Finance Newsletter Writer Builds AI Story Scout, Exposing Wall Street's Data Paradox
A finance newsletter writer who hadn't touched code in 40 years just built his own AI research assistant—and in doing so, stumbled into the question now haunting every CFO with a Bloomberg terminal: what exactly are we paying for?
Marc Rubinstein, who publishes Net Interest to finance professionals, spent last week using Anthropic's Claude Code to create software that indexes his five years of back issues, then scans financial news daily to flag stories matching his editorial patterns. The system now sends him a morning email with curated story suggestions. "It's been remarkably well attuned," he wrote this week, describing the AI's ability to learn from 250-plus archived newsletters what topics he covers and how he frames market events.
The experiment matters less for what Rubinstein built than for what it reveals about the infrastructure finance runs on. His advantage, he notes, is proprietary data—those 250 back issues serve as a training corpus unavailable to generic AI models. The system learned his patterns because it had his patterns to learn from.
That observation landed the same week Euronext CEO Stéphane Boujnah delivered what may become the quote of 2026. "For years, I was told Euronext is missing the data revolution. Data is the new oil. You are missing the data boat," Boujnah said in February. "As I said earlier, all of us are finding out that maybe we missed the data boat—that maybe this data boat was a Titanic boat, that we missed the Titanic boat."
Translation: the financial industry spent a decade being lectured that market data would print money, only to watch AI models potentially commoditize the entire value chain. If an AI can learn to spot relevant financial news from a newsletter archive, what happens to the premium pricing on professional research services?
Rubinstein's homemade system can't yet replace him entirely. It can't access information "hidden in private repositories like my personal experiences or those of my network of contacts," he writes, nor can it write in his voice. (He adds, with apparent satisfaction: "Let that sink in.") But it handles the pattern-matching work that used to require human judgment about what's "on brand."
The adoption numbers suggest finance may be late to this shift. Anthropic reports the median Claude Code session lasts just 45 seconds, with only 0.1% of users running sessions longer than 40 minutes. Active users may number no more than one million. In a survey of 150 quants and research analysts in the financial community, adoption appears even thinner—though Rubinstein suggests that within regulated finance, he may actually be early to the tool.
The implication for CFOs is straightforward: if a newsletter writer can build functional AI tooling in a week after a 40-year programming hiatus, the barrier to automating research workflows has effectively collapsed. The question isn't whether finance teams can build these systems—it's whether the data vendors they currently pay can justify their pricing when clients can increasingly roll their own.
Boujnah's Titanic metaphor captures the vertigo. The data revolution happened. It just may not have happened the way anyone in finance expected—with premium providers charging premium prices. Instead, it's happening in piecemeal, as individual users with domain expertise and proprietary archives build narrow AI tools that chip away at the traditional research stack.
For finance leaders evaluating vendor contracts, the test case is now clear: what can this service do that I couldn't replicate with Claude Code and my own data in a week?
Responses (0 )