Living Tracker
The Agentic AWS Landscape, Refreshed.
Most consulting firms publish a quarterly PDF of trends. We publish a page that refreshes. AWS ships agentic capability. The landscape changes. Our delivery learns. This page is updated in cycle so executives can read a current view rather than a stale one.
How to Read
The structure.
The tracker has three columns. The first column lists current AWS agentic capability : services, models, frameworks, partner programs : by category. The second column states the capability as AWS describes it. The third column is Mactores' view: what we are seeing in delivery, what we recommend, what we would not commit to yet. The third column is what makes this page worth keeping open.
The Tracker
Current agentic AWS landscape.
Foundation Models on Bedrock
Capability
Claude 3.5 Sonnet
AWS Description
Anthropic's flagship model for complex reasoning, analysis, and long-context tasks
Mactores View
Our default for agentic orchestration. Strong reasoning, reliable tool use. Production-ready.
Capability
Amazon Titan Text
AWS Description
First-party embeddings and text generation optimized for enterprise
Mactores View
Solid for retrieval embeddings. We use Titan for embedding layers in production RAG pipelines.
Capability
Llama 3.1 (405B)
AWS Description
Meta's open-weight model available on Bedrock
Mactores View
Strong alternative when customers want open weights. Heavier inference cost than Claude for comparable quality.
Bedrock Agents
Capability
Bedrock Agents
AWS Description
Managed agent orchestration with action groups and knowledge bases
Mactores View
Good for single-agent patterns. Multi-agent orchestration still requires custom scaffolding in most production scenarios.
Capability
Bedrock Guardrails
AWS Description
Policy enforcement for AI behavior, content filtering, and safety
Mactores View
Essential for regulated industries. We bake guardrails into every HIPAA and FSI deployment.
Capability
Bedrock Knowledge Bases
AWS Description
Managed RAG with automatic chunking, embedding, and retrieval
Mactores View
Fast path to RAG. Production tuning usually requires custom retrieval for precision.
Data & Retrieval
Capability
Amazon OpenSearch (with k-NN)
AWS Description
Hybrid search with vector and keyword capabilities
Mactores View
Our go-to for hybrid retrieval. Vector + BM25 with cross-encoder re-ranking is the production pattern.
Capability
Aurora Serverless v2
AWS Description
Serverless PostgreSQL with pgvector support
Mactores View
Strong for applications needing both relational data and vector search in one store.
Capability
Amazon S3 + Textract
AWS Description
Document storage and extraction pipeline
Mactores View
Standard pattern for document-heavy agentic apps. Textract quality is production-grade for most legal/financial docs.
Orchestration & Compute
Capability
AWS Step Functions
AWS Description
Serverless workflow orchestration with visual debugging
Mactores View
Our default for multi-step agentic workflows. Audit trail by construction.
Capability
Amazon EKS
AWS Description
Managed Kubernetes for containerized workloads
Mactores View
Required for EMR on EKS migrations. Also our pattern for high-throughput inference serving.
Capability
AWS Lambda
AWS Description
Serverless compute for event-driven functions
Mactores View
Glue layer for agentic architectures. Cold start latency matters less than people think for most use cases.
Want the Mactores view on your specific stack?
This tracker covers the general agentic landscape. Your architecture has specific constraints. Let's talk about what applies to you.
Start a conversation →