debate. And it will take more experience with gen AI coding.
Adrian was specifically referring to “agent swarm” coding, in which a group of AI agents break up a task and work in parallel to achieve the desired outcome, such as creating a new program or improving an existing program.
In the comments on Adrian’s post, he clarifies that he is not saying it will take fewer people, rather people with different skill sets.
This does get to the heart of the question as to whether people have inherent skill sets that either apply to gen AI coding or do not.
Until the industry creates the “perfect abstraction” and computers completely adapt to humans , however, humans are going to continue to have to adapt to computers and understand how to communicate.
Copyright © Intellyx BV. Intellyx is an industry analysis and advisory firm focused on enterprise digital transformation. Covering every angle of enterprise IT from mainframes to artificial intelligence, our broad focus across technologies allows business executives and IT professionals to connect the dots among disruptive trends. None of the organizations mentioned in this article is an Intellyx customer. No AI was used to write this article. Image credits: Google Gemini, Quang Nguyen Vinh and Tom Fisk on Pexels.com.
