Define the infrastructure beyond a prototype

Start with the generated system

Consider a fictional prototype platform. It creates a web container, a managed PostgreSQL database, and a public URL. The workflow runs correctly with example records. This is a useful result: people can assess the feature before funding a larger implementation.

Now the company wants to store confidential contracts and use its employee identity provider. The required system has changed. A successful container deployment does not establish authorization, approved data handling, recoverability, or service ownership.

Different development platforms provide different capabilities. Inspect the actual service and configuration. Do not assume that every prototype tool has the same limits or that a familiar cloud name satisfies company policy.

Ask seven production questions

The answers can use existing enterprise services. You do not need to build a new identity system or monitoring platform for every application. Connect to the approved capabilities and record the remaining gaps.

AWS Well-Architected considers operation, security, reliability, performance, cost, and sustainability together. This is a useful reminder that a functioning deployment is only one part of an architecture assessment. Read the framework.

Define boundaries between environments

Identify development, test, and production resources. Define which identities can cross these boundaries. Do not copy production records into a convenient preview environment without an approved handling process.

Inspect outbound connections as well as inbound access. A private database can still feed a public logging service through the application. The coding agent’s model calls are another flow to assess separately.

Record who owns the cloud account, DNS, certificate, encryption keys, and billing relationship. A project that depends on a departing employee’s personal account has an ownership problem even when the application code is available.

Test the responsibility split

A managed database provider may operate the underlying service while your organization controls users, data access, schema changes, and retention settings. The exact split depends on the service and contract. Ask for it explicitly.

For the contract application, run a fictional restore exercise. Measure the actual recovery time and identify possible data loss. Compare the result with the business requirement. A checkbox labelled “backups enabled” is not the same evidence.

Also test offboarding. Remove a fictional employee from the identity source and verify the intended access change. Include active sessions, administrator roles, and automation identities in the design.

Connect infrastructure to the delivery system

Infrastructure definitions, environment configuration, pipelines, and application code need coordinated changes. An agent should plan against the real target environment. Otherwise, it can generate a deployment that conflicts with networking, identity, or ownership requirements.

This is where platform engineering and a software factory meet. The platform supplies supported capabilities and boundaries. The delivery system must use them, produce evidence, and preserve a clear operational handover. Continue with platform engineering.