Yesterday’s debt was code. Today’s is ungoverned decisions.
The ubiquitous hype around the power of AI reminds us that every time a new technology is introduced, businesses are promised previously unheralded increases in speed, scale and efficiency. Early adopters and evangelists inundate us with tales of transformation and ROI that leave us desperate to jump on the band wagon. But only when it’s too late do the risks and unexpected costs.
We were at the Digital Construction expo in London in early June, and alongside all the new AI businesses promising the world, a number of speakers were highlighting the first indicators of incoming problems. Kira Barrett explains in
Publicis Sapient’s Guide to Next 2026 Industry Trends Report that the costs of AI agents are becoming apparent, with rising technical debt from unmanageable and ungoverned code. But perhaps more importantly, they’ve also introduced decision debt – untraceable automated judgements that change policy, conflict with each other based on out-of-date or poor quality data.
What makes this warning different is that the debt has now spread from software architecture to the frontline. When an agent makes the wrong call on pricing or customer service, the cost shows up in trust, compliance and revenue. The challenge has shifted from ‘can we build an agent for this?’ to ‘are we able to govern them properly?’. Without strict standards for new models and prompts and the data used for training them, new agents aren’t scaling intelligence. They are exponentially increasing confusion and business risk.
