Custom Compute, Autonomous Payments, and the Work Behind Better Software

Who controls the agent stack?

OpenAI’s Jalapeño project puts custom compute on the agenda alongside model development. An interview about the internal chip effort covers design tradeoffs and the rationale for building more of its own compute rather than relying solely on merchant silicon. That choice could affect inference economics and demand for chip talent over time, though the project itself does not establish those outcomes.

At the transaction layer, Mastercard says its AI agents can pay for digital services through Coinbase’s x402 protocol, with on-chain payments routed through Cloudflare’s network and no human approval required for each transaction. That gives developers a concrete example of payment infrastructure for autonomous agents. Workflow vendors are also adjusting their positioning: n8n’s brand refresh emphasizes its role in an agent-driven world. The refresh explains how the company sees its automation product fitting into that work; it does not establish new orchestration capabilities.

Managed ingress still needs recovery goals

AWS announced Amazon ECS Gateway, a managed ingress solution for ECS services, bringing routing and load balancing into the ECS control plane. For teams already using ECS, the appeal is less separate configuration and patching of load balancers or custom ingress setups. Operational simplification still leaves the question of what adequate recovery means: a CIO.com article argues that growing infrastructure complexity makes that judgment harder. Its concern gives engineering leaders a reason to revisit recovery-time and recovery-point assumptions, and to test whether their exercises reflect current dependencies.

Useful features and broader experience design

Lightroom mobile’s Open Eyes Quick Action detects and opens closed eyes in a photo places a narrowly targeted AI edit inside an existing workflow, removing a trip to another editing tool. It is a concrete convenience with a clear task. The essay “You can’t hill climb your way to good experience” addresses a broader design problem: incrementally improving a single metric cannot substitute for rethinking the experience itself. A useful local improvement and an overall product direction require different kinds of judgment.

That judgment also matters to the people building the software. “The New Age of Software” reflects on changing skills and workflows as AI enters everyday development. It offers an opinionated perspective on the craft rather than technical guidance, complementing the product-design essay’s challenge to treating faster iteration as a sufficient measure of progress.

Geometry worth keeping close at hand

For engineers building canvas interfaces, tldraw’s guide to computing bounding boxes offers a more concrete reference: algorithms for rotated rectangles and curves, including edge cases and performance tradeoffs. Those details matter wherever hit-testing and layout depend on fast, accurate shape bounds, including graphics editors and CAD-like tools.