frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

I'm building a 30k‑line V12 codebase solo with a "team" of 4 AIs

7•garylauchina•6h ago
I’m a solo developer working on a “complex systems measurement” project that has grown to over 30k lines of code and is now at V12. Every line so far has been written by one person (me), with the research notes and design docs in a separate repo: https://github.com/Garylauchina/Prometheus-Research.

I’ve been using Cursor heavily along the way. The models are genuinely good and the local code they generate is often excellent, but on a large, evolving codebase I kept running into the same problem: context limits caused subtle architectural drift. The AI would write clean functions that were globally wrong, quietly breaking earlier design decisions and long‑range invariants.

What finally helped was to stop treating “AI” as a single assistant and instead treat different models as different team members with clear roles and constraints.

My current setup looks like this:

Perplexity + ChatGPT → “product / research brains” I use them for requirements, trade‑offs, and high‑level architecture sketches. They live outside the IDE and exist to clarify what I’m building and why before any code is touched.

Cursor, window 1 (GPT‑5.2) → “architect” This instance is not allowed to write production code. It is responsible for architecture and module boundaries, writing design notes and developer guides, defining interfaces and contracts, and reviewing diffs. I treat it like a senior engineer whose main output is prose: mini‑RFCs, comments, and checklists.

Cursor, window 2 (Sonnet 4.5) → “programmer” This one only implements tasks described by the architect: specific files, functions, and refactors, following explicit written instructions and style rules. It doesn’t get to redesign the system; it just writes the code.

The key rule is: architect always goes first. Every non‑trivial change starts as text (design notes, constraints, examples), then the “programmer” instance turns that into code.

This simple separation fixed a lot of the weirdness I was seeing with a single, all‑purpose assistant. There is much less logical drift, because the global structure is repeatedly restated in natural language. The programmer only ever sees local tasks framed inside that structure, so it’s harder for it to invent a new accidental architecture. The codebase, despite being tens of thousands of lines, feels more coherent than earlier, smaller iterations.

It also changed how I think about Cursor. Many of my earlier “Cursor is dumb” moments turned out to be workflow problems: I was asking one agent, under tight context limits, to remember architecture, requirements, and low‑level implementation all at once. Once I split those responsibilities across different models and forced everything through written instructions, the same tools started to look a lot more capable.

This isn’t a Cursor ad, and it’s not an anti‑Cursor rant either. It’s just one way to make these tools work on a large solo project by treating them like a small team instead of a single magical pair‑programmer.

One downside of this setup: at my current pace, Cursor is happily charging me something like $100 a day. If anyone from Cursor is reading this – is there a “solo dev building absurdly large systems” discount tier I’m missing?

Comments

zkmon•6h ago
How does the hand-off work among the team roles? Is it a water-fall or iterative? How does the programmer complain about something that's wrong or inconsistent in architecture or requirements?
garylauchina•3h ago
It’s iterative, not a strict waterfall, but the constraint is “text first, code second”.

A typical loop looks like this:

I talk with Perplexity/ChatGPT to clarify the requirement and trade‑offs.

The “architect” Cursor window writes a short design note: intent, invariants, interfaces, and a checklist of concrete tasks.

The “programmer” Cursor window implements those tasks, one by one, and I run tests / small experiments.

If something feels off, I paste the diff and the behavior back to the architect and we adjust the design note, then iterate.

There is a feedback channel from “programmer” to “architect”, but it goes through me. When the programmer model runs into something that doesn’t fit (“this API doesn’t exist”, “these two modules define similar concepts in different ways”, etc.), I capture that as:

comments in the code (“this conflicts with X”), and

updates to the architecture doc (“rename Y to Z, merge these two concepts”, “deprecate this path”, etc.).

So the architect is not infallible. It gets corrected by reality: tests failing, code being awkward to write, or new edge cases showing up. The main thing the process enforces is that those corrections are written down in prose first, so future changes don’t silently drift away from whatever the system used to be.

In that sense the “programmer” complains the same way human ones do: by making the spec look obviously wrong when you try to implement it.

nunobrito•5h ago
You really seem on top of your game, just to mention that 30k code bases aren't typically complex on my experience. When they, it just means you need to simplify the overall struture, unify as much as possible the functions
garylauchina•3h ago
Fair point – 30k LOC by itself isn’t “complex” in any absolute sense. In my case the difficulty isn’t the raw size, it’s:

long‑running experiments and measurement contracts that span many modules,

an evolving research agenda (I’m at V12 now, with quite a few pivots along the way),

and the fact that I’m doing all the design + implementation solo.

I agree that when something feels complex at this scale, it’s often a sign the structure needs to be simplified and unified. A big part of what I use the “architect” agent for is exactly that: forcing myself to rewrite and consolidate interfaces instead of just piling on more code.

I'm having the worst career winter of my life

17•mariogintili•1h ago•18 comments

Ask HN: Who is using local LLMs in a production environment here?

6•Haeuserschlucht•4h ago•2 comments

Tell HN: Happy New Year

437•schappim•2d ago•205 comments

How to use AI to augment learning without losing critical thinking skills?

13•mintsuku•11h ago•12 comments

I optimised my vibe coding tech stack cost to $0

7•udit_50•5h ago•7 comments

Ask HN: When do we expose "Humans as Tools" so LLM agents can call us on demand?

36•vedmakk•21h ago•27 comments

A quantum-resistant RNG powered by collective human entropy

4•EntropyGrid•6h ago•0 comments

Security breaks during partial failures – design notes from distributed systems

6•sandhyavinjam•14h ago•1 comments

I'm building a 30k‑line V12 codebase solo with a "team" of 4 AIs

7•garylauchina•6h ago•4 comments

Ask HN: Building a tool to ensure things get done on time

3•Vishal19111999•12h ago•2 comments

Ask HN: What did you read in 2025?

335•kwar13•1w ago•444 comments

Ask HN: How did you learn to code?

26•chistev•1d ago•74 comments

Ask HN: Why is Apple's voice transcription hilariously bad?

7•keepamovin•17h ago•4 comments

Ask HN: Loneliness at 19, how to cope?

61•yresting•4d ago•105 comments

Ask HN: What is the best microVMs for AI agents?

9•zfoong•1d ago•8 comments

Ask HN: How Are You Handling Auth in 2026?

10•joshcsimmons•22h ago•14 comments

Ask HN: Is being hungry enough to win?

7•meysamazad•1d ago•4 comments

Ask HN: Any example of successful vibe-coded product?

79•sirnicolaz•2d ago•117 comments

I built a public skill registry and MCP server so Codex can install new skills

3•iluxu•1d ago•0 comments

Ask HN: Which AI productivity tools are you using in 2026?

4•Vishal19111999•12h ago•0 comments

Semantica – Open-source semantic layer and GraphRAG framework

8•kaifahmad1•1d ago•0 comments

Ask HN: Does reading HN make you happy?

47•yakattak•2d ago•38 comments

Ask HN: How to do a Personal Cybersecurity audit

24•preciousoo•3d ago•12 comments

Tell HN: Happy New Year!

4•realberkeaslan•1d ago•2 comments

Ask HN: How long before the first civilian cargo flights are AI piloted?

3•givemeethekeys•1d ago•13 comments

Ask HN: How did you make yourself more marketable?

11•ronbenton•2d ago•13 comments

Happy New Year HN!

11•thunderbong•1d ago•4 comments

A curated directory of open-source AI projects

12•doanbactam•3d ago•2 comments

Ask HN: How to go back to listening to MP3s?

9•muratsu•5d ago•25 comments

TP-Link only works with a permanent internet connection

8•roscas•3d ago•7 comments