Aporia Cuts AI Coding Mismatches by 79% by Making Hidden Decisions Visible
A new VS Code extension forces coding agents to ask you design questions first, cutting mental-model mismatches by 79 percent in a user study.
- Researchers introduce decision-oriented programming, a paradigm making design choices explicit between humans and coding agents.
- Aporia is a VS Code extension that elicits yes/no design questions before implementation and tracks answers in an editable Decision Bank.
- Each committed decision is auto-encoded as an executable test suite that validates the agent's implementation.
- In a 14-participant study, mental-model mismatches with actual code were 79% less likely vs. Claude Code baseline.
- Participants articulated 2.99x more design decisions overall and 13.5x more elicited decisions than with the baseline agent.
- Limitations include high latency, UI complexity, and underused test suites; both tools used Claude Sonnet 4.6.
AI coding agents routinely choose data flows, error behavior, access rules, and edge-case policies while turning a brief prompt into a patch. Those choices often remain implicit, leaving developers with code they can inspect but a weaker account of why it behaves as it does. Aporia, a research prototype for VS Code, inserts a structured decision step before implementation. In a controlled study of 14 programmers, the interface increased articulated decisions and reduced mismatches between participants’ descriptions and the resulting code.
Researchers at UC San Diego, the University of Pennsylvania, Cornell, Technion, and the University at Buffalo describe Aporia in a preprint about decision-oriented programming (DOP). Under DOP, an agent identifies choices embedded in a task, asks the developer to resolve them, records the answers, and connects each decision to executable tests before changing the codebase.
Agents accumulate hidden choices
The paper cites 2025 survey data showing that 84% of developers used AI tools and 47% used them daily. Coding agents now extend beyond autocomplete: they can inspect repositories, edit multiple files, run shell commands, and revise their own implementations. In the common prompt-generate-review workflow, the agent may settle policy questions that the developer never explicitly considered.
The authors call the resulting loss of system understanding cognitive debt. As implicit choices accumulate, a team’s mental model can diverge from the code’s actual behavior. Developers often compensate by limiting each change, reviewing generated code line by line, and writing plans or design documents manually. Those safeguards require time and sustained attention.
Decisions become first-class artifacts
- Explicit and structured. Each choice becomes a persistent, editable record with its own identity and status.
- Interactively co-authored. The agent surfaces unresolved choices and asks the developer to commit an answer.
- Traceable to code. Each decision connects to executable tests that encode the intended behavior.
From goal to tested code
Aporia is a TypeScript extension for VS Code that keeps goals, questions, decisions, and generated tests inside the editor. The agent begins modifying the repository after the developer commits the decision set through a four-step workflow.
This story is for Pro members
You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.
Comments
