Seven Ways Tech Leaders Are Designing AI Agents To Prevent Mistakes

Gopichand Mannava, Independent Researcher, Chief Data & AI Architect, State of Connecticut.

As AI agents take on more complex tasks and consequential decisions, preventing mistakes is becoming a critical design priority. From validating assumptions and preserving context to enforcing permission boundaries and human oversight, technology leaders are finding that reliability depends on more than model intelligence alone.

To help, I ask members of the Software Architect Group, a community that I lead through Forbes Technology Council, to share the design techniques they’re incorporating into AI agents to minimize errors and build safer, more trustworthy systems.

1. Preserve Context By Reasoning Close To The Source Data

One lesson we’ve learned is that AI rarely fails all at once, but what we know is that it drifts. Small pieces of context disappear as data moves through different systems, and by the time a model reaches a conclusion, those tiny losses can compound into a poor decision. We design AI agents to reason as close to the original data as possible, preserving the signals that are often invisible once information has been aggregated or transmitted elsewhere. That shift has improved both reliability and trust because we’re reducing uncertainty before the model ever starts reasoning. - Tiffany Chang, Kneron

2. Design For Managing Mistakes, Not Eliminating Them

Treat AI like a human, since it will also make mistakes the same way a human does. Mistakes do not need to be prevented, but rather need to be handled as a human’s mistake would be. Look at how you handled human mistakes a few years back. Now handle AI mistakes just the same way. - Peter Smulovics, Morgan Stanley

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