# Superblocks 3.0 Brings Vibe Coding Into the Customer’s AWS Account *The app builder moves code, data, databases, and model inference behind existing cloud controls, while remaining a managed service.* By [Tech Tab](https://paragraph.com/@tech-tab) · 2026-08-04 developer tools, cloud computing, artificial intelligence --- AI app builders make it easy to turn a prompt into an internal dashboard, approval tool, or workflow. They also create an uncomfortable question for IT teams. Where do the generated code, business data, database, model prompts, and application logs actually live? Superblocks 3.0 offers a concrete answer for companies already committed to Amazon Web Services. Its new Cloud-Prem deployment places the platform inside the customer’s AWS account. Applications use the company’s network boundaries, identity rules, logging, and approved models instead of sending every part of the project through a separate consumer app-building cloud. That is more specific than a generic “private AI” label. It changes the operating boundary of the product, but it does not make Superblocks self-hosted software that the customer must maintain alone. What the product builds ----------------------- Superblocks is designed for internal business applications rather than public consumer products. A user describes an app in natural language, and its Clark agent generates the interface, workflows, and connections to databases or APIs. The AWS Marketplace listing says the generated applications use React and TypeScript, can be deployed with Git, and remain available for engineers to inspect and extend. The intended users include analysts, operations teams, and other employees who understand a business process but may not build a full application from scratch. IT configures shared requirements such as single sign-on, role-based access, secrets management, approved integrations, audit logs, and design systems. New applications inherit those controls. Superblocks 3.0 also imports prototypes from tools including ChatGPT, Claude, Lovable, and Replit. This creates a path for an experimental app to enter a managed development process instead of remaining an untracked link or a collection of scripts on an employee’s laptop. What “inside your AWS account” means ------------------------------------ In the Cloud-Prem model, Superblocks runs as a dedicated single-tenant deployment in the customer’s AWS account. The control plane and data plane can share one region, or the control plane can operate in one region while local data planes sit closer to data in other regions. Communication from those data planes is outbound-only, so customers do not need to expose inbound firewall ports. When an app needs storage, the platform can provision Amazon Aurora or S3 resources inside that account. Moving an application from development to production can trigger database migrations across the company’s environments. AI inference runs through Amazon Bedrock using models and regions allowed by the organization. This arrangement gives security teams familiar levers. AWS Identity and Access Management policies govern identity and permissions. VPC controls define network access. Existing encryption, monitoring, and audit systems can observe the platform. Prompts, model outputs, application data, and runtime components remain within the selected AWS boundary, according to the product documentation. The word “customer-owned” still needs care. Superblocks continues to operate the platform as a managed service. It handles upgrades, patches, reliability, and incident support under least-privilege access and customer-defined boundaries. The customer controls the cloud environment and policies, but remains dependent on Superblocks for the product lifecycle. Security is a pipeline, not a badge ----------------------------------- Running inside a VPC reduces some data movement. It does not prove that AI-generated code is correct or secure. Superblocks addresses that second problem with both deterministic scanners and specialized security agents before deployment. Static analysis looks for known patterns such as SQL injection, embedded secrets, and unsafe data flows. The agent-based layer examines application context, including authentication, authorization, APIs, and business logic. Organizations can add policy agents to check their own rules. In production, the platform records a software bill of materials and continues scanning dependencies for newly disclosed vulnerabilities. It can also connect to a private package registry and block public NPM access at the network layer. That matters because a generated application can be perfectly contained and still import a compromised dependency. Controlling where packages come from is different from checking the code the model wrote, and the product treats them as separate controls. These protections should be evaluated rather than assumed. Superblocks describes the security agents’ capabilities, but has not published a public benchmark showing their detection rate or false-positive rate. Human review, testing, and conventional change controls remain necessary for applications that move money, modify sensitive records, or enforce important decisions. Model routing adds another trade-off ------------------------------------ The platform’s Smart Router divides a build into tasks and sends routine coding work to lower-cost open models while reserving frontier models for planning and difficult reasoning. AWS and Superblocks claim this can reduce model costs by as much as 30 percent. That figure is a vendor estimate, and actual savings will depend on the workload, chosen models, and Bedrock pricing. The useful part is administrative model choice. A company can approve several models through Bedrock and change the mix without rebuilding its application platform around one AI provider. The trade-off is deeper dependence on AWS services, including Bedrock, Aurora, IAM, and the VPC architecture itself. Pricing is also not public. The AWS Marketplace lists contract and usage-based charges but directs buyers to request terms. Infrastructure consumption sits alongside the Superblocks contract, so evaluating the tool requires counting both platform fees and the AWS resources it creates. Superblocks 3.0 is therefore less about making prompts more impressive than making the resulting software governable. It gives business teams a faster builder while placing the runtime, data, inference, and audit trail where an AWS security team already works. That does not remove software risk. It makes the risk visible in systems the company can inspect and control. Sources ------- * [Superblocks 3.0 announcement](https://www.superblocks.com/blog/superblocks-3-0-secure-private-vibe-coding-on-aws-built-for-the-age-of-ai-cyberattacks) * [Superblocks Cloud-Prem architecture documentation](https://docs.superblocks.com/enterprise/cloud-prem/aws) * [AWS collaboration announcement](https://press.aboutamazon.com/aws/2026/8/superblocks-and-aws-announce-strategic-collaboration-to-bring-secure-enterprise-ai-app-development-to-amazon-bedrock) * [AWS Marketplace listing](https://aws.amazon.com/marketplace/pp/prodview-kllccta3zgs2q) * [TechCrunch: AWS brings Superblocks into customer private clouds](https://techcrunch.com/2026/08/03/aws-is-helping-vibe-coding-startup-superblocks-and-the-implications-are-big/) * [Specification-driven development paper](https://arxiv.org/abs/2607.16680) --- *Originally published on [Tech Tab](https://paragraph.com/@tech-tab/superblocks-3-brings-vibe-coding-into-customer-aws-account)*