Ship AI-written code you can still explain: own the design, make the agent show its assumptions

One developer's rule: "I don't let the LLM design the code at all, I let it implement a design I have created." He decides "at the UML level, how abstractions will sit and interact", then has the agent implement the gaps. The payoff: "I may not know how each unit is implemented, but I know each unit's responsibilities and how they interact with one another." The design doc becomes a living document.

Another developer writes a scratch Python file next to his Rust+Tauri app because "Python reads like pseudocode, and pass and ... (Ellipsis) let me skip whatever I don't care about." Where he cares, he states method signatures, domain-model fields and invariants; elsewhere he leaves an ellipsis. It runs both ways: "I ask the agent to sketch the current system in Python, with some details elided and others spelled out. Then I edit the sketch and have the agent push the changes back into the much bigger Rust+Tauri app."

A third developer reviews plans as "rendered diagrams. Data structures, flowcharts, uml sequences, and component architecture docs." His tweak: "making the AI look the plantuml and the rendering and reviewing before presenting it to me results in much more legible designs."

One developer has the agent "Write a templated description of the change: interactions with external systems, sequence diagram, happy path, corner cases, deploy checklist, settings change, etc." Then: "I review high level choices, and not the code." He did not share the template itself; the fields above are the full list he gave.

The same developer has the agent add review threads "everywhere it made suppositions and related to untested assumptions." Each one says "what could not be teste[d], what the llm thinks, what happens if I don't change anything and release the code." For the parts he wants to be sure of, he asks the agent to build a test bench, for example "provide me the command to fetch data, and a script to anonymise it" for a before/after comparison.

One commenter's answer to the original question: "just use AI to explain the code to you before committing." He adds that he still wants to "understand every line of code that's committed with my name on it." Another developer plans iteratively with the model, has it write tests, and says "I try to review changes myself and ask it to explain the changes." Check the explanation against the diff rather than taking it on trust.

One developer keeps "a /docs folder that holds the full specification of the code" (block diagrams, edge-case behaviour, problem statement) and makes sure "the agent follows these specs and ensures the specs are updated after every code change." Another keeps it lighter: "The README defines the expected behavior of my API endpoints and the overall program in natural language," and Claude Code tests the endpoints it touched against that.