AI Has Made Data Verification More Important Than Ever

The hackathons in 2026 look like 9 parallel dashboards pulling source-of-truth data in different shapes from different AI agents, trying to reason about what’s right and what’s wrong.

Even with deterministic systems, I always default to:

1. manually clicking and looking into the original dashboard of the ERP/CRM/ecom network/whatever system exists

2. pulling raw API requests at the lowest possible level

3. vetting raw data pulled into the data warehouse and SUM() for the given day (against time zone differences, duplications, returns, cancellations…)

Before consolidating all into three different dashboards that need to agree with each other + at least 2 red teams vetting the numbers on top.

In our software engineering classes 20+ years ago, enterprise engineering professors always reminded us that software engineering is roughly 20% new features and 80% maintenance and building on top of legacy code.

With vibe coding, it may be 30-70, but it may end up being 2-98 if coded quickly and then unfortunately plugged into production.

Ultimately, a piece of software (a system of record, a dashboard, or any other web property) remains largely a core suite of solutions and features, and ensuring that it works at all times no matter what.

If the underlying data fails, the software is practically unusable.