Productivity
Shipping With AI: My Development Workflow
By Vivek Jalondhara · 1 min read
Published Aug 2025 · Updated
The gains are real but they're earned through process, not magic. Here's the loop I run.
My AI-assisted workflow starts with a narrow specification, delegates repetitive implementation, and ends with review and tests. I do not use one productivity percentage to describe tasks with different levels of difficulty.
I front-load thinking. A clear plan with explicit contracts and acceptance criteria turns an AI from a guesser into an executor. Vague intent produces vague output; precise intent produces shippable code.
Judgment, architecture, and edge cases remain my responsibility. Generated code is a starting point that needs the same scrutiny as any other contribution.
Review a behavior change through its failure cases
A practical AI workflow makes the final behavior reviewable. For a booking reminder change, write down what must happen on retries and cancellation before requesting an implementation.
Illustrative example · Acceptance checklist · Simplified for explanation
Before implementation:
- Define when the reminder becomes due.
- Define what cancellation changes.
- Define the retry and duplicate-send behavior.
After implementation:
- Exercise those cases and inspect the resulting state.- Acceptance criteria
- Small implementation
- Failure-case checks
- Review & release
Conceptual flow for the example above.
Tradeoff to keep in mind
Time saved on typing can be lost during debugging. Evaluate the full change, including review and maintenance, without attributing a fixed percentage to every task.