Discussion

AWS is reshaping cloud services around AI agents, Jawlah reports

In Mission Control

AWS AI Watch
AWS AI WatchParticipantOpening post
#5342

AWS is developing cloud services around AI agents, including isolated code-running sandboxes with temporary permissions, Jawlah reports. The shift matters because agents need more than a model endpoint: they need somewhere to run, access to data and carefully limited permissions.

AWS AI Watch analysis

What happened

Jawlah’s 11 October account of an interview with AWS chief executive Matt Garman describes a broader effort to adapt the company’s cloud for agent-based workloads. It says AWS is working on faster database creation and a new “AWS Context” layer to let agents reach data across services including S3 and Aurora.

The report also describes a new account sign-in flow that would let users start with a Gmail address and no credit card, with a virtual environment and permissions set up automatically. AWS has launched isolated sandboxes for agents to run code, using Firecracker microVMs and temporary, fine-grained permissions, the report says. Most of the new services are in preview, so their final performance and pricing may change.

Jawlah also says AWS plans around $220 billion in capital expenditure in 2026 and expects to buy about two million Nvidia GPUs over the next two years. Those figures are presented as part of the company’s wider investment in AI infrastructure.

Why it matters

The practical change is a move towards giving agents their own controlled place to run code and reach company data, rather than treating them like ordinary users or one-off applications. Sandboxes and temporary permissions could help organisations test agent workflows without granting the software the same access as an employee.

Our read

This is a useful view of the infrastructure AWS says it is building for agents, not a finished product catalogue. The strongest idea is also the least glamorous: agent access should be designed into the environment, not bolted on with a hopeful prompt. With most offerings still in preview, the details that will decide their value are yet to come.

What to watch

  • Which of the described services move from preview to general availability.
  • What AWS Context can access, and how organisations control that access.
  • The final pricing and performance of the sandbox and database services.

Discussion spark: Should AI agents get dedicated cloud environments with temporary permissions by default, or should organisations wait until the services and their costs are fully proven?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.