Build Your Own Bloomberg Terminal With AI

Brooker Belcourt, Every’s head of financial services consulting and former Perplexity finance lead, walks through four levels of AI-augmented investment analysis using a Shopify earnings preview as the running example.

Level 1: Custom ChatGPT project with SEC filings — solid but limited by 8K character instruction cap. Level 2: Claude with Skills (reusable prompt files) and MCP data connectors like Daloopa for live financial data. Level 3: Claude Cowork with local file access — reads your proprietary Excel models, call notes, and thesis documents. Level 4: Claude Code with a single /work command that runs continuously, building an interactive Bloomberg-style dashboard covering news, previews, reviews, and thesis tracking across an entire coverage universe.

Belcourt estimates Level 4 compresses 160 hours of quarterly analyst work into the time Claude Code takes to run plus about an hour of human review. The piece is fully ungated — a rare free article.

RDCO mapping: Exceptional reference for anyone building domain-specific AI workflows. The four-level progression from ChatGPT to Claude Code mirrors our own maturity curve. The MCP + Skills + local files architecture is directly relevant to how we structure the COO agent. Strong candidate for cross-reference in any content about AI tool selection and progressive automation.