Recognizing the need for careful human oversight, Fractional Chief Technology Officer and Consultant Jevin Maltais turned to AI to speed up coding tasks and tackle unfamiliar programming languages.jessica deeks
Jevin Maltais remembers the first time his eyes were opened to the capabilities of generative AI.
It was November 2022, and Maltais was working as an engineering director at Humi, a payroll software company. OpenAI just released ChatGPT version 3.5, and a colleague interviewing programmers for a job told Maltais how candidates were using it to complete coding tests that typically took an hour.
“He said people were just feeding them.” [the test] “I logged into ChatGPT and within five minutes I was back. I had no idea how this worked,” he says.
Maltais began experimenting. I started by asking ChatGPT for feedback on my code. He found that the early results were not reliable for deeper study. But when version 4.0 was released in March 2023, he realized how meaningful AI could be, especially when making changes to existing code.
“I’m not going to manually change the text” [anymore]“I ask the AI to do that because it can actually work more thoroughly and see other areas that might be affected that I wasn’t looking at.”
This technology is compressing timelines that once spanned months. Traditionally, implementing a new feature for an app has required engineers and sales and business development teams to get together in a room and discuss what needs to be done, then the engineers have to spend a day or two planning how to implement the feature and how long it will take (often months), then come back to the team to present their findings and decide whether to develop the feature.
Now, Maltais is using AI to build working prototypes in 15 minutes.
“The versions may not be very good, but instead of taking months to perfect the first version, it’s something you can show and play with and iterate on continuously,” he says.
Thanks to AI, Maltais can now work in programming languages that are unfamiliar to him. In his current role as a fractional CTO and consultant, he recently worked on a product using Kotlin, a language for which he lacked deep experience.
“I didn’t have much language or technical experience,” Maltais says. However, he proposed new feature ideas using Claude code. “What used to take weeks to learn can now be accomplished in a day or two.”
That acceleration is just one dimension of change. AI will also fundamentally change how developers collaborate, and whether they even need to collaborate at all.
Jevin Maltais says AI can build an initial prototype of an app or new feature in just 15 minutes. “Maybe the version isn’t very good, but it’s something we can show and play with,” he says.jessica deeks
Stephen Doxsee, software developer, chief technology officer (CTO), and co-founder of SuccessionHR, a software company that creates succession planning tools for businesses, realized that AI could act as a partner he could bounce ideas off of.
Doxsey said he may have worked with another programmer in the past, with one person “navigating” and sharing insights and ideas while the other “managed” the keyboard. Well, AI is that virtual programming partner.
“AI helps foster curiosity by spitting out code and digging deep to understand problems,” he says. “It’s like a super senior developer who knows everything inside and out.”
However, this technology has significant limitations.
“It’s not perfect,” Doxey said. “AI can be overconfident, and it’s not always accurate. We need to make sure we know what the AI does and put guardrails around it so it doesn’t mislead us, build the wrong thing, or introduce bugs into our code.”
These concerns were echoed by Jyoti Kunal Shah, independent researcher and director of application development at ADP. Jyoti Kunal Shah explores refactoring, the process of restructuring code to be cleaner or more efficient without changing its behavior.
“Providing explanations for refactoring decisions can significantly increase developer trust and transparency,” Shah said in a paper published in July 2024. American Journal of Engineering and Technology. “Such explanations tie the behavior of the AI to well-known software engineering principles and make it easier for developers to evaluate and approve changes.”
She describes a collaborative workflow in which an AI suggests improvements, explains why, and waits for a human developer to approve or reject the suggestions. When that happens, Shah writes, “developers’ expertise combined with the speed and pattern recognition of AI can produce optimal results when decisions are transparent.”
Despite AI’s limitations, both Maltais and Doxsee see it as an irreversible change in their industry, one that enhances rather than replaces human programmers.
“While AI will certainly continue to evolve, I believe humans are essential to the art of coding,” Doxsey says. “We want to take full advantage of the insights that AI provides, but we want to keep humans in the driver’s seat.”
Maltais admits he is worried about AI taking his job. But he says developers who learn how to leverage AI will continue to be in high demand.
“The most powerful thing we can do as engineers is use it regularly and find creative ways to be more productive and provide more value,” he says. “It’s hard to predict how things will turn out, but finding new ways to use AI is a major way we can become more powerful as individuals.”
