Phase 12 · Months 10 to 12

Get teams to adopt the platform

Watch developers use the platform

Internal platforms fail when they suit the platform team instead of the developers. Measure setup time, local feedback time, CI wait time, and the number of manual steps. Pair with users and watch where they hesitate. Treat every confusing step as a bug.

Make feedback loops fast for agents

Agents run your tests and builds many times for each task. A slow test suite costs one person a few minutes a day, but it costs a team that relies on agents hours. Fast, repeatable tests, clear error messages, local environments that start the same way every time, and accurate context files are now the most valuable developer-experience work you can do.

Set up local development

Build a local environment that starts with one command, using containers and local copies of cloud services. Seed it with realistic data that is safe to share. Let tests mock or record model calls, so that neither developers nor agents spend tokens on every unit test. For AI platforms this matters more than any single emulator.

Aspire is worth a weekend even if you do not use .NET. Microsoft renamed it from .NET Aspire with version 13 in November 2025. It now supports Python and JavaScript fully, runs a multi-service app locally with built-in telemetry, and includes an MCP server that lets Claude Code query running services, logs, and traces. That makes it a working example of a dev environment that agents can use. Its deployment tools favour Azure, so use it for local development only.

LocalStack runs copies of AWS services on your machine, and platform job postings mention it, so learn it before you apply to AWS teams that use it. It is no longer open source. In March 2026 LocalStack merged the free Community edition into a single image that needs an account and an auth token. The free tier has no CI credits, and paid plans start at about US$39 a month. Learn the pattern of local cloud emulation rather than depending on the product. Moto mocks many AWS services inside tests.

Make CI fast and rollouts safe

Run unit tests, integration tests, targeted evals, and full eval suites as separate stages. Create preview environments for significant agent changes. Roll out prompts and workflows with feature flags and canary releases, as you would code. Speed up builds and CI, because each minute you save, you save on every run.

Drive adoption without mandates

Run pairing sessions and office hours. Demo real results, publish before and after numbers, collect complaints, and act on them. Give teams a way out when the default does not fit. Measure adoption honestly by active users, returning users, and completed tasks, not by installs.

Test onboarding with a real person

Ask a colleague to build a small agent from your docs without help. Separately, ask a colleague to use your coding-agent setup for a week. Watch every problem they hit. The target is that they can create a project, run it locally, add a tool, run an eval, inspect a trace, and deploy to a test environment with little custom setup.

You are done when

You can show with numbers that the platform cuts the time to a first working agent or removes repeated setup work. Developers use the default setup because it is easier, not because a policy makes them.

Exit checklist