AI Prompt for Sync-to-Async Conversion

Convert sync code to async with this AI prompt: correct async/await conversion, independent calls parallelized, and every semantic change flagged.

The prompt

Role: You are an async programming expert converting blocking code to non-blocking [TO_MODEL].

Context: From: [FROM_MODEL: sync/threaded]. To: [TO_MODEL: asyncio / async-await / reactive]. Language: [LANGUAGE].

Code:
[CODE]

Constraints on what must stay sync:
[STAY_SYNC]

Task: 1) Convert to [TO_MODEL], keeping [STAY_SYNC] synchronous with clear sync/async boundaries. 2) Identify independent operations and parallelize them; state the failure semantics chosen (fail-fast vs gather-all). 3) List semantic differences: execution order, error propagation, resource usage, backpressure. 4) Provide the converted code plus a usage example. 5) Warn about common async pitfalls in this specific code (e.g. blocking calls smuggled into the event loop).

Format: Converted code first, then the parallelization decisions, then semantic differences, then the pitfalls warning.

Ask the AI to draw the new execution timeline for your main flow: which operations now overlap and where the awaits serialize. Visualizing concurrency prevents the 'it runs but weirdly' class of bugs.

Constraints: No blocking calls inside async code — flag every one. Preserve [STAY_SYNC] boundaries exactly. Error handling must be explicit for the parallelized parts.

Variables to fill in

Replace each [variable] in the prompt with your own details:

[FROM_MODEL] [TO_MODEL] [LANGUAGE] [CODE] [STAY_SYNC]

Example input

FROM=synchronous Python, TO=asyncio, CODE=script fetching 50 URLs sequentially with requests, CONSTRAINTS=keep function signatures where possible

How to use this prompt

Say which calls are independent — parallelization is the main win of async. List constraints on what must stay sync; not everything should convert.

Tips for the best results

Audit for blocking calls in the converted code — they silently destroy async benefits.
Choose fail-fast vs gather-all deliberately; the AI defaults need your confirmation.
Load-test the converted code; async changes resource profiles significantly.