You know what, folks? I think AI has become pretty darn good at writing code. Apparently, IBM agrees. You see, the company now believes the bigger problem is what happens after the code is generated.
Today, IBM announced a major update to its Bob AI software development platform. While the release includes plenty of new features, IBM’s biggest message is that reviewing and validating AI-generated code has become the new bottleneck for software teams.
The company points to a survey showing that 85 percent of DevSecOps professionals believe AI has shifted the bottleneck away from writing code and toward reviewing it. That’s an interesting statistic, although it’s worth remembering the research was cited by IBM as part of its announcement.
To help address that challenge, IBM has added multi-agent capabilities to Bob. Rather than relying on a single AI assistant, the platform can coordinate multiple specialized agents across different development tasks.
IBM also introduced Bobalytics, a new analytics tool designed to give organizations visibility into AI usage, productivity, performance, and costs. The idea is to help enterprises decide which AI models should handle which tasks while keeping spending under control.
Another major focus is modernization. IBM is rolling out pre-built AI workflows for IBM Z mainframes, IBM i systems, and Java applications. The packages include AI-assisted tools for modernizing COBOL, PL/I, RPG, JCL, and large Java codebases.
For organizations still relying on decades-old software, that could end up being more valuable than simply generating code faster.
Of course, IBM also highlighted customer success stories. Financial technology company Jack Henry said Bob has helped its developers work more efficiently with a large RPG codebase.
Cloud consulting firm Blue Pearl made the boldest claim. It said a legacy modernization project that was originally estimated to take nine months with a team of 14 engineers was completed in just three days. Woah. That’s certainly eye-catching, but IBM didn’t provide enough technical detail to independently evaluate exactly what was accomplished.
Will every organization will see those kinds of results? That’s the real question, I suppose. Still, IBM’s broader argument makes sense. AI can generate code surprisingly well, but that code still has to be reviewed, tested, secured, and maintained.
If IBM is right, the next big challenge in AI-powered software development won’t be writing more code. It’ll be making sure the code AI writes is actually ready for production.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz