Earlier this year, LinkedIn co-founder and venture capitalist Reed Hoffman issued a warning mixed with amazement about artificial intelligence. “There is real magic going on,” Hoffman said, speaking to tech executives across sectors of the economy.
Some of this magic is becoming more evident in creative spaces, such as the visual arts and the idea of “Generative Technology” has captured the attention of Silicon Valley. Artificial intelligence even recently Prizes won in art galleries.
But Hoffman’s message was aimed squarely at executives.
“Artificial intelligence is going to change all industries,” Hoffman told CNBC’s Technology Executive Board members. “So everyone should think about it, not just data science.”
The rapid progress you make MicrosoftAI co-pilot, and GitHub Open Source Automated Code Writing ToolOne example was directly cited by Hoffman as an indication that all companies are better prepared for artificial intelligence in their world. He said that even if they do not make significant investments today in AI, business leaders must understand the pace of improvement in AI and upcoming applications or else they will be “sacrificing the future.”
“100,000 developers got 35% of their coding suggestions from KoPilot,” Hoffman said. “That’s a 35% increase in productivity, and it’s outside of last year’s model. … Across everything we do, we’re going to have amplification tools, and we’ll get there in the next three to 10 years, and it’s the foundation for everything we do,” he added.
Copilot has already added another 5% to the 35% mentioned by Hoffmann. GitHub CEO Thomas Dohmke recently told us that Copilot is now handling up to 40% of coding among AI programmers in beta testing over the past year. In other words, for every 100 lines of code, 40 lines are written by the AI, With total project time shortened by up to 55%.
The co-pilot, trained in massive amounts of open source code, monitors the code the developer writes and acts as a co-pilot, taking input from the developer and making suggestions about the next line of code, often multi-line coding suggestions, often “the standard code required but a waste of time for a human to recreate it.” We all have some experience with this kind of AI now, in places like our email, where both Microsoft and Google mail programs suggest the next few words we might want to type.
AI can make sense about what might come next in a text string. But Dohmke said, “He can’t do more, he can’t grasp the meaning of what you want to say.”
He says whether the company is a supermarket working on payment technology or a banking company working on customer experience with an app, they are all already software companies. They all build software, and once the C-suite has developers, it needs to consider developer productivity and how to continually improve it.
This is where 40 lines of code come in. “A year after Copilot, about 40% of the code was written by an AI where Copilot was enabled,” Dohmke said. “And if you show that number to CEOs, it’s amazing for them. … they do the math about how much they spend on developers.”
With projects completed in less than half the time, the logical conclusion is that there will be less work for humans. But Dohmke says another way to look at a software developer’s job is that they do many more “high value” tasks than simply rewriting code that’s already out there. “The definition of ‘highest-value’ work,” he said, “is to remove the menial work of writing things that have been done over and over again.”
The goal of Copilot is to help them “stay in the flow” when they are on the job of coding. Domky said that some of the time spent writing code is spent searching for existing code to plug in from browsers, “extracts from someone else”. This can distract programmers. “Eventually they are back in edit mode and copy and paste a solution, but they have to remember what they were working on,” he said. “It’s like a surfer on a wave in the water and they need to find the next wave. The co-pilot keeps them in the editing environment, in the creative environment and they come up with ideas,” Domki said. “If the idea doesn’t work, you can reject it or look for the closest idea and you can always modify it,” he added.
The GitHub CEO expects more of these Copilot code suggestions to be taken in the coming years. “In the next five years, it’s going to be 80% or so,” Domky said. “It’s not an exact science…but we think it’s going to grow exponentially,” he said of those predictions, unlike much that happens in the computer field. After being on the market for a year, he said new models are getting better fast.
As developers reject some of the code suggestions from Copilot, the AI is learning. “As more developers adopt Copilot, it gets smarter by interacting with similar developers to a new coworker, and learning from what is acceptable or not,” Domke said. He said that new AI models aren’t released every day, but every time a new model becomes available, “we might make a leap.”
But it is still far from replacing humans. “Today’s co-pilot can’t do 100% of the job,” Domky said. “He’s not conscious. He can’t create himself without user interaction.”
With Copilot still in private beta testing among individual developers — hundreds of thousands of developers — GitHub has not announced any institutional clients, but it expects to start naming business clients before the end of the year. Enterprise pricing information hasn’t been revealed yet, but in beta testing, Copilot pricing has been set at a flat rate per developer – $10 per person per month, which developers often spend on corporate cards. “And you imagine what they earn per month, so it’s a marginal cost per month,” Domki said. “If you look at 40% and think about improving productivity, and you take 40% of opex spending on developers, then $10 is not a relevant cost. … I have 1,000 developers and more money than 1,000 x 10,” he added.
The CEO of GitHub sees what’s happening now with AI as the next logical stage for the productivity advancement in the coding world he’s been a part of since the late 1980s. It was a time when programming was out of the punch card stage, there was no internet, and programmers like Dohmke had to buy books and magazines, and join computer clubs for information. “I had to wait to meet someone to ask questions,” he recalls.
That was the first phase of developer productivity, then came the Internet, and now it’s open source, allowing developers to find other developers on the Internet who have already “rolled the wheel,” he said.
Now, whether the task of coding is related to payment processing or logging in to social media, most companies – whether they are startups or enterprises – put open source code. “There is a huge dependency tree of open source already out there,” Domky said. It is not uncommon for up to 90% of the code on mobile applications to be pulled from the Internet and open source platforms such as GitHub.
In the age of “everything that’s already available” coding, that wouldn’t be what stood out for a developer or an app.
“AI is only the third wave of this,” Domki said. “From punch cards to building everything ourselves to open source, to now within a lot of code, AI is writing more,” he said. “With 40%, soon enough if AI spreads across industries, innovation will be generated on the phone with the help of AI and the developer,” he added.
Today, and for the foreseeable future, Copilot is still a technology that trains on code, and makes proposals based on searching for things in the code library. Domky said that he does not invent any new algorithms, but that given the current pace of progress, “it is quite possible that with the help of a developer he will create new ideas for the source code.”
But even that still requires a human touch. “Copilot is getting close, but it will always need developers to create innovation,” he said.