Code generation has emerged as a top use case for generative AI. Is your organization ready for it?
Generative AI’s ability to create text and image-based collateral for marketing, product design and other business functions is well known. Yet that’s not the use case drawing the most buzz of late.
That distinction goes to software development, where AI “copilots” have captivated the coding world. Organizations are using GenAI copilots that generate or edit code to streamline routine software tasks such as testing, debugging and language translation.
GenAI coding has caught on at NVIDIA, the chipmaker whose GPUs have an outsized influence on the market. “We use generative AI for coding quite extensively here at NVIDIA now,” said NVIDIA CEO Jensen Huang during the company’s second quarter earnings call.
As AI copilot use grows, it requires organizations to reconsider how to build software holistically, including educational initiatives aimed at reskilling and upskilling developers. In the meantime, individuals and businesses alike are raving about the productivity boost these GenAI tools provide.
Coders Are Coding Less
Eureka Labs Founder Andrej Karpathy shared how he’s accelerating software creation using AI code editor Cursor in conjunction with the Claude 3.5 Sonnet large language model (LLM). “It’s a bit like learning to code all over again but I basically can’t imagine going back to ‘unassisted’ coding at this point, which was the only possibility just ~3 years ago,” Karpathy said.
Inspired by this man-machine collaboration, Theory Ventures General Partner Tomasz Tonguz used Cursor, Claude 3.5 Sonnet, and other tools to build a go-to-market survey analysis. The tools helped Tunguz move away from memorizing programming libraries by simply using English language prompts.
“Coding this way, I explored the data much more deeply, more rigorously, and& more quickly than I would have otherwise.”
Organizations can reduce by roughly half the time it takes to generate, refactor and document code, McKinsey estimated a year ago. That may be conservative when you consider statistics reported by large enterprises.
Amazon used its Q digital assistant to cut the time it takes to upgrade Java applications from 50 developer days to a few hours, or the equivalent of 4,500 developer years’ worth of work, according to CEO Andy Jassy.
Such results have the company thinking more broadly—on the scale of workforce transformation. Amazon Web Services CEO Matt Garman went as far as to suggest that AWS software developers may spend most of their time coding in two years.
Instead, Garman expects his coders to upskill, including learning new technologies and engaging more with customers.
Guardrails for the Coding Copilot Future
Abstracting software development wholesale may prove aspirational for organizations still trying to use GenAI to build bespoke digital assistants. Even so, Garman’s position merits exploration and provides an opportunity for IT leaders and their business peers to rethink their operations.
IT departments can prepare for AI copilots by partnering with their HR counterparts to create formal reskilling and upskilling programs for existing and future software programmers.
Ideally, organizations will identify and tab “power users” who are fluent in the latest GenAI coding tools to lead instruction. Power users will be essential in articulating the value of copilots to senior leadership and developers—those in the trenches who will benefit from using them to augment their work.
Research suggests employees want to grow their skill sets: Eighty percent of employees surveyed by EY said they would be more comfortable using AI at work if they were afforded more opportunities for training and upskilling.
Organizations will still need engineers who can think programmatically, from designing algorithms to guiding teams through shifts in the evolving GenAI technology landscape. For example, even now some organizations are thinking about how to run their business using autonomous agents, which will be a significant force in the future. One imagines such engineers might work alongside copilots to ensure that agents are doing what they’ve been designed to do.
These engineers will also help articulate sound software architecture and governance, including the safeguards that must be built around GenAI to reduce reputational risk.
Moreover, since LLMs are well known to produce incorrect information and feature human biases, having humans in the loop helps ensure the outputs produced by AI-assisted software development are accurate and relevant. Human monitoring is table stakes for organizations planning to input proprietary code into their models.
Copilots may be the hot new AI software today, it’s unclear what tomorrow will bring.
The reality is there is no easy button for reorienting software development in the AI era—either at the strategic or keyboard level. Regardless, you want to ensure you have the right technology foundation in place, as well as access to partners who can help guide you on your GenAI journey.
That is why Dell created the AI Factory, a comprehensive approach to bringing AI to your data, infrastructure to support your GenAI applications and access to the growing ecosystem of AI partners, as well professional services and use cases.
Learn more about the Dell AI Factory.