What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
For AI-assisted software work, Derek Wang’s proposal is to keep durable project rules in files the coding model is instructed to read before generating code, rather than relying on chat prompts alone. The files can make intent and boundaries easier to carry from one generation to the next, but they do not guarantee compliance: rules still need working checks and regular maintenance.
Why put project rules in files?
A chat prompt can guide one interaction, but Wang argues it is a poor place to store a project’s lasting architecture and conventions. His concise formulation is, “A prompt is a memory, not an architecture.” In practice, the prompt can point the model to the project’s rules; the rules themselves belong in maintained project files.
As an Amazon Associate I earn from qualifying purchases.
This is an engineering thesis, not a demonstrated universal result. Wang’s essay, published September 16, 2026, describes a practice and one project anecdote, not an independent evaluation of whether file-based instructions reliably improve every coding project.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallHow Wang’s constraint pyramid is meant to work
Wang connects project intent to code through five layers. Each layer should inform the one below it, while experience from implementation can reveal when an upper layer needs revision.
#1 Best Overall
- Strategy: State the project’s purpose and guiding principles in a README or constitution.
- Architecture: Record accepted architectural decisions in ADRs (architecture decision records).
- Contracts: Specify how work should be carried out in specifications and runbooks.
- Gates: Turn verifiable requirements into pre-submit checks.
- Implementation: Have the model produce code within those constraints.
The pyramid is Wang’s framework, not a settled industry standard. Its practical point is to connect high-level intent with rules a developer or automated check can inspect, rather than expecting a model to infer all of that from a growing conversation.
Make each rule’s strength explicit
The essay divides constraints into three levels. This distinction helps teams avoid treating a suggestion as if it were enforced, or leaving a critical requirement as informal prose.
- L1 — Hard constraints: Requirements enforced through code and gates, such as automated checks that reject a prohibited change.
- L2 — Architecture constraints: Accepted decisions represented in ADRs and the project’s file layout. They shape where and how code belongs, but documentation alone does not enforce them.
- L3 — Execution suggestions: Advisory guidance in a runbook, skill, or checklist. These can help direct work but remain suggestions unless backed by enforcement.
These labels describe Wang’s proposed classification. A team should make the distinction visible in its own project files instead of assuming that a model will recognize how binding each sentence is.
A practical starting point for a project
Wang recommends beginning with a file that the coding model is instructed to read before each generation. Keep the first version focused on the information needed to make changes safely: the project’s purpose, module boundaries, and naming rules. As the project grows, connect that entry point to deeper architecture records, contracts, and checks rather than turning one document into an unwieldy catch-all.
- Write down the durable rules. Describe the project’s purpose, module responsibilities, and naming conventions in a project file.
- Label the rules. Indicate whether each requirement is a hard constraint, an accepted architecture decision, or advice.
- Tell the model to read the file. Make this part of the generation workflow so the instructions are available in each relevant session.
- Automate what must not be violated. Add a pre-submit check for requirements that can be tested mechanically, and ensure that the check actually runs.
- Use repeated deviations as feedback. If the same problem keeps appearing, consider whether an advisory step needs a gate, a contract is missing an instruction, or an architectural boundary needs reconsideration.
What the reported 636-line example does—and does not—show
Wang says a README in one project grew to 636 lines as it accumulated boundaries, path conventions, naming rules, and layering guidance. He reports that after the model read the file before each generation, weekly style drift “nearly vanished.” Both the line count and the outcome are his account of that project. The essay provides no study design, sample size, independent measurement, or comparison group, so the anecdote does not establish that a long README—or file-based rules generally—will produce the same result elsewhere.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Files can drift; gates can fail to run
Written rules help only while they remain connected to the implementation. Wang warns that documentation can drift from the code, that an ADR not mapped back to implementation can become a diary, and that a gate that never executes is only a wish. A rule may be clearly written and still be ignored if it is not loaded into the session or checked in the development workflow.
Rank #4
Keep the files and enforcement mechanisms in the same maintenance loop as the code. When architecture changes, update the relevant decision and implementation together. When a check is added, confirm that it runs at the intended point in the workflow. When deviations recur, decide whether to strengthen enforcement, clarify the contract, or revise the boundary rather than merely adding more prose.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
