IBM Bob helps an experienced architect code again

AI coding tools are putting senior software engineers back behind the keyboard after years spent managing teams and designing systems.

John Duigenan, an IBM Distinguished Engineer and Technology Sales Leader, began coding at age 11 but spent less time building software as he moved into senior roles—a path he says many seasoned technologists follow. His architectural experience remained, he said, but he lost some of the fluency that comes from writing code every day.

Now, agentic AI coding tools have allowed Duigenan to jump back in, turning architectural ideas into working applications himself. He uses IBM Bob, an AI-assisted development tool that helps users plan a system before generating code, to test whether companies could rebuild selected functions of specialized software-as-a-service offerings.

“The instincts and the patterns were still there, but they had become abstract, living in my head rather than in my hands,” Duigenan told IBM Think in an interview. What he had lost, he said, was “the effortlessness that used to let me move from idea to working code without friction.”

Duigenan started coding at age 11. As he took on more responsibility and moved into senior technology roles, he spent less time building software himself. He said many technology professionals follow a similar path, keeping their knowledge but losing some of the fluency that comes from daily practice.

AI tools help him close that gap by giving him a faster way to act on ideas he already understands, Duigenan said. “It gave those abstract instincts a vehicle again and let me act on them at a pace that still surprises me.”

The renewed coding fluency lets Duigenan move beyond architectural sketches and start testing ideas himself. Projects that once might have stayed theoretical can now become working software. One idea involved breaking down the functions provided by software-as-a-service (SaaS) products and rebuilding selected parts as software that a company owns and controls.

Duigenan is now testing whether AI coding tools can help companies replace some smaller SaaS subscriptions with software they build and control themselves. The effort remains an experiment, rather than a finished enterprise deployment.

“The real opportunity sits in the long tail most firms don’t think about,” he said.

Some firms have dozens or even hundreds of those services, each handling a narrow task, according to Duigenan. AI can help teams examine what each product does, identify which functions they still want to rent and rebuild others internally.

Regulated industries could find the approach especially useful when weighing data residency, compliance and vendor risk, he said.

“A few years ago, that would have stayed a thought experiment for me,” Duigenan said. “Now it’s something I can sit down and build myself.”

Bob also encourages developers to settle architectural questions before generating code, Duigenan said.

The process starts with “Why?” before moving to “How?” and “What?” The sequence pushes teams to define a system’s purpose and record major design decisions early.

“We don’t start construction without a design that’s been tested, or build a bridge without proving it can carry the load,” Duigenan said. “Starting with ‘Why?’ brings that same discipline to software.”

Faster development makes that discipline more important, he said. Teams still need to plan for security, scalability, resiliency and maintenance before they commit an architecture to code.

Duigenan said he now spends more time reviewing what AI produces than creating everything himself. Developers need to understand why the code followed a particular structure, judge whether it fits the task and decide whether they can stand behind it.

“The leaders who get real value out of AI-assisted development are the ones who invest in structure and judgment first, then let AI amplify it,” Duigenan said.