AIToday
Large Language ModelsAI Business & IndustryAmazon AI BlogPublished: Nov 12, 2026, 06:00 JST

DXC OASIS on Amazon Bedrock cuts resolution time 95%

Send on LINE
DXC OASIS on Amazon Bedrock cuts resolution time 95%

3 Key Points

  1. What happened

    DXC Technology built DXC OASIS on Amazon Bedrock, separating a governed multi-model layer from its own orchestration harness; across 56 measured use cases since March 2026, resolution time fell 95% on average.

  2. Why it matters

    Regulated operators of production systems get a template for bounded, auditable agent authority—every persona starts read-only, and the harness, not the model, records who approved what.

WHO IT HITSEnterprise IT and platform teams running production systems in banks, hospitals, and government agencies—plus their CIOs and risk officers—get a concrete model for separating model access from agent governance.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

DXC Technology's starting point is a specific failure mode: an autonomous agent holding more authority than its task requires, with nothing in the architecture to stop it. The company's answer was to split the problem in two. Amazon Bedrock supplies a governed gateway to models—Claude, Nova, Llama, and open-weight families—with Guardrails, PrivateLink, and serverless scaling applied uniformly. On top sits DXC's own harness, carrying 60 years of runbooks, escalation logic, and ITSM data, plus the policy enforcement and human-oversight gates that Bedrock does not provide.

The division of labor is deliberate. DXC's team lists four design goals—resilience, flexibility, regulatory sovereignty, and safety by design—and argues no single control delivers any of them. Serverless infrastructure, model-layer guardrails, private networking, and approval gates compound into defense in depth. The harness also owns assurance: OpenTelemetry traces for every action, each with a unique execution ID, so auditability comes from DXC's layer rather than the agent's own account of what it did.

DXC began the build as a prototype in early 2025, reached general availability in March 2026, and is scaling toward 130 customer environments on Amazon EKS with per-tenant namespaces. Beyond the 95% average reduction in resolution time, diagnostic accuracy is 92% or better against senior engineer judgment. The company is now evaluating Amazon Bedrock AgentCore alongside its current orchestration layer, which suggests the substrate-harness boundary may be redrawn as AWS's own agent tooling matures.

FAQ
Which AI models does DXC OASIS use?
All models are reached through Amazon Bedrock model inference: Anthropic Claude, Amazon Nova, Meta Llama, and open-weight families. Because they sit behind one API, a provider outage or deprecation becomes a routing decision rather than a system rebuild.
How does DXC keep agents from taking dangerous actions?
Every persona starts read-only, and authority is granted in four tiers: observe, propose, act with approval, and act within scope. In the incident-resolution workflow, the agent can diagnose freely but cannot change a production server without an explicit, recorded approval.
Does customer data leave the AWS boundary?
No. Prompts and completions are not shared with third-party model providers and are not used to train the underlying foundation models. Inference traffic stays inside the customer's VPC via AWS PrivateLink.
Amazon AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleLgram joins Claude connector list as Community app