{ "video": { "id": "uA4ah-dVHvU", "title": "AI Cut Our Research Time From 7 Days to 15 Minutes — Gareth Yoder, Vanguard", "duration": 939, "upload_date": null, "channel": "AI Engineer", "source": "AI Engineer", "event_format": "" }, "analysis": { "video_id": "uA4ah-dVHvU", "title": "AI Cut Our Research Time From 7 Days to 15 Minutes — Gareth Yoder, Vanguard", "one_liner": "Vanguard exposed its legacy mainframe codebase to AI tools through an MCP server, which cut the research feedback loop for modernizing its cash ledger from 7 days of mainframe-engineer analysis to a 15-minute AI prompt.", "summary": "Gareth Yoder, the cash program technical lead at Vanguard, describes modernizing the cash ledger behind Vanguard brokerage accounts. The ledger handles up to 30 million service calls and half a million ledgers a day. Because the work was an enterprise priority, data, producer and consumer migrations ran in parallel. The project began before AI tooling was available and adopted it along the way: an MCP server over the legacy codebase and migration docs sped up research and reduced producer support, and AI compiled consumers' business-case descriptions into a picture of the modernized use cases and the impact of SLA breaches. He closes with five leadership lessons: make legacy code AI-queryable, treat migration docs as product, reimagine instead of lift and shift, adopt AI incrementally, and treat observability as non-negotiable.", "key_points": [ "The cash ledger covers all money moving into Vanguard brokerage accounts and between accounts, over payment rails such as ACH and wires, through web, mobile and batch channels. It runs active-active, multi-region and fault-tolerant, with up to 30 million service calls and half a million new ledgers per day.", "The modernization was split into three pillars (data migration, producer migration and consumer migration), and all three had to run in parallel because it was an enterprise priority.", "Making the legacy codebase available to AI tooling through an MCP server was the turning point. Research that took a mainframe engineer 7 days now takes one AI prompt, so the feedback loop went from 7 days to 15 minutes.", "Data migration used standard change data capture with daily audits. Row-level counts checked consistency, and column-level checks (amounts, account information, state and federal withholding) checked integrity. His advice: don't over-engineer this code, because it is deprecated once modernization finishes.", "Legacy producers wrote data three ways: service requests, an API wrapping those requests, and direct table inserts, the last of which left data in the database with no clear producer. The modern design put a common data schema on a gateway that wraps the cloud insert APIs, using idempotency borrowed from FinTechs and a Snowflake-ID-like paradigm for consistency across multiple writers.", "35 producers had to migrate. Migration docs moved from manual documentation into the company wiki and were exposed through the MCP server, which greatly reduced chats and producer support. AI tools also mapped the legacy ledger status codes to the new human-readable enums.", "50 consumers read the data through a shared schema registry over three channels: real-time GraphQL (for example, funds available, which drives the displayed balance), batch data-share views from the data warehouse scoped to each consumer, and a Kafka topic with at-least-once delivery and idempotency-safe operations, which fraud teams use for real-time ingestion.", "Each consumer had to supply a description and a business-case justification in addition to its three-character app code. AI compiles these into a list of modernized use cases and top business functions, and can estimate the impact of an SLA breach (a batch job may give hours to respond; a real-time API with a client receiving errors gives minutes). Today 90% of cash program engineers use AI tools daily, with multi-agent workflows and a shared 'team brain' knowledge base." ], "takeaways": [ "Make your legacy codebase AI-queryable by exposing it through an MCP server, so developers stop waiting days for mainframe experts.", "Treat migration docs as part of the product and expose them through MCP, so producers and consumers can self-serve onboarding.", "Reimagine rather than lift and shift. Use the migration to add idempotency, consistency across multiple writers and per-consumer data views. The consumers and producers that were hardest to onboard were the ones that wanted a direct lift and shift.", "Fix forward, not in the legacy system: let fixes in the modern state be the incentive to migrate, and keep whitelisting so audit logs still make sense.", "Collect business-case justifications from every consumer and feed them into observability, so incident response knows who is affected and how urgent it is. Adopt AI incrementally instead of waiting for a perfect starting point." ], "topics": [ "legacy modernization", "mcp", "mainframe migration", "data migration", "event streaming", "observability", "ai adoption", "fintech" ], "tools": [ "Vanguard", "MCP (Model Context Protocol)", "GraphQL", "Kafka", "Snowflake ID", "change data capture (CDC)", "schema registry", "ACH", "company wiki" ], "quotes": [ { "text": "What was taking 7 days for a mainframe engineer to go into the code, do their analysis, and then kind of and come back to us for an answer, now could all be done via an AI tool prompt.", "at": "02:43", "url": "https://www.youtube.com/watch?v=uA4ah-dVHvU&t=163s" }, { "text": "We are reimagined and we didn't do a lift and shift.", "at": "05:08", "url": "https://www.youtube.com/watch?v=uA4ah-dVHvU&t=308s" }, { "text": "Don't fix the past.", "at": "12:33", "url": "https://www.youtube.com/watch?v=uA4ah-dVHvU&t=753s" }, { "text": "There's never going to be a perfect starting point in the world that we live in today.", "at": "14:44", "url": "https://www.youtube.com/watch?v=uA4ah-dVHvU&t=884s" } ], "words": 2526 }, "summary_url": "/#uA4ah-dVHvU", "transcript": { "html": "/transcripts/uA4ah-dVHvU.html", "txt": "/transcripts/uA4ah-dVHvU.txt", "vtt": "/transcripts/uA4ah-dVHvU.vtt" } }