Blog
Amazon Bedrock, AgentCore, SageMaker & Q: Agents in Microsoft Teams
FDE playbook: which AWS service to use, first Teams agent path, and calendar estimates for RAG, MLOps, and LLM apps
On this page
How Amazon Bedrock, Bedrock AgentCore, SageMaker, and Amazon Q fit together — and how an FDE ships a first business agent into Microsoft Teams. For the full setup guide, tooling list, and calendar estimates, read the long article.
Who does what in the AWS stack
| Service | Best for | Not for |
|---|---|---|
| Amazon Bedrock | Serverless FMs, Guardrails, Knowledge Bases (RAG), Agents APIs | Training large models from scratch |
| Bedrock AgentCore | Custom agent runtime, Gateway tools (incl. Teams Graph), Identity, Memory | Low-code “chat with my docs” only |
| Amazon SageMaker | Train / fine-tune / host custom models, Pipelines, Feature Store, MLOps | Fastest path to a Teams chatbot MVP |
| Amazon Q | Employee knowledge assistant; developer coding help on AWS | Deeply custom multi-agent products |
Recommended business pattern
- Week 0–2: Stand up Bedrock + Knowledge Base RAG on a pilot corpus; enable Guardrails.
- Week 2–4: Package a production agent on AgentCore Runtime; connect tools via Gateway (SharePoint, CRM, Teams Graph).
- Week 3–5: Ship a Teams bot that relays messages to AgentCore; keep sessions with
teams-{aadObjectId}-{conversationId}. - Parallel / later: SageMaker for custom classifiers or fine-tunes; Amazon Q Business for org-wide Q&A with less build.
First agent in Teams (FDE snapshot)
Minimum path that works in production-minded pilots:
- Azure Bot + Teams app registration (Entra ID)
- AWS Lambda (or container) as Bot messaging endpoint
POSTto AgentCore Runtime/invocationswith JWT- Bedrock Knowledge Base for RAG; Guardrails for safety
- Pilot channel, human feedback loop, CloudWatch traces
Expect roughly 3–6 weeks for a scoped internal agent (one department, one knowledge source, read-mostly tools) with an FDE + platform engineer + security review. Full MLOps + multi-agent + write-actions into line-of-business systems is a quarter+.
Time boxes (indicative)
| Workstream | Calendar (FDE-led) |
|---|---|
| Bedrock chat + Guardrails MVP | 3–5 days |
| Knowledge Base RAG (one corpus) | 1–2 weeks |
| AgentCore agent + 2–3 tools | 2–3 weeks |
| Teams bot + admin publish pilot | 1–2 weeks (overlap) |
| Amazon Q Business (managed assistant) | 1–3 weeks |
| SageMaker fine-tune + endpoint + MLOps | 4–8+ weeks |
Read the long version
The long article covers architecture diagrams in prose, step-by-step Teams first-agent setup, Entra ID with AgentCore Identity, tool Gateway targets, FDE training curriculum, requirements/tooling checklists, and phased estimates for LLM apps, RAG, and MLOps. Published by TenthMatrix.
More from the blog
Observability Is Not Optional: Make It a Business Standard
Why fragmented monitoring creates blind spots, duplicate spend, and slow incidents — and how to make centralised observability a non-negotiable company standard.
WSL Vault for Compliance: One Tenant Never Opens Another’s Secrets
How WSL Vault keeps app, team, and environment tenants cryptographically isolated — so a stolen root-style token cannot empty the estate — and why dual-tenant control beats secrets sprawl.
Workstation AI Shop: Promoting the AI Sales Storefront Through int → test → acc → prod
Workstation AI Shop sells GPU machines with a fully autonomous AI Sales Agent that configures the basket. See how Ring Promoter moves each build through int → test → acc → prod — with Orbit screenshots release engineers and stakeholders share.