Back office ops · Production

Coinbase builds RAPID-D: a multi-agent AI decision support system to augment the RAPID decision framework

The problem

Coinbase's RAPID decision framework provided accountability structure but lacked a systematic way to surface unseen risks, mitigate cognitive bias, and provide a transparent, auditable layer of analysis for critical strategic decisions.

Workflow diagram · grounded in source
1
RAPID document submitted
trigger
“At Coinbase, we use RAPIDs for making critical decisions where inputs are taken from key stakeholders across functions and an accountable "decider" makes the final decision based on these insights”
2
Analyst agent baseline review
ai_action
“The agent generates a baseline recommendation based strictly on the facts and arguments presented in the document”
3
Seeker agent context retrieval
ai_action
“It then leverages our enterprise search tool to find answers across all internal knowledge sources. By synthesizing these findings, it provides a deeply informed decision with wider organizational context that might otherwise be missed”
4
Contrarian agent challenges recommendation
ai_action
“This agent's sole purpose is to build the strongest possible case against the initial recommendation. It deliberately probes for weaknesses, unstated assumptions, potential risks, and unintended consequences”
5
Synthesizer produces final recommendation
ai_action
“It then produces a comprehensive final recommendation for the human Decider, complete with a detailed explanation of its reasoning and the trade-offs it considered”
6
Human Decider reviews AI output
human_review
“It is not designed to replace human oversight but to augment it with a scalable, AI-driven diligence engine”
7
Feedback captured and integrated
feedback_loop
“Comments or corrections — whether provided by the user during an active session or later by any stakeholder in the RAPID document — are captured and analyzed against the assistant's original recommendation. This evaluation is then used t…”
Reported outcome

RAPID-D makes the decision-making process transparent, consistent, and reproducible; Claude 3.7 Sonnet was selected after benchmarking across leading models for its quality, stability, and reliability.

Reported metrics
RAPID-D recommendation accuracy benchmarkbenchmark scores across leading models
Model selection rationalestrong balance of quality, stability, and reliability
Reported stack
Claude 3.7 Sonnetenterprise search tool
Source
https://www.coinbase.com/en-it/blog/making-smarter-decisions-faster-with-AI-at-Coinbase
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

RAPID-D makes the decision-making process transparent, consistent, and reproducible; Claude 3.7 Sonnet was selected after benchmarking across leading models for its quality, stability, and reliability.

What tools did this team use?

Claude 3.7 Sonnet, enterprise search tool.

What results were reported?

RAPID-D recommendation accuracy benchmark: benchmark scores across leading models; Model selection rationale: strong balance of quality, stability, and reliability (source-reported, not independently verified).

How is this back office ops AI workflow structured?

RAPID document submitted → Analyst agent baseline review → Seeker agent context retrieval → Contrarian agent challenges recommendation → Synthesizer produces final recommendation → Human Decider reviews AI output → Feedback captured and integrated.

Related back office ops cases