Source: AWS Machine Learning Blog — 2026-08-03
Summary
Formula 1 and AWS built the "Data Accelerator," a system that uses Amazon Bedrock AgentCore agents to automate onboarding new MarTech data sources into F1's data platform — a process that previously took engineers up to 8 weeks of manual work and now takes roughly 40 minutes of automated code generation plus a short deployment step. The same agentic system also handles automated schema-drift/anomaly detection in production and tracks data-platform and agent lineage in a unified view, opening self-service data onboarding to analysts and scientists beyond the core data engineering team.
Key Takeaways
- Onboarding flow: a team member uploads a business requirements document (BRD) to Amazon S3, which triggers an AWS Lambda function that invokes a Bedrock AgentCore Runtime agent.
- The agent reads the BRD and generates the configuration files needed to onboard the new data source — the step that used to take engineers weeks of manual work.
- The agent then pushes the generated files as a pull request to F1's standardized Git repository via a GitHub App, and opens a linked Jira ticket through Jira's REST API — no engineer writes the initial boilerplate.
- Headline metric: onboarding time dropped from as much as 8 weeks of manual engineering to about 40 minutes of agent-driven code generation.
- Beyond onboarding, the system also detects data-source anomalies/schema drift in production and gives teams a single window into data-platform operations and agent lineage.
Reel Script
Hook (~18s, ~40 words) Formula 1 used to spend up to eight weeks just wiring up a single new marketing data source — approvals, configs, tickets, the works. Then they handed the job to an AI agent. Now it takes about forty minutes.
Core Concept (~70s, ~155 words) This isn't a chatbot answering questions about F1 data — it's an agent that does the actual engineering work of connecting a new data source into a pipeline. The problem it's solving is boring but expensive: every time F1's marketing team needed to plug in a new ad platform or partner feed, someone had to write config files, get them reviewed, deploy them, and track the whole thing through a ticket — classic multi-week enterprise data-onboarding grind. AWS built this on Bedrock AgentCore, which is basically a managed runtime for letting an agent take real actions — read a document, generate files, call other systems — rather than just chat. The trigger is dead simple: someone uploads a plain business requirements document, and that single upload kicks off the whole automated chain.
Hands-On (~80s, ~180 words) Walk through what actually happens after that upload. The BRD lands in an S3 bucket. That triggers a Lambda function, which wakes up the Bedrock AgentCore agent. The agent reads the requirements doc like an engineer would, and generates the actual configuration files needed to onboard that data source — the code that used to be written by hand. But it doesn't stop at generating files: the agent uses a GitHub App to open a real pull request against F1's standardized repository, and separately calls the Jira REST API to file a ticket that references that PR. So a human still reviews and approves the PR — this isn't a black box shipping straight to production — but the entire drafting, generating, and ticketing chain that used to eat weeks of engineering time now runs in minutes. And the same agentic layer keeps working after onboarding: watching for schema drift and data anomalies in production, and tracking lineage across the whole platform in one place.
Takeaway (~22s, ~55 words) The real win here isn't "AI writes code" — it's that F1 turned a multi-week cross-team workflow into a single document upload with a human approval gate at the end. That's the pattern worth stealing: agents doing the tedious middle of a workflow, humans still signing off. If your team has an 8-week onboarding process buried somewhere, ask what your BRD-to-PR step could look like.