Tech companies lead the way on generative AI: Does code deserve the credit?

Are gen AI coding tools the thin edge of the wedge that can lead to wider use of gen AI across all industries?

Faruk Muratovic

United States

Technology companies (hardware, software, semis, and more) surveyed worldwide believe that generative AI tools will be transformative to their companies more than non tech companies. In part, this may be because tech companies are using gen AI tools to write software code at a much higher rate than non-tech industry respondents. This could suggest that software coding tools are an important early use cases for gen AI.

Gen AI tools for writing and testing software appear to be among the most compelling and user-ready use cases for enterprise adoption in tech companies.1 Tech companies are code heavy and have greater need for these tools compared to non-tech companies: Almost half (46%) of all developers work for tech companies.2 Technology companies are the fastest adopters of AI in the United States as of 2024: Almost 20% of those surveyed are already using AI, with well over 20% expecting to use it in the next six months—more than double the rate of all industries but one.3

As depicted in the figure, tech companies are significantly more optimistic (by 15 percentage points) than non-tech companies that gen AI will transform their organization in the next 12 months or is already doing so. A big reason for this optimism is likely the applicability of gen AI for software code. 

Developers are currently using these tools to (among other examples) write boilerplate or routine code, maintain code accuracy, and streamline the process of finding coding shortcuts.5 Gen AI coding tools are often seen as more like autocorrect: A tool that is used dozens of times per day and enhances productivity by roughly 10 to 20 minutes per day.6 That said, this is rapidly evolving, and co-piloting tools are often used not only for code autocomplete and suggestions, but for interpretation of code designed by someone else, accelerating creation of user stories and acceptance criteria from simple business language, automation of test script generation, and there are already early-stage tools in exploration that are expected to write code like a human (versus assist in coding).7

An industry executive interviewed in April of 2024, quoted one of his coders as saying, “I’m not touching any code anymore without my [coding] co-pilot.”8

What might this leading indicator mean for the near future?

Tech may be leading, but all other industries may follow soon, at least in terms of coding tools. Although tech companies are early adopters and are more likely to use gen AI coding tools at present, if the efficiency gains being seen are real and significant, it seems likely that other industries will likely catch up in a year or two.

Gen AI coding tools are expected to show positive returns on investment and could be worth billons in the United States alone. There are many (many) different gen AI–powered coding tools available, and they likely deliver different productivity gains. Studies suggest that the gains from these tools are highest for relatively junior developers.9 However, if one assumes even a 5% productivity gain, a median US software developer salary of US$130,000,10 and 1.8 million software developers in the United States,11 the productivity gain from gen AI coding could be worth US$12 billion annually in the United States alone. Globally, the number would be far larger.

It would be unlikely that all that potential value could be captured by software companies that provide gen AI coding tools. Some large technology companies use a mix of third-party and in-house tools.12 One European media company has built their own gen AI coding tool, which they believe outperforms third-party tools on local language, domain-specific knowledge, and price.13

Gen AI coding tools are not perfect. Tools vary, tasks vary, and studies vary—But in general, gen AI coding tools produce code that reaches 30% to 40% acceptance rates (on the low end,) or up to 50% to 80% of usable code at the high end.14 Software developers should check AI-generated code for accuracy, since gen AI tools for coding (in common with gen AI for other applications) can “hallucinate,” and produce answers that sound right, but are in fact incorrect or biased.

Gen AI coding tools could increase technical debt. Gen AI tools allow developers to write more code, faster. At one level, that can be a good thing. Care should be taken to ensure that writing more code faster doesn’t mean lower quality or insecure code, which could increase technical debt—an implied cost of not modernizing systems and working with suboptimal performance.15 There are tools to help manage this, but it’s a potentially large concern, especially if coding tools allow nondevelopers or less experienced developers to create more unnecessary/less efficient code.

Training talent on gen AI coding tools can help. As companies accelerate adoption of gen AI coding tools, what can this mean in terms of how they think about software development talent? For one, they may need to provide more guidance and more training to ensure that the productivity gains are maximized. That’s in the near term. In the long-term, this could cause them to rethink career progression and engineering talent models.

BY

Faruk Muratovic

United States

Endnotes

  1. Deloitte interviews with tech executives, conducted from January to April 2024.

    View in Article
  2. Erin Yapis, “Hopping instead of hustling: Survey tells us how developers are taking care of business,” Stack Overflow, October 26, 2023. 

    View in Article
  3. Kathryn Bonney, Cory Breaux, Catherine Buffington, Emin Dinlersoz, Lucia Foster, Nathan Goldschlag, John Haltiwanger, Zachary Kroff, and Keith Savage, Tracking firm use of AI in real time: A snapshot from the business trends and outlook survey, US Census Bureau, Center for Economic Studies, March 2024.

    View in Article
  4. Deloitte, The State of Generative AI in the Enterprise: Getting real—Q2 report, accessed July 24, 2024.

    View in Article
  5. Deloitte interviews with software developers, conducted in May 2024.

    View in Article
  6. Ibid.

    View in Article
  7. Katherine Kostereva, “How generative AI And no-code development impact innovators,” Forbes, February 27, 2024.

    View in Article
  8. From executive interviews conducted as part of Deloitte’s State of Generative AI in the Enterprise research (2024). 

    View in Article
  9. Lucas Mearian, “Just how good is AI-assisted code generation?,” Computerworld, April 3, 2024.  

    View in Article
  10. US Bureau of Labor Statistics, Occupational outlook handbook: Software developers, quality assurance analysts, and testers, accessed July 24, 2024.

    View in Article
  11. Ibid.

    View in Article
  12. Deloitte interviews with software developers, conducted in May 2024. 

    View in Article
  13. Deloitte interview with European media company, conducted in March 2024.

    View in Article
  14. Mearian, “Just how good is AI-assisted code generation?.”  

    View in Article
  15. Mike Bechtel and Bill Briggs, “Core workout: From technical debt to technical wellness,” Deloitte Insights, December 5, 2023; Bill Doerrfeld, “Will the rise of generative AI increase technical debt?,” DevOps.com, January 25, 2024.

    View in Article

Acknowledgments

The authors would like to thank Julie Shen, Jeff Loucks, Karthik Ramachandran, David Jarvis, Andy Bayiates, Prodyut Borah, and Blythe Hurley for their help in preparing this article for publication.

Cover image by: Jaime Austin