Domain-Driven Agents

📝 Discussion Summary (Click to expand)

Three prevalent themes from the discussion

Up‑front architecture and specifications are essential for LLMs in greenfield work
“The single thing that seems to have helped is that we all agreed to use OpenSpec early on, and to commit the specs alongside the code.” – fcarraldo
“If you start with architecture docs that outline the high level goals and what the core abstractions should be … you can get pretty good results.” – spicyusername
LLMs perform better on existing projects where structure and conventions already exist
“My experience has been the opposite. They work well on existing projects but are not so great at new ones (unless you are just vibe coding something simple).” – owaiswiz
“I hate the start of a project with an LLM. The code it writes is always expletive, and wrangling it … is just a series of half measures.” – soganess
Keeping documentation tightly coupled with code improves LLM reliability
“We've had a lot of success with having the agents write and maintain a docs.md file in each folder, and requiring that they both read and update that file whenever they make a change in that folder.” – theahura
“I've also been a little‑d DDD fan, and we've had luck with per‑entity md files to language‑independent document domain behavior/quirks/usages.” – stephen

🚀 Project Ideas

Generating project ideas…

A CLI that bootstraps a new greenfield project with an OpenSpec-inspired architecture document and templates that keep high‑level goals alongside code.
Core value: gives LLMs and developers a shared “north star” from day one, reducing spaghetti and the need for constant hand‑holding.

Key	Value
Target Audience	Startup teams, indie hackers, and any developers beginning a new codebase who want LLM assistance without architectural decay
Core Feature	Generates a project skeleton with an `ARCHITECTURE.md` (OpenSpec style), per‑folder `docs.md` placeholders, and example code that references these docs; includes a watcher that prompts the LLM to update docs when code changes
Tech Stack	Node.js (or Go) for CLI, TypeScript templates, optional integration with OpenAPI/Swagger for spec validation
Difficulty	Medium
Monetization	Hobby

HN users highlighted that “having the intent develop alongside the code… ensures that agents have a more holistic view” (fcarraldo) and that “architecture docs that outline the high level goals… give the first few agents the 'north star'” (spicyusername).
Provides a concrete way to implement those insights, likely sparking discussion on best practices for LLM‑friendly project initialization.

A language‑agnostic VS Code extension (or Language Server) that automatically creates/updates a sibling .md file for each source file, keeping documentation in sync with code changes.
Core value: eliminates the “fistfight with a cheerful drunk keyboard” by ensuring agents and humans always see matching docs without manual effort.

Key	Value
Target Audience	Developers using LLMs for code generation or refactoring who struggle with outdated or missing documentation
Core Feature	On file save, extracts comments, type signatures, or JSDoc/docstrings and writes/refreshes a corresponding `<file>.md`; can also parse the `.md` to inject stubs or reminders into code
Tech Stack	TypeScript/VS Code API, tree‑sitter for parsing, optional Rust backend for high‑perf extraction
Difficulty	Medium
Monetization	Hobby

Commenters praised per‑entity .md files (stephen) and folder‑level docs.md (theahura) as ways to give agents context; DocSync automates exactly that workflow.
Enables practical utility: reduces friction in PR reviews and encourages adoption of documentation‑first habits among LLM‑augmented teams.

A lightweight service that parses a repository’s architecture decisions (from ARCHITECTURE.md, OpenSpec files, or annotated code) and exposes them via an API for LLM agents to query context on demand.
Core value: provides agents with a holistic, up‑to‑date view of the project’s design without bloating the prompt with entire codebases.

Key	Value
Target Audience	Teams employing AI coding assistants that need scalable, token‑efficient context for large monorepos
Core Feature	Indexes architecture documents and code comments, builds a searchable graph of components, responsibilities, and decisions; offers `/context` endpoint returning relevant snippets for a given file or feature request
Tech Stack	Python (FastAPI) or Node (Express), Neo4j or SQLite for graph storage, embeddings via sentence‑transformers for semantic search
Difficulty	High
Monetization	Revenue-ready: SaaS subscription (tiered by repository size/API calls)

HN discussion emphasized that agents perform better when they can rely on “years of established structure/conventions” (owaiswiz) and that having “intent develop alongside the code” improves outcomes (fcarraldo); ArchCtx makes that intent programmatically accessible.
Could become a central piece in LLM‑augmented dev workflows, prompting discussion on optimal context windows and architecture‑first AI assistance.