AI-driven coding assistants have amassed nearly $1bn of funding since the start of last year, a signal that software engineering is becoming the first “killer app” for generative artificial intelligence.
Companies such as Replit, Anysphere, Magic, Augment, Supermaven and Poolside AI raised $433mn so far this year alone, bringing the total since January 2023 to $906mn, according to Dealroom.
The rush to pour money into AI coding assistants is an indication that computer programming is the first job function to be transformed by the latest wave of AI technology.
“Today, software engineering and coding is the number-one area impacted by AI,” said Hadi Partovi, chief executive of education non-profit organisation Code.org and a long-time Silicon Valley investor and adviser to Airbnb, Uber, Dropbox and Facebook. “At this point, software engineering without AI is a little bit like writing without a word processor.”
Growing conviction in Silicon Valley of the benefits of AI coding stands in contrast to questions among some investors about the economic benefits of generative AI and likely returns on Big Tech’s projected trillion-dollar investment into computing infrastructure to support the technology over the coming years.
Hannah Seal, a partner at Index Ventures, which has invested in start-up Augment, alongside Eric Schmidt and others, said it was “much easier to monetise AI if you can embed your product into an existing workflow, and make the benefit instantly visible”.
For AI tools to make money, the questions for Seal are: “What is the time to value, and how meaningful is that value-add?”, while she added that “with coding co-pilots, the answer is very clear”.
AI enthusiasm has prompted start-ups and tech giants Microsoft, Amazon, Meta and Google to vie for dominance in a crowded sector, building AI assistants and agents that can write and edit computer code.
An executive on Code.org’s board, which includes David Treadwell, Amazon’s head of ecommerce, and Kevin Scott, Microsoft’s chief technology officer, recently told Partovi their company would stop hiring people who code without AI by the end of the year, he said.
“The easier [programming] becomes, the more demand goes up, because so much more technology can be built,” Partovi added.
Microsoft-owned GitHub, the world’s biggest software development platform, was one of the first to turn a large language model — software that underpins ChatGPT, which can generate text, images or code — into a coding assistant.
“When using GPT-3, OpenAI’s first major model, we figured out relatively quickly that it was so good at writing code that we could build a product around this,” said Thomas Dohmke, chief executive of GitHub, which was acquired for $7.5bn by Microsoft in 2018.
The prototype turned into GitHub Copilot, an AI coding assistant that was launched widely in 2022 and has nearly 2mn paying subscribers. “Now, the model writes better code than the average developer,” Dohmke said.
As of April, GitHub’s revenue was up 45 per cent year on year and, according to Microsoft chief executive Satya Nadella, its annual revenue run rate was $2bn at the start of this month.
“Copilot accounted for over 40 per cent of GitHub revenue growth this year and is already a larger business than all of GitHub was when we acquired it,” he said on a July 30 earnings call.
More than 77,000 organisations — from BBVA, FedEx and H&M to Infosys and Paytm — had adopted the two-year-old tool, Nadella said, a figure that showed a 180 per cent rise year on year.
IT departments of large companies nonetheless retain some reservations around the security implications of using automated programming tools to produce production-grade code.
Dohmke said though that he would not expect AI-generated code to be deployed without manual checks and balances.
“In general, we see between 20-35 per cent productivity gains from enterprises that have reported internal statistics,” Dohmke said, referring to customers such as Latin American ecommerce giant Mercado Libre and professional services group Accenture.
A McKinsey analysis from last year found that the direct impact of AI on the productivity of software engineering could range from 20 to 45 per cent of current annual spending on the function, with benefits including generating initial code drafts, code correction and refactoring.
“By accelerating the coding process, generative AI could push the skill sets and capabilities needed in software engineering toward code and architecture design,” McKinsey said.
Software engineers say they have already integrated AI assistants into their daily workflow, and it helps not only to be quicker but also more creative.
“I personally code every day with GitHub Copilot, oftentimes alongside ChatGPT,” said Marc Tuscher, a deep learning scientist and chief technology officer of Sereact, a German robotics start-up.
GitHub’s tool is most useful for “repetitive tasks”, such as for user interfaces and the back end of products, he added, while he uses ChatGPT to help with more abstract problem solving.
“ChatGPT will come up with some classical ideas, some new papers and then you can ask, ‘how would this be done in Python?’ and it produces code,” Tuscher said. “Both tools are very, very cool.”
While all programmers he knows use these products, and “it changes fundamentally how we work”, Tuscher said the tools were no more than powerful helpers, rather than replacements, for coders.
“No GenAI knows about good software architecture, or how to put systems together,” he added. “That’s still the thing we have to think through ourselves.”
Read the full article here