How ARIA Works

The Thesis

Howard Marks wrote in his February 2026 memo "AI Hurtles Ahead" that AI has commoditized quantitative data processing. The remaining source of investment alpha is qualitative judgment — correctly interpreting what management says, identifying contradictions between competitors, and spotting narrative shifts before the market prices them.

ARIA — Autonomous Research Intelligence Agents — was built to do exactly this. Not to replace human judgment, but to process the volume of qualitative information that no human team can cover — and surface the contradictions, patterns, and signals that matter.

The Architecture: Six Agents, Six Perspectives

ARIA runs six specialized agents in parallel against the same source material — earnings transcripts, SEC filings, and news narrative. Each agent analyzes from a different angle. Then a seventh agent fact-checks every claim before anything publishes.

How It Runs

Each week, ARIA processes a set of target companies through the full pipeline. The system ingests the latest earnings transcripts, 10-K/10-Q filings, news coverage, and analyst commentary. All six agents run in parallel, each producing their analysis independently.

The outputs converge into a synthesis layer that identifies where the agents agree, where they disagree, and — most importantly — where the contradictions are. The fact-check gate then verifies every claim before the final report compiles.

The result is a weekly publication — The Qualitative Edge — distributed to a curated list of investment professionals. Contradiction-first analysis. Every insight sourced. No consensus summaries.

Sample Output: Three Bets on the Same Table

The following is the opening analysis from Issue 2 of The Qualitative Edge, where ARIA examined JPMorgan, Goldman Sachs, and BlackRock. This is what the system produces — unedited agent output after fact-checking.

From The Qualitative Edge — Issue 2, March 2026

The three financial institutions ARIA examined this week are making fundamentally incompatible wagers on the same market — and only one side can be right.

JPMorgan's Jamie Dimon warned of "peak private credit" on the Q2 2025 earnings call. By Q3 he was describing cockroaches in the non-bank financial system — "when you see one cockroach, there are probably more." By March 2026, JPM had marked down loan portfolios to private credit funds and restricted further lending. A qualitative warning became a credit event in under nine months.

BlackRock's Larry Fink is making the opposite bet — the largest strategic wager of his tenure. The $12B HPS acquisition anchors a private markets push targeting $400B in fundraising by 2030, with new asset fee yields running 6-7x higher than 2023 vintage. When pressed on private credit default rates across four consecutive earnings calls, management declined to provide specific metrics. That opacity is a choice, not an oversight.

Goldman's David Solomon sits at the intersection. He framed the M&A cycle as "not yet in the middle" and wagered publicly that "2021 is not the ceiling" for deal activity — a specific, falsifiable claim from a measured CEO. If the credit cycle turns hard enough to validate Dimon, Solomon's IB pipeline stalls and the alternatives expansion hits institutional risk-off behavior.

That's the kind of analysis ARIA produces — contradiction-first, sourced to specific transcripts and filings, with every claim verified by the fact-check gate. Here are more examples of what the system surfaces:

Why This Matters For Your Firm

ARIA is built for equity research as a demonstration. The real opportunity is building the same kind of agent architecture around your firm's own data, research workflow, and investment edge.

The architecture underneath ARIA — multi-agent pipelines, fact-checking gates, persistent memory per company, contradiction-first analysis — is the same architecture we deploy for clients. Different data sources. Different analytical angles. Same framework.

From concept to working system in weeks, not months. That speed comes from the architecture being purpose-built for exactly this kind of work.