How debate between AI economists creates a verification loop that improves analysis quality. The economists are AI agents powered by Claude Opus 4.6 Agent Teams.
For a personal macroeconomic analysis project, I tested whether adversarial review could reduce errors. I gave two AI economists (powered by Claude Opus 4.6) the same task: analyze the current state of the U.S. economy. But instead of working independently, they had to debate each section before publishing. Only when both agreed could the section be published.
<aside>
Economist A (Drafter) - Writes initial analysis
</aside>
<aside>
Economist B (Challenger) - Reviews and challenges claims
</aside>
Claude Code - Working with Agent Teams (Opus 4.6)
In Section 2 (Leading Indicators), Economist A analyzed bank lending standards using the DRTSCILM indicator (Net Percentage of Domestic Banks Tightening Standards for Commercial and Industrial Loans to Large and Middle-Market Firms) :
<aside>
Economist A's Draft:
"Bank Lending Standards (DRTSCILM): 5.3% [15] - Easing significantly - The net percentage of banks tightening lending standards has dropped from 6.5% to 5.3%, indicating improving credit availability for businesses. This is a positive signal for investment and expansion over the next 6-12 months."
Signal count: 2 green (expansion), 3 yellow (mixed), 1 red (recession risk)
</aside>
<aside>
Economist B's Challenge: DRTSCILM Interpretation Completely Backwards (CRITICAL)
"MAJOR ERROR: You classified DRTSCILM as 'easing' and counted it as a GREEN expansion signal. But DRTSCILM measures NET TIGHTENING. Positive values = more banks tightening than easing.
5.3% means banks are STILL net tightening credit—just less tight than 6.5%. This is not 'easing.' It's 'less tight but still restrictive.'
This error reverses your interpretation: Credit is contractionary (YELLOW/RED), not accommodative (GREEN). Your signal count is wrong."
</aside>
<aside>
Economist A's Pushback:
"But the TREND is improving—6.5% down to 5.3%. Doesn't the improving direction matter?"
</aside>
<aside>
Economist B Holds Firm:
"The trend matters, yes. But positive = net tightening. You can call it 'improving from very tight to moderately tight,' but you CANNOT call it 'easing' or count it as green. That's factually wrong."
</aside>
<aside>
Economist A's Concession:
"You're right. I misread the indicator. Positive = net tightening. I'll reclassify this as YELLOW (improving but still restrictive) and update the signal count."
</aside>
Revised version:
"Bank Lending Standards (DRTSCILM): 5.3% [15] - Improving but still restrictive - The net percentage of banks tightening lending standards has dropped from 6.5% to 5.3%. Positive values mean more banks are tightening than easing, so credit conditions remain contractionary, though the improvement suggests conditions are gradually normalizing."