A Door Closed After 2 Years and 8 Months
Amazon Bedrock Agents, launched in November 2023, was renamed Bedrock Agents Classic on July 30, 2026 and stopped accepting new customers as of that date. Existing customers keep working as before, but the service entered maintenance mode and its model catalog froze as of the same effective date. Any model released after July 30 is reachable only through the successor service, AgentCore — not through Classic.
AgentCore Is Not "Bedrock Agents 2"
Classic was a managed-resource model: you configured an agent in the console, AWS owned the orchestration loop, and your code only existed at the edges, inside action-group Lambdas. AgentCore instead composes Runtime, Gateway, Memory, Identity, Observability, Policy, Evaluations, Browser, and Code Interpreter as independent primitives, which flips the orchestration loop into a program the developer owns outright. It also plugs into frameworks like LangGraph and Strands — another point where it departs from Classic.
From Design to Operations: A Classic-to-AgentCore Migration Checklist
Anchor planning on the catalog freeze date (July 30) and work backward to a migration deadline, so no agent is left stranded without access to new models. Start by fixing a table of agents to migrate with a deadline for each. AgentCore Runtime bills $0.0895 per vCPU-hour and $0.00945 per GB-hour, and idle time spent waiting on a model response is not billed — so calculate the expected cost delta against Classic's simpler agent-hour billing before you commit to a timeline.
Set two minimum operational targets: a 24-hour canary observation window after each cutover, and a 100% trace-field migration rate — Classic's one-click console trace has to be fully replaced by CloudWatch metrics before that agent goes live. Cutting traffic over before both numbers are met leaves a gap where incident root-causing falls back from one console click to assembling a CloudWatch dashboard from scratch.
The first common failure pattern is treating AgentCore as a straightforward Classic upgrade. Because ownership of the orchestration loop itself moves from AWS to the developer, simply relocating an action-group Lambda does not work. The second is mistaking the output of the AgentCore CLI's import-agent command for a finished migration — it is only a starting point that carries a Classic configuration into a LangChain, LangGraph, or Strands framework, and any feature with no validated harness path is halted with a suggested alternative rather than silently migrated.
Build the recovery branch in two stages. When import-agent stalls on a specific feature, quarantine only that feature for manual reimplementation and keep the rest of the migration moving. If an incident surfaces during canary observation, the Classic agent still runs normally for existing customers even in maintenance mode, so route traffic back immediately and diagnose the AgentCore-side cause separately. Make explicit to the team that a rollback to a catalog-frozen Classic is a stopgap, not an end state.
For the operations checklist, standardize the migration environment on either Node.js 20+ with the AgentCore CLI, or Python 3.10+ with boto3, and pull Gateway and Memory's billing units — MCP operation call counts, memory record create/retrieve counts — into a single dashboard so they don't blur into Classic's flat agent-hour billing. Revisit PII-masking rules against CloudWatch's log fields as well.
Run the improvement loop by tallying migrated agents against remaining Classic agents every week. Since the catalog is already frozen, any agent still on Classic keeps falling further behind on model access, so use that widening gap as your migration-priority signal. Keep a running list of features import-agent has halted on — it cuts the re-validation time the next migration batch spends on the same feature.
Quick-Reference Checklist
Bedrock Agents Classic stopped taking new accounts on July 30, and its model catalog froze the same day. Accept that AgentCore's orchestration model is a genuine shift from managed resource to owned program, treat import-agent as a starting point only, track Runtime, Gateway, and Memory billing units separately, and keep Classic as nothing more than a temporary rollback path — in that order, migration failures drop.
References
Amazon Bedrock Agents Classic maintenance mode — AWS official docs