# This Week in Cloud: Major Updates That Make Building Easier
This week brought a wave of meaningful updates across cloud platforms that simplify how developers take their applications from concept to live production. Let’s walk through what happened.
—
## A New Way to Run Web Applications Without Rewrites
One of the biggest announcements this week centers on running standard HTTP-based web applications directly on serverless infrastructure. Previously, developers had to rewrite their web apps into event-driven handlers to work with serverless platforms. That’s no longer the case.
You can now take an existing web server built with popular frameworks — including Express, Hono, Fastify, or Next.js — deploy it directly, and have it live on a secure HTTPS URL in less than a minute. There’s no need for adapters, no infrastructure setup beforehand, and no commitment to any single framework.
Under the hood, this new offering supports up to 128 concurrent requests per execution environment. This means that during traffic spikes, warm environments are shared efficiently rather than spinning up a new environment for every single request. Billing is also more fair, as it tracks the actual CPU time your code uses instead of the total elapsed time — which is a big deal for applications that spend most of their time waiting, such as server-side rendering, API backends, or AI inference endpoints.
Deployment is streamlined too. Adding a single flag to your deploy command can replicate your code across multiple regions and set up latency-based routing behind one global URL. You can also route traffic between different versions of your app by weight, making it simple to test changes with a subset of users before a full rollout. The service is currently available across five regions, running the latest version of Node.js on energy-efficient Arm64 processors, with full support for virtual private cloud connectivity.
## Controlling AI Agents Without Losing Momentum
As artificial intelligence agents become more autonomous, a critical question emerges: how do you let them do more without losing oversight? A widely discussed piece this week argues that having a human approve every single action by an agent works well at small scale but breaks down as usage grows. People get fatigued and start missing important signals.
The proposed solution involves four checkpoints evaluated at different stages: scope (what the agent is allowed to do), know (what information it has access to), bound (what limits are hard-coded), and observe (what happens after an action is taken). These safeguards are applied outside the agent’s own reasoning, enforced at the system level so the agent itself cannot bypass or disable them. Major cloud providers have already used automated reasoning at enormous scale for years, processing billions of access requests per second. Applying that same proven approach to the rules governing AI agents is a logical and powerful step forward.
## Notable Product Updates and Launches
Several other product announcements caught attention this week:
– **Cloud storage is now directly accessible from popular development tools.** Developers can provision storage buckets, data tables, and vector databases directly from their favorite frontend platform, all within a single cloud account and a single bill. Authentication is handled securely through short-lived tokens that are exchanged at runtime, meaning no static credentials are stored anywhere in the codebase.
– **New cost-effective virtual machines are now available in South America.** These latest-generation burstable instances offer up to 30 percent better price performance compared to the previous generation, with significantly higher CPU output and network bandwidth. They use the same credits system as their predecessors, making migration straightforward. The smallest sizes are even included in the free tier for new users.
– **An open-source AI model designed for coding and long-running tasks is now available for deployment.** This sparse mixture-of-experts model has over 500 billion total parameters and supports a context window of up to 1 million tokens. It can be deployed for real-time use or scaled up for batch processing, and it has been optimized for specialized hardware accelerators designed for training and inference.
– **A software development kit for building resilient workflows in Go is now generally available.** Developers can now write multi-step processes that automatically save their progress and pause for extended periods — up to a full year — while waiting for external events. This is particularly useful for workflows that require approval gates or that call an AI model and then pause for human review. The toolkit integrates naturally with the Go programming experience, allowing developers to test full workflows locally without needing network access or cloud credentials.
– **Enhanced logging capabilities have arrived.** Users can now define automated routes that direct log entries to different storage locations based on conditions like severity level, source, or event type. Additionally, log entries can now be processed through a chain of parsing tools, where each parser in the sequence can use data produced by the one before it — for example, extracting a nested field from a JSON structure and then applying a pattern-matching tool to that specific field.
—
## Upcoming Events
Mark your calendars for the biggest gathering of cloud builders this year, returning to Las Vegas from the end of November through the first week of December. The event will feature over 2,200 sessions, and registration is now open. Additionally, community-organized conferences continue to expand globally, with new events planned throughout the year in regions around the world.
—
## Frequently Asked Questions
**Q: What is the benefit of running a standard HTTP server on serverless infrastructure?**
A: It eliminates the need to rewrite your application into event-driven code. You deploy your existing web app and get a live URL quickly, with built-in scaling and pay-per-use billing.
**Q: How are AI agents kept under control if they are given more autonomy?**
A: Controls are applied at the system level outside the agent’s reasoning. This means the agent cannot override or disable its own guardrails, similar to how access policies work in identity management systems.
**Q: Are the new virtual machines in South America suitable for production workloads?**
A: Yes, they are designed for a wide range of workloads including login gateways, small databases, CI/CD pipelines, and low-traffic websites. They also offer a significant price-to-performance improvement over previous generations.
**Q: Can the open-source coding AI model be used for real-time applications?**
A: Yes, it can be deployed through real-time endpoints for interactive use, or scaled using container-based infrastructure for higher throughput and Kubernetes-native management.
**Q: What happens if my Go workflow needs to wait for a user to approve something?**
A: The Durable Execution SDK allows your workflow to checkpoint its progress and pause for up to a year, resuming exactly where it left off once the external event occurs.
—
## Conclusion
This week’s updates reflect a clear trend in cloud computing: making it easier, faster, and more secure to move from development to production. Whether it’s removing friction from web deployments, introducing smarter controls for AI agents, or expanding global infrastructure with better pricing, these changes empower developers to focus more on building and less on managing complexity. As these platforms continue to evolve, the barrier to creating robust, scalable applications continues to drop.
Thank you for reading
