AI Can Pay Down Technical Debt, But It Can't Touch Cultural Debt

Laureen Knudsen Technology Magazine’s #2 Chief Transformation Officer, founder of The Essential Professional & author of Becoming Essential.

gettyWhen ChatGPT launched in late 2022, I saw two ways companies responded.

One group had spent years modernizing how they worked. Within weeks, they saw ChatGPT’s potential and pulled a few teams off non-strategic work to learn the technology and share what they learned. When the tool was ready for real work, those companies implemented that knowledge immediately.

The second group was still running projects the old way, with annual budgets and project managers tracking allocation. Most didn’t fund an AI project until their 2025 fiscal year, which put them approximately two years behind.

The fast movers’ advantage was that they knew which work was non-strategic, so they could move teams without waiting for a budget cycle. That knowledge came from transformations they had finished years earlier.

I’ve watched companies move through agile, DevOps, cloud and then the effort to tie strategy to execution. In each wave, some companies changed how work and decisions flowed. Others bought the tools, renamed their meetings and called it done.

Agile is the clearest example of this, and the biggest sign was the data. In companies that played at agile, the teams themselves didn’t change. Backlogs filled with one-line stories, and no one required teams to update their work data each day, so no one knew what was done and what wasn’t. These companies still needed traditional tracking: status meetings, project managers with old tools and unreliable status reports. Companies that did agile well had real-time updates on every initiative and project. Anyone could see what was done, what was stuck and what had changed, without calling a meeting.

The companies that played at agile shared other traits, too. Quality was deferred: Testing happened after the sprint, and there was no shared definition of done, the agreed standard for when work is finished. Technical debt, the accumulated shortcuts that make every future coding change slower, was never tracked. Leaders didn’t understand agile fundamentals, so no one could tell the teams were not changing.

The same rhythm repeated in every wave: DevOps never reached fully automated deployments, cloud migrations left silos intact, and strategic goals at the top never connected to the work teams did each day.

Each of these movements, done well, gave organizations the ability to run on data. Real agile produces work-in-progress data, with no need for status meetings. DevOps produces data on every deployment and every failure. The cloud makes system data accessible. Strategy-to-execution connects spending to outcomes. Over time, companies that build the data trail can truly understand their businesses and eliminate unnecessary oversight.

For decades I’ve watched companies that played at each wave end up without that trail, and the trail is what AI needs to be useful. In a recent Deloitte survey of U.S. business and technology leaders, 72% said they lack unified, accessible data. A July 2026 BCG survey of CEOs found that only 14% have clearly defined the P&L impact of their AI initiatives. The strategy-to-execution work those companies skipped is showing up again.

Over those 25 years, I watched every new tool magnify what was true about how a company worked. Teams with strong practices got faster. Teams without them got faster at producing problems. Google Cloud’s DORA research program found the same pattern with AI in its 2025 report: AI acts as an amplifier, improving throughput but often at the cost of stability when the foundation underneath is weak.

I also watched companies bolt tools onto broken processes, and many are doing the same with AI today. Deloitte, in their study linked above, found organizations largely layering AI agents on top of existing processes, with only 21% of leaders saying their processes were ready. McKinsey reported in March 2026 the same pattern from the budget side: Companies are adding AI on top of existing systems, which increases technical debt rather than reducing it.​

​One of the hardest problems in every transformation I’ve led has been tribal knowledge: How a system really worked lived in a few people’s heads. When those people were busy or gone, the transformation stalled. That problem is now solvable.

QuantumBlack, AI by McKinsey, described a European roadside assistance company running a 25-year-old system that only two developers understood in depth. Over five weeks, AI tools helped extract more than 4,000 business rules from roughly 300,000 lines of code. Morgan Stanley has modernized more than 17 million lines of legacy code with an in-house platform, saving developers more than a million hours, the firm’s head of technology strategy told CIO Dive.

AI can also mine tickets, project records, commits and deployment logs to rebuild part of the missing data trail. It can help reconstruct some of the trail that was never built.

In my experience, agile and DevOps have stalled on cultural debt. In companies that struggled, most leaders never understood how these methodologies were supposed to work, so they couldn’t recognize when teams were only going through the motions, and they never made the changes only leaders can make.

Decision rights stayed at the top, trust between leaders and teams never formed, and leaders were unwilling to let teams operate differently.

AI can’t fix these problems. It can’t create the habit of running the business on data. That takes leaders willing to look at what the data says and change course.​

For leaders who recognize these patterns in their own company, my advice is simple. Point AI at the foundation before you build on top of it. Use it to surface the business rules in old code, find and reduce technical debt and show workflow problems. Then do the harder part yourself: Give teams the decision rights, trust and room to work differently.

The companies that were able to use AI in December 2022 did the real work years earlier. It isn’t too late to do it now, and this time, AI can carry part of the load.​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?