Amazon upgraded the Bedrock AgentCore harness to build on its open-source Strands harness, giving managed agents recommended models, task-oriented system prompts, automatic prompt caching, and proactive context management by default. It also adds built-in tools for planning, web retrieval, filesystem access, context offloading, and task delegation, plus lifecycle hooks that let developers inject deterministic checks and notifications at defined points in the agent loop.
Prompt Caching
提示词缓存
Reuses computed state of a matching prompt prefix to avoid reprocessing long context, cutting latency and cost.
Curated Intelligence
Life-sciences SaaS vendor ArisGlobal runs its NavaX engine on Amazon Bedrock to automate pharmacovigilance adverse-event case processing. As multi-tenant compliance requirements pushed the shared prompt prefix up roughly 150% in six months and per-case inference costs up over 200%, the team reversed its caching strategy to cache the shared prefix instead of each source document, cutting effective input cost and enabling tenant growth without proportional spend.
Amazon Bedrock now supports explicit prompt caching for open-weight models, letting applications reuse repeated prefixes such as system prompts, reference documents, tool definitions and conversation history to cut input costs and latency. The cache is fully managed by Bedrock, offers implicit and explicit modes, and exposes cache read/write metrics for monitoring.
Baoyu outlines a layered routing pattern for multi-agent sessions: the main session only classifies and dispatches tasks, so a mid-tier model like Sonnet 5.5 suffices, while stronger models such as Opus 5.5 handle work inside threads. Paired with differentiated prompt-caching tactics, sending only recent messages plus session retrieval after cache expiry and compressing thread context near cache TTL, overall token consumption stays manageable.
AWS's official blog notes that even after prompt and application tuning, four request-level settings—cache duration, service tier, reasoning effort, and maxTokens—still materially change Amazon Bedrock's token volume and billing rates. It documents how each setting was tested and how to verify impact via response metadata, logs, and billing data.
AWS published the second part of its Bedrock cost series, arguing that agent economics should be measured by cost per task rather than cost per token, with caching, model routing and output constraints as the main levers. It also outlines a discipline for proving agent ROI through human baselines, holdout experiments and a five-dimension scorecard, aligned with the Well-Architected Agentic AI Lens.
Anthropic has introduced a beta feature to the Claude API that lets developers adjust effort mid-conversation by inserting an empty system message with output_config.effort. The setting applies from the next user message onward without altering already-generated content, and prompt caching still hits, helping balance quality, latency and cost when switching between casual chat and complex analysis.
Amazon Bedrock introduces in-country inference for the open-weight multimodal model Kimi K3 in India to meet strict data residency requirements. This allows customers to process sensitive local workloads while integrating frontier coding and agentic capabilities via existing APIs.
Anthropic has launched a beta for 'Message Threads' in the Claude Messages API, allowing the server to cache conversation context so developers no longer need to resend full history each turn. This feature significantly reduces request payload size (from ~800KB to 1.8KB) and cuts time-to-first-token by ~35%, though token pricing remains unchanged. It is particularly beneficial for long-running agents and multimodal interactions.
Arena.ai analyzed over 26K agent sessions, revealing that removing prompt caching increases typical task costs by 260%. For models like Claude Fable 5.1, a single task's cost jumps from $7.44 to $27 without caching, while cache invalidation adds another 5% to bills. The study highlights the critical role of caching in optimizing agent economics.