I ask, not to condemn, but to find out what your process was for developing the requirements. Clearly it was done with LLM help but what was the refinement process?
One main thing I did was to use the deep research feature of Claude to get a good understanding of what other tools are offering (features, integrations etc.)
Then each feature in the specs document got refined with manual suggestions and screenshots of other tools that I took.
> Before writing any code, I spent time on detailed specs, an architecture doc, and a style guide. All public: https://github.com/brightbeanxyz/brightbean-studio/tree/main...
> It took me tho 4 full days to get the specs to the level I was happy with.
When I click on history there I see only a single commit for these docs. Would you be willing to share some or all of the conversation you had with the LLM (in a gist or in the repo) that led to these architecture docs? Understand if you can't, but I'm sure it would be super instructive for people trying to understand the process of doing something like this and the types of guide rails that help to move the process forward productively.
1. You mentioned developing tasks in parallel—how many agents were you actually running at the same time? Did you ever reach a point where, even if you increased the degree of parallelism, merging and reviews became the bottleneck, and increasing the number further didn’t speed things up?
2. I really relate to the idea of “80% of features in 20% of the time, then 80% on polish.” Did you use AI for this final polishing phase as well? In other words, did you show the AI screenshots of the screens and explain them? Also, when looking back, do you feel that if you had written the initial specifications more carefully, you could have completed the work faster?
First I triggered all work streams per layer and brought them to a level of completion I was happy with. Then you merge one after another (challenge in github with the @codex the implementation and rebases when you move to the next work stream.
This is roughly how it looked like:
Layer 0 - Project Scaffolding
Layer 1 — Core Features Stream A — Content Pipeline Stream B — Social Platform Providers Stream C — Media Library Stream D — Notification System Stream E — Settings UI
T-0.1 (Scaffolding)
│
T-0.2 (Core Models + Auth)
│
┌───────────────────┼───────────────────┬──────────────┐
│ │ │ │
Stream A Stream B Stream C Stream D
(Content) (Providers) (Media) (Notifs)
│ │ │ │
T-1A.1 Composer T-1B.1 FB/IG/LI T-1C.1 Library T-1D.1 Engine
│ T-1B.2 Others │ │
T-1A.2 Calendar │ │ Stream E
│ │ │ T-1E.1 Settings UI
T-1A.3 Publisher ◄──────┘ │
│ │
└──────────◄───────────────────────────┘
(Publisher needs providers + media processing)
Layer 1 complete
│
┌───────────┼───────────┬──────────────┐
│ │ │ │
Stream F Stream G Stream H Stream I
(Approval (Inbox) (Calendar+ (Onboarding)
+ Portal) Composer
│ enhance)
T-2F.1 Approval
│
T-2F.2 Portal
Additionally, your token usage is very high since you are having so many agent do work at the same time, hence I very often reached my claude session token limits and had to wait for the next session to begin (I do have the 5x Max plan)
Also calling HTMX old makes me feel old.
Yes, and Django feels like more of a Drillator than a simple screwdriver to me.
It's simple, it works, it's efficient, safe, and there are tons of online resources for it. Excellent choice, even more so when using a coding agent.
And for hmtx I simply wanted to have something lightweight that is not very invasive to keep things simple and dependencies low.
In my head this was a good consideration to keep complexity low for my AI agents :-)
Svelte even older (2016, SvelteKit was just an new version in 2022)
SQLAlchemy is ancient (2006)
Use newer tech, like HTMX (2020)
(/s obviously)
What you describe has also been my experience so far with building projects mostly with AI but with detailed specs but Rails instead of Django.
Questions: why no X? Do you have a feature to resize (summarize?) to the text to fit into short boxes?
If you're building anything serious and your data integrity is important, use Postgres.
Postgres is much stricter, and always was. MySQL tried to introduce several strict modes to mitigate the problems that they had, but I would always recommend to use Postgres.
Probably Postgres is there because you can use it as a queue (https://livebook.manning.com/book/just-use-postgres/chapter-...)
Isn't this a SaaS-pocaplyse testament? What's stopping anyone from doing the same to BrightBean? What's stopping anyone with a little of domain knowledge and a $200+ Claude to clone your app and build yet another gap-filling, slightly improved content-syndication version and go-to-market? Is it worth taking it to the market when anyone can perpetuate the cycle?
I'm genuinely interested in knowing your thoughts.
It being open source doesn't help it either, so easy to malus/chardet it.
This will be 2 types of products.
1. A product which requires tech that is not inside the training data distribution of the models underlying coding agents. This is usually then very cutting edge.
2. A product that uses data/insights to generate value for a customer to which a coding agent has no access.
These are the only abstract moats I can think of, the rest will be a race to the bottom
- "In the AI Age, Data is your only Moat!"
Thanks for sharing such valuable insights JanSchu! B-)
It's been running flawlessly for months without a single error. In fact, spooky good. I still feel nervous about it, though and check it every morning.
It is built mostly as a web app in .NET 10 Razor with SQLite db.
The APIs for the social networks are the hazy bit. As OP mentions, some are badly documented. Some are a pain to get access to. I was using a driven browser to post to Twitter, but they opened their API recently, which was nice.
[0] I used Claude in Github Copilot, so total cost was less than the $10/month in credits.
The tool shows all pipelines, builds, buildagents and approval ready in one place at a glance. For the most actions I use the public api but approval I use the internal api because they are still not public, since 4 years.
The tool runs so smooth it scares me too! My guess is that the apis are so stable that it will work for the next 10 years.
Before writing any code, I spent time on detailed specs, an architecture doc, and a style guide. All public: https://github.com/brightbeanxyz/brightbean-studio/tree/main...
I broke the specs into tasks that could run in parallel across multiple agents versus tasks with dependencies that had to merge first. This planning step was the whole game. Without it, the agents produce a mess.
I used Opus 4.6 (Claude Code) for planning and building the first pass of backend and UI. Opus holds large context better and makes architectural decisions across files more reliably. Then I used Codex 5.3 to challenge every implementation, surface security issues, and catch bugs. Token spend was roughly even between the two.
Where AI coding worked well: Django models, views, serializers, standard CRUD. Provider modules for well-documented APIs like Facebook and LinkedIn. Tailwind layouts and HTMX interactions. Test generation. Cross-file refactoring, where Opus was particularly good at cascading changes across models, views, and templates when I restructured the permission system.
Where it fell apart: TikTok's Content Posting API has poor docs and an unusual two-step upload flow. Both tools generated wrong code confidently, over and over. Multi-tenant permission logic produced code that worked for a single workspace but leaked data across tenants in multi-workspace setups. These bugs passed tests, which is what made them dangerous. OAuth edge cases like token refresh, revoked permissions, and platform-specific error codes all needed manual work. Happy path was fine, defensive code was not. Background task orchestration (retry logic, rate-limit backoff, error handling) also required writing by hand.
One thing I underestimated: Without dedicated UI designs, getting a consistent UX was brutal. All the functionality was there, but screens were unintuitive and some flows weren't reachable through the UI at all. 80% of features worked in 20% of the time. The remaining 80% went to polish and making the experience actually usable.
The project is open source under AGPL-3.0. 12 platform integrations, all first-party APIs. Django 5.x + HTMX + Alpine.js + Tailwind CSS 4 + PostgreSQL. No Redis. Docker Compose deploy, 4 containers.
Ask me anything about the spec-driven approach, platform API quirks, or how I split work between the two models.