The emergence of large language artificial intelligence (AI) models for programming languages opens up the possibility of AI developers and AI assistants to human software developers.
While it has quickly become the must-have programmer’s aid, at a recent roundtable hosted by GitLab, Simon Dawson, head of engineering at Atom Bank, discussed how AI can also be used for upskilling and training, and help managers identify inexperience software developers with potential.
“We’ve got lots of senior software engineering staff that we need in our organisation, but we are trying to look to grow more at the junior level and help those people to become senior developers, maybe sooner than they would have,” he said.
While it is unclear how far AI technology will go to help junior software developers grow their skills, Dawson said: “We are making an assumption around the technology, but there will be some unintended consequences of AI that we don’t yet know about yet.”
During the roundtable discussion, Kishor Toshniwal, enterprise architect at Community Fibre, discussed how he recently went back to coding thanks to the help of an AI, which assisted him in developing some scanning software for the equipment the telco installs.
He recognised that having the ability to scan the equipment via a mobile phone would remove one of the biggest headaches Community Fibre’s installers face. “I had no clue how to scan using the phone’s camera. Neither did my developers. But by using an AI code generator, we were able to identify the code and bring it into our environment.”
The experience offered him a way to return to coding. “To be honest, for the last 15 years, I have not written a single line of code. But I was excited and so happy that I had something working. This has changed my perspective. I can now put 25% of my work time towards core software development. I believe large tech companies have a lot of senior managers who are also writing code again, and this is a good thing,” he said.
Having an AI assistant help with coding is seen by those in senior managerial roles as a way to dabble in coding. David DeSanto, chief product officer at GitLab, added: “Our VP of engineering still writes code at GitLab.”
At another event – the recent Microsoft AI Tour in London – Microsoft CEO Satya Nadella discussed how he uses Microsoft’s own AI coding engine in GitHub. “I always joke that every weekend I can go back to programming with GitHub Copilot and actually finish a project in a finite time,” he said.
Measuring the benefits of AI
Although it is an existing GitLab customer and GitLab would very much like Atom Bank to pilot its AI technology, called Duo, Dawson believes it is very important to assess the outcomes the bank would want to achieve from such a project.
For Dawson, one of the challenges in deploying any AI system in the workplace is how to measure the business benefits. “It is quite difficult and can be quite subjective. But you can look at regular tasks. How long would it take you to do a regular task unassisted versus assisted?” he said.
The company’s experience with the Google Gemini AI assistant in the workplace for office productivity showed that AI is very good at note taking and summarising meetings, which he said means people are much more engaged in meetings. From his experience of using Gemini, Dawson suggested that IT and business decision-makers look at the baseline for a business process and then decide on the output they want to achieve by deploying AI.
Discussing an approach to measuring productivity achieved by deploying the GitLab Duo AI programming tool, DeSanto said: “When people start using GitLab Duo, people ask questions about measuring productivity. We encourage them to not look at things that are the vanity metrics you’ll hear online, such as counting the number of lines of code.” Instead, he urged IT managers to be creative. “Maybe you want to have less code, but have higher quality code.” He said Duo can also be used to help developers new to a project get up to speed quicker.
AI can be used to automate operational tasks, which DeSanto said “allows the brain to focus on the harder problems”.
There are clearly short-term and medium-term benefits AI can offer software development teams. However, for Dawson, AI also presents a societal risk. “We need to be careful that the technology doesn’t make human beings not quite as intelligent as they are today,” he warned. While he admits there is no evidence to support this, Dawson added: “It is something that I have on my mind, because you do learn by doing things.”
In other words, if AI takes away the on-the-job learning that new software developers experience, they may miss out on some of the fundamental knowledge building blocks they need to succeed in their career. On the other hand, there is no shortage of senior managers who want to dabble in coding again and AI gives them a chance to do so in their spare time, by filling gaps in their knowledge.