AWS This Week: Flat-Rate Direct Connect, PrivateLink Tunnels and 2-Second Agent Cold Starts (Video)

The week’s AWS announcements that actually matter for architects — in 79 seconds, with the “so what” for your architecture. This week the theme is network economics and faster agents. Watch first, details below.

This week’s picks

1. AWS PrivateLink Tunnel Endpoints — a new type of VPC endpoint: you describe a CIDR range of your network as a single Resource Configuration, share it through AWS Resource Access Manager, and the other party creates a tunnel endpoint that reaches into that segment over GENEVE encapsulation. Until now, sharing resources with an external vendor meant one Resource Configuration per resource; now it is one per network segment. Architect’s note: this is the pattern for managed-service providers and appliance vendors who need to reach many private resources — price it first, because there is an hourly charge per tunnel endpoint plus a per-GB charge for data processed. Available in 28 Regions at launch, including Paris, Frankfurt, Ireland, London, Stockholm and Zurich. (announcement)

2. AWS Direct Connect flat-rate pricing for 10G and 100G dedicated connections — a single fixed monthly price that eliminates data-transfer-out charges within the selected tier, with five geographic tiers from same-metro to global. The structural change for FinOps: DTO stops being the line item that grows with your traffic. The structural change for resilience: the new port-pair provisions two dedicated connections on different devices or locations sharing the same bandwidth and scope, and the second connection is included at no additional charge. Dedicated connections only, all commercial Regions except China, and you can switch billing mode per connection at any time. (announcement, flat-rate user guide)

3. The new AgentCore Runtime is generally available — the next generation of the serverless microVM compute in Amazon Bedrock AgentCore. Two mechanisms matter: the agent environment is prepared once and snapshotted, so every new instance restores from that snapshot (AWS reports a P75 cold start of 1.9–2.0 s for container images from 200 MB to 2 GB, versus 5.4–30 s on V1), and memory is allocated on demand and reclaimed during the session, so you pay for what the workload actually touches rather than its peak. Image size stops being a cold-start tax. Opt in by setting platformVersion to V2 when creating or updating a runtime; available in us-east-1, us-east-2, us-west-2, eu-west-1 and ap-northeast-1. (announcement, deep dive on the ML blog)

4. Amazon EC2 T8i burstable instances are generally available — custom sixth-generation Intel Xeon (Granite Rapids) processors, up to 30% better price-performance and up to 70% higher compute performance than T3, four sizes from t8i.nano (2 vCPU, 0.5 GiB) to t8i.medium (2 vCPU, 4 GiB), same CPU-credit model with Standard and Unlimited modes. The architect’s takeaway is the migration cost: zero. Dev and staging fleets, low-traffic services and login gateways move by changing the instance type. (announcement)

Industry signal — Claude Code now reads AGENTS.md when a folder has no CLAUDE.md (from version 2.1.277, toggleable with /config). AGENTS.md is the tool-agnostic instruction format OpenAI contributed to the Agentic AI Foundation under the Linux Foundation, already used by more than 60,000 open-source projects at the end of 2025. For teams running several coding agents on the same repository, the CLAUDE.md/AGENTS.md symlink workaround can go: one instruction file, read by every agent. (coverage)

Which one changes your roadmap? The video also runs on LinkedIn — previous edition: September 7. If the new Runtime caught your eye, see how five agents with per-role models run on AgentCore.