In the week of 8 to 14 June 2026, AWS renamed its rehost engine to AWS Transform MGN, which tells you where the migration product line is going: the agent is the product, and the replication service is a part it calls. The same week delivered a cluster of cost tooling that reads your bill for you (the FinOps Agent in preview, AI-powered anomaly investigations, and intelligent cost explanations), the first Graviton5 instances for general-purpose work, and a maintenance notice for AWS CLI v1 that anyone with a decade of shell scripts should read twice.

Also this week: Anthropic's Claude Fable 5 launched on Bedrock on 9 June and had its access revoked on 12 June under a US export directive. That story is told at ercan.ai; the infrastructure lesson is the usual one: do not pin production to a single upstream you do not control.

AWS Transform MGN: the rename that is a roadmap

AWS Application Migration Service is now AWS Transform MGN, dated 8 June. The announcement is explicit that it is a name change reflecting MGN's role as "the proven replication engine powering AWS Transform", lists no API, pricing, or feature changes, and keeps the compliance certifications (FedRAMP High, HIPAA, PCI DSS, ISO, SOC 1, 2, 3). It is available in all commercial Regions and both GovCloud Regions.

You now get two rehosting experiences. The AWS Transform MGN console gives direct control over replication and cutover, the way MGN always did. The AWS Transform agentic workflow has an agent do discovery, wave planning, landing zone setup, network creation, and the rehost or containerization itself, calling MGN underneath. Read it as a statement of intent: the console path is for people who know what they are doing and want the knobs; the agent path is where AWS expects volume to go.

On the same date, AWS Transform added a migration cost assessment for RDS for SQL Server. Feed it RVTools exports, CMDB data, or output from the Transform discovery tool, and it recommends instance configurations, models Bring Your Own Media against License Included (BYOM shipped on RDS last week), prices in Database Savings Plans (up to 20% off On-Demand), and flags MAP eligibility. What-if scenarios let you vary Region, utilization, and term. The trade-off with any TCO tool: the output is as honest as the utilization data you fed it. RVTools peak numbers produce a conservative bill; 95th percentile numbers produce the one finance will hold you to.

Practical note: the announcement says nothing about IAM action prefixes, SDK namespaces, or Terraform resource types, so assume mgn stays put until a changelog says otherwise.

Three agents for one bill

Cost management got more agentic attention in one week than in the prior year. AWS FinOps Agent entered preview on 9 June in US East (N. Virginia), at no charge during the preview. It answers cost questions, surfaces rightsizing, idle resource, and Savings Plans recommendations from Cost Optimization Hub and Compute Optimizer, investigates anomalies, runs scheduled FinOps workflows, opens Jira tickets, and posts findings to Slack. Its data covers all Regions except GovCloud and China.

AI-powered cost investigations (8 June) sit inside Cost Anomaly Detection: choose "Investigate with Amazon Q" on a detected anomaly and it determines whether the change is usage-driven or rate-driven, names the contributing services, accounts, and Regions, and for usage changes correlates against CloudTrail to attribute the spike to specific API calls and IAM principals. With an organization trail it covers member accounts. No extra charge, except that the cross-account path reads an organization trail delivered to CloudWatch Logs, so Logs Insights scan charges apply. Cost Explorer intelligent cost explanations (9 June) and target coverage analysis in the Savings Plans Purchase Analyzer (8 June) round it out.

If you are going to let an agent near production, cost is the right first domain: the data is read-only, the blast radius of a wrong answer is a wrong Jira ticket, and the alternative is a human doing the same CloudTrail correlation at 11pm after an anomaly alert. The failure mode is noise. An agent that opens a ticket for every anomaly trains the team to close tickets without reading them within a month. Scope the schedules and thresholds before you turn on the integrations, the same way you would scope a budget alert.

Graviton5 arrives in general purpose

M9g and M9gd instances went generally available on 10 June. AWS claims up to 25% better compute performance than Graviton4 overall, up to 35% for web applications and ML inference, up to 30% for databases, and a 5x larger L3 cache. The largest sizes reach 192 vCPUs and 768 GiB; M9gd carries up to 11.4 TB of local NVMe. Available in US East (N. Virginia), US East (Ohio), US West (Oregon), and Europe (Frankfurt).

Four Regions is the constraint. If your Savings Plans are EC2 Instance plans locked to M8g in a Region not on that list, there is nothing to do yet. If they are Compute Savings Plans, M9g is a drop-in experiment: same ARM64 binaries, one launch template change, and a week of CloudWatch to see whether the 25% shows up for your workload or only for the benchmark.

AWS CLI v1 goes into maintenance on 15 July

The AWS CLI team announced on 10 June that CLI v1 enters maintenance mode on 15 July 2026. From then, v1 bundles its own copies of botocore and s3transfer instead of depending on the standalone packages; upgrading the CLI no longer upgrades them, and installing them no longer changes what the CLI uses. New v1 releases are limited to critical bug fixes and security issues.

The subtle break is environments where CLI v1 and boto3 share a Python installation. After the change there are two copies of botocore in that environment, and any script that assumed "pip install awscli" kept boto3's dependencies current is wrong. Pin boto3's dependencies explicitly, then plan the move to v2. The v1 versus v2 differences are mostly output and pagination defaults, which is exactly the kind of thing that breaks a cron job silently.

Also this week

The throughline

Every headline this week is AWS moving a human task into a service: wave planning into an agent, anomaly investigation into Q, TCO modelling into Transform. The engines did not change; what changed is who reads the output. Decide deliberately which of those readers you trust, and keep the console path warm for the day the agent is wrong about your estate.

Read this next

For the model and agent side of the same week, including the Fable 5 interruption, read ercan.ai. For consulting on AWS, cloud, CI/CD pipelines, Linux, GitHub/GitLab, and more, the hub is ercanermis.com.

References