In 2023, 92 percent of programmers reported using AI tools at work, according to a survey and internal data analysis from GitHub. Since then, the emergence of prompt-based tools has allowed anyone to “vibe code” a website or software application, and developers are being asked to take advantage of new tech in order to be more productive.
At the same time, computer-science majors and early-career tech workers are saying their job market has dried up. Prominent tech leaders are making stern declarations about a potential decrease in headcount, as Amazon CEO Andy Jassy has suggested, or the threat of falling behind, as the CEOs of Shopify, Fiverr and Duolingo have stated. Salesforce founder and CEO Marc Benioff said he does not plan to hire new engineers in 2025.
This leaves software development teams with fewer resources and less weight in negotiations around salary, benefits and working conditions. Over the last few years, major tech companies have pulled back on benefits spending as they also laid off thousands.
Newly released AI tools for developers are promising to help programming teams identify errors and even generate code. Some of the major players in the industry, such as OpenAI (GPT-5), Google (Jules) and Anthropic (Claude Code) have such offerings, as do startups like Windsurf and Cursor.
This technology is “actually pretty good at generating code, but it generates somewhat simplistic code,” Jay Preall, senior consultant at Segal, told Newsweek. He’s been working with clients for over 30 years in systems integrations and new-tech adoption. “It probably gets you 60 percent of the way there, but you still have to have that expertise to understand, ‘How do I make it better and faster?'”
Windsurf’s leadership was recently “reverse acqui-hired” by Google, as the rest of the company sold to Cognition a few days later.
At the time of acquisition, Windsurf had over 350 enterprise customers, including JP Morgan Chase and Dell, and hundreds of thousands of daily active users. In July, Google paid $2.4 billion for a perpetual, nonexclusive license to Windsurf technology and to hire its founders and a small team of senior engineers for its DeepMind division. The remaining employees and assets became part of Cognition.
In an interview with Newsweek conducted before the acquisition activity, the Windsurf head of product at the time, Anshul Ramachandran, explained how the company’s tool was making developers more productive.
“It was the first truly generally accessible tool for software engineers that had this agentic capability,” he said. “Obviously that’s the buzzword of today, but we were the first to introduce this back in November [2024].”
Windsurf’s product has since been folded into Cognition’s Devin AI offering, and Ramachandran is now at Google. In his conversation with Newsweek, he shared that senior developers were the first beta testers of Windsurf. After they found it useful, the company also saw results from deploying it with nontechnical teams (after building some games, of course).
“All these folks started building their applications,” he said. These were not meant to be public-facing apps but “for internal tools, they were domain experts; they knew exactly what they wanted to build.”
Allowing nontechnical teams to create technical tools could represent savings on software licenses for sales, marketing, legal or finance teams.
“Instead of having to pay six figures for an off-the-shelf, B2B SaaS tool that is really just like a wrapper around a database, they could build exactly what they want,” Ramachandran said.
Getty Images
Windsurf cut around “half a million dollars” of software spending because teams across all functions were developing their own apps in-house.
While noting a decline in the hiring of engineers at this time, Ramachandran said that the startup world remains ripe for opportunity and that the definition of who is a developer may also expand, as anyone with a basic understanding of prompts and the power of large datasets can build something for use within their own teams.
“I don’t think we’re getting rid of developers,” he said. “What we call a developer is going to change, for sure, but still there’s going to be more developers. Software is inherently a value-creation activity. Now that the ability to generate software has become significantly easier, any company in the world that uses technology as a differentiator will be looking at an opportunity cost of not investing more into software that has just gotten higher.”
Preall points out that companies still need technical expertise in the room, because even internal workflows require compliance and other necessary steps that a marketing manager or finance associate may not be aware of. He added that part of the value of a junior engineer is that eventually they become an experienced one.
“You still need that expert to go back and verify, update and correct everything,” he said. “At some point, particularly for the people who are getting rid of all the entry-level jobs, they’ll never develop experts.”
Among the reported challenges of these tools, outputs have been so error-filled that a person would have been better off doing the task on their own, Preall said. He also pointed to an incongruency in responses that again only adds to the editing and quality-control effort of using the AI partner.
“I can take the exact same prompt and ask it twice in a row and get different answers,” he said. “That’s a really big problem. Both answers can be somewhat correct, and I have to have the expertise to figure out which parts are correct from each of the answers that merge them to give me the final answer.”
Some are questioning the value of these tools, given their mixed results. Preall said he does think that eventually the technology will get there, even if it is currently falling somewhat short.
For now, engineering firms have reported limited benefits from AI, and other survey data suggest that engineering teams may not be able to “ship products faster,” given all of the other moving parts associated with a product or feature development cycle.
“Suppose I write a program as a developer, and I use all kinds of variable names. … It doesn’t know my variable names when it goes to write code,” Preall said. “So it’ll write nice code, but if it doesn’t match the standards I already have, I then have to change it all. That’s what I’ve seen is the biggest issue.”
