scalable/ai/agentcore cph / /
8 lessons · ~109 min

agentcore

eight lessons for engineers new to aws bedrock agentcore. everything runs in your browser against a mocked typescript sdk that mirrors the real one — write the same shape of code you'd deploy, without an aws account.

written forbedrock-agentcore 0.2.2 (typescript)last updatedsdk moves fast — verify against current aws docs before you ship.
agent invoke() yield done runtime events memory context gateway tools identity auth observability traces code interpreter code browser web

what is agentcore

aws bedrock agentcore is a managed runtime for production ai agents on aws — framework-agnostic (langgraph, strands, the vercel ai sdk all work), model-agnostic (any bedrock-resident model plus openai and anthropic via api keys), priced per invocation, with cold starts in the 3–8 second range. it ships seven primitives — runtime, memory, gateway, identity, browser, code interpreter, observability — that compose into the parts of an agent that normally take a quarter to build yourself. this course covers all of them, with running code.

who this is for

engineers who already ship typescript and now need to ship agents in production. you should be comfortable with async/await, http, and at least one llm api. you do not need an aws account, access to bedrock, or any prior agentcore experience — every lesson runs against a browser-side mock of the sdk that mirrors the real public surface. by the end of the course you'll be able to read a production agentcore codebase, reason about which primitive owns which problem, and write the deploy.

why mock, not real

cold starts and bedrock pricing make a real-aws teaching loop slow and expensive. the mock removes both. the deliberate constraint: every public function on the mock has the same signature as its real counterpart, so the typescript you write in lesson 02 is the typescript you'd commit to a production repo. lesson 08 walks the diff explicitly — what changes when you swap bedrock-agentcore-mock for the real package, what doesn't, and the iam and region setup you'll need on real aws.

how it works

lessons

  1. 01 what is agentcore the seven primitives, the boring parts aws decided to own, and the shape of a minimal agent. 10m
  2. 02 runtime — your first agent define an entrypoint, handle a payload, emit events. the agent app object, up close. 15m
  3. 03 memory — state that survives the turn session memory vs long-term memory. events, strategies, retrieval across sessions. 14m
  4. 04 gateway — turning apis into tools take an openapi spec (or a lambda) and make it callable from your agent as an mcp tool. 16m
  5. 05 identity — inbound and outbound auth workload identity, oauth to third parties, iam without writing policy by hand. 14m
  6. 06 browser & code interpreter two managed tools you get for free — when to reach for each. 12m
  7. 07 observability — reading the trace spans, tool-call boundaries, cost attribution. the parts cloudwatch doesn't show you. 10m
  8. 08 shipping to real aws what changes when you leave the sandbox. cli, iam, costs, and the smallest deploy that works. 18m
scalable labs·cvr 30091604·github·linkedin·hello@scalable.dk