Where AI Stands in Software Development
Episode hosts
What we discussed on the recording
The first 3 AImigo episode establishes an AI-development baseline. “We adopted AI” may mean autocomplete, agents, or process automation. Cheap execution does not create value: start from a concrete problem, current process, context, and outcome.
Uneven adoption creates bubbles: an engineer mistakes a workflow for the norm, leaders mistake licenses for transformation, and employees hide practice. AI champions join product teams, get several iterations, and spread tested techniques. Those become policies, permissions, and platform capabilities.
Agents need machine-readable specifications, accessible repositories, reproducible environments, and checks. Tests, linters, and mocks become conditions for delegation. In a Pydantic migration, one engineer can automate changes only when service owners provide review and smoke tests; otherwise speed exposes context debt.
Cases include visa evidence, illustrations, virtual personas, and interview-built specifications. Choose an observable outcome and verifiable environment: proof checkers for mathematics, tests for code, and telemetry for products. Agent swarms remain forecasts, but the split is visible: AI executes while people retain intent, architecture, critique, and accountability.