DXC built a multi-model AI substrate on Bedrock, cutting resolution time 95%
AWS ML Blog · rss · 2026-11-12
An AWS Blog case study co-written with DXC Technology (115,000 employees serving banks, hospitals, and governments) on building DXC OASIS, the governed multi-model substrate for mission-critical IT-ops agents:
- Two layers: Amazon Bedrock as the governed substrate (Claude, Amazon Nova, Llama, open-weight families via one API); DXC's own orchestration/governance harness on top
- Safety: Bedrock Guardrails enforce content filters, PII redaction, and denied topics uniformly; PrivateLink keeps inference inside the customer VPC; prompts never leave the AWS boundary or train models
- Resilience: a provider outage or deprecation is a routing decision, not a rebuild—behavioral differences are assurance-checked before rerouting
- Results: since GA in March 2026, average resolution time down 95% across 56 measured use cases, scaling toward 130 customer environments
Key thesis: the real risk isn't the model but an agent holding more authority than its task requires with nothing to stop it—governance must live at the infrastructure layer.
More from Companies & People
- Andrew Ng at Stanford: AI aces tasks with answers, lags badly on judgment calls — stanfordnlp · 2026-11-12
- Nathan Calvin's ant quip about Jensen Huang scandal becomes the viral take — ShakeelHashim · 2026-11-12
- NYT reports second affair by Nvidia CEO Jensen Huang involving a reporting executive — felpix_ · 2026-11-12
- Stanford NLP amplifies case for teaching agents by writing the raw loop from scratch — stanfordnlp · 2026-11-12
- Stanford's Agentic Development course praised as the only one covering fundamentals — stanfordnlp · 2026-11-12
- Epic's 2026 AI MegaJam partners include Claude, Google, OpenAI, NVIDIA and AMD — jfischoff · 2026-11-12