Stakeholders are worried about the future of software engineering and everyone is busy making predictions. While some claim that most coding jobs will cease to exist, others say that there will be an abundance of software engineers in the future. What is the truth?
Remember Devin? The AI software engineer made by Cognition Labs propelled people to talk about job displacements. Now, Russell Kaplan, who has joined as the president of Cognition Labs, is making several predictions about the future of software engineering.
“Models will be extraordinarily good at coding very soon,” he stated, emphasising that research labs are prioritising coding and reasoning improvements above all else, as was evident with OpenAI, Meta, and Google releasing models for maths.
This investment will soon pay off, transforming the landscape of software development. Kaplan highlights a unique advantage in coding: the potential for superhuman data scaling through “self-play”.
A Team of AI Agents
If models can write code, test it, and check for self-consistency, a form of automatic supervision that is not possible in most domains due to the limits of human expertise will usher in. This capability would allow code to be tested empirically and automatically by AI agents.
As a result of this, software engineering will see radical transformation. “True coding agents, which do tasks end to end, will complement today’s AI copilots,” Kaplan explained.
This will make every engineer akin to an engineering manager, delegating basic tasks to coding agents while focusing on higher-level aspects such as understanding requirements, architecting systems, and deciding what to build.
“Any value that humans can bring here is either in leading the direction of deciding what to work on (eg. product), or in working at the cutting edge of what AI cannot do yet (researching new models),” agreed Jeremy Bernier, founding engineer at Canvas.
This is similar to what Franoic Chollet, the creator of Keras, said a few months ago. He predicted that there would be around 10 million more coding jobs in the next five years, but those would be for people who have the expertise in writing code in Python, but most of the coding would be done by AI.
Chollet had earlier said, “If you could fully automate software engineering (my job), I think that would be great, since I could then move on to higher-leverage things. Making software is a means to an end, not the end.” He added that software engineering is not just about copy-pasting code, but about developing mental models of problems and their solutions.
The Age of Abundance
What’s certain is that the role of software engineers will evolve significantly. “There will be way more software engineers in the future than the present,” Kaplan noted. But “the job will just be very different: more English, less boilerplate coding”.
According to this prediction, engineers will adapt, much like they did when transitioning from assembly language to Python. Funnily, it’s likely that English teachers would make the best engineers.
Interestingly, this could also usher in the era where advertisements will be focused on AI agents, and not developers. Companies that market to developers will start targeting coding agents as well. “Your agent might decide what cloud you use and which database you choose,” said Kaplan. And there would be an army of those agents.
Just as Reid Hoffman predicted the decline of traditional 9-to-5 jobs due to the rise of the gig economy and a focus on results, the standards for product quality will increase as developers will be able to ship products much faster. This would call for better testing infrastructure, where agents can automatically train, test, and verify models and software.
“There’s never been a better or more productive time to be a builder,” he concluded, adding that softwares like Devin would usher in the age of abundance for software.
“Code is largely worthless, more of a liability than an asset. Problem-solving is where the value is,” Chollet added in his thread. Because software engineering is not just about writing or copying code or giving prompts to AI models, it is much more than that.
Offices in the future would probably have ‘Human Software Engineers’ managing a team of AI engineers, just as we predicted.
