What developers using AI chatbots can tell us about how it’s changing work

0
35

In her first job since graduating from college, Eknoor Kaur works at a company where using AI chatbots is not unusual. At first the software engineer at Pathlight, which makes automation tools, was skeptical. But after a colleague mentioned that ChatGPT helped him work better and faster, she eased into the idea—and today, she doesn’t spend a workday without it.

Kaur keeps ChatGPT open on her desktop, typically posing the bot four or five questions a day. She doesn’t use the tool to write code because she’s worried about the hallucinatory, or made-up, answers that AI chatbots can provide. Instead, Kaur uses the system like a search engine, asking it programming-related questions she doesn’t want to burden coworkers with.

Not surprisingly, some of the earliest adopters of generative AI at work are software developers. Alongside ChatGPT maker OpenAI, companies like Microsoft and Salesforce have rolled out AI copilots, or digital assistants, for writing code. And while a slew of employers, including Apple, Bank of America, and Goldman Sachs, have blocked or limited ChatGPT on the job, things are different at plenty of tech companies—and startups in particular. Tech workers use a range of AI chatbots. Amazon developers, for instance, have their own version of ChatGPT called CodeWhisperer.

ChatGPT is currently in the “giant room-size machine phase,” said David Baggett, founder of cybersecurity firm Inky. He likens our current chatbots to the computers of the 1950s: early-stage and used for a narrow range of tasks. 

AVIDAC, one of the early computers, began operating in 1953.
Photo: Argonne National Laboratory via Wikimedia Commons

But engineers are leading the charge in using those chatbots at work, even in their limited capacity. Developers Quartz spoke with use ChatGPT to generate code for software—saving themselves anywhere from minutes to hours a day on writing—or to find information faster than using traditional online search methods.

Instead of starting on Google or Stack Overflow, a popular Q&A site for developers, which could take several pages or clicks to land on the right piece of code, developers can ask ChatGPT or another chatbot and get what they need with one prompt. “I do a lot less Googling,” said Amin Ahmad, CEO of search software company Vectara and a former researcher at Google.

Developers can also prompt chatbots to write code for them—and adjust from there. “Everything hasn’t worked on the first try,” said Cody De Arkland, head of technical marketing at tech management platform LaunchDarkly. De Arkland said he uses ChatGPT as one of his final steps to see if there’s a better way to optimize his code, like writing it more efficiently. He uses a few AI chatbots, including GitHub Copilot, which is paid for by his employer.

Generative AI doesn’t always work for Baggett, either. In his experience, ChatGPT sometimes spits out an answer that doesn’t work at all.

The speed of coding with chatbots

At LaunchDarkly, De Arkland recalls a teammate who estimated that coding a complex pricing calculator would take roughly two months—but after using ChatGPT, wrote the code in just a week and a half. The obvious case for coding with chatbots is speed: projects are finished faster, and engineers say they’re shifting freed-up time into building better features.

“We’re not going to end in a place where there’s not enough work to go around,” De Arkland said. “There’s always going to be projects and new things that have to be built to fill up that space.”

But there’s a limit to what software engineers will share with AI chatbots. For example, none of the developers Quartz spoke with said they would paste entire blocks of code into ChatGPT or other chatbots, wary that the AI tool could compromise data privacy, or have trouble understanding large volumes of text. For some, it wasn’t clear if their employer had guardrails to prevent people from entering personal data into a chatbot.

Could AI assistance make us like our work more?

In general, developers say that ChatGPT takes away boring baseline work. Catherine Yeo, an engineer at coding software maker Warp, has used her company’s AI chatbot for nine months. Even today, she “always marvels when it does return an answer” and solves her problems.

Vectara’s Ahmad notes that a chatbot allows him to find new solutions to a problem he wouldn’t have initially considered when writing code. But as a developer working on AI technology, he—like many non-technical workers—worries his job could be automated away.

LEAVE A REPLY

Please enter your comment!
Please enter your name here