Hosted version: https://coastmath.com/
iOS app: https://apps.apple.com/app/id6809539979
This post is partially inspired by this thread: https://news.ycombinator.com/item?id=50048344
``` > Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt.
>> No, they can't, at least not any semblance of quality. The cases we're seeing where this does kinda work is in ports and translation where all the rules are already documented in the best specification language possible with a way for the LLM to verify itself: code. We saw this close to a year ago now with Cloudflare and NextJS ```
This post mostly outlines how I am personally building with agents, as I am almost entirely hands-off.
Desired feature-set determined by having agents research existing products and feedback / feature requests on FIRE subreddits.
Features built using Opus 5.5 (xhigh) / Fable 5.1 (med/high) as orchestrator and Opus-5-5 low/medium as worker subagents.
Quality is attempted by running Opus-5-5 agents in a perpetual audit + fixer loop on the calculation / compliance engine until no medium+ severity can be found. The origin of each rule / logic must be backed up and documented in the codebase with a source.
Sonnet subagents run UX / QA audit on iOS simulators within the Claude Code desktop app, and Opus agents resolve all medium+ sev issues; also run on a perpetual loop.
I have not manually tested or used the app(s) for more than 30 minutes. For UX quality, agents' screenshots + associated context are fed to a combination of GPT / Gemini / Claude agents until consensus is achieved on quality (7/10+).
Early feedback from beta testers report extremely close results to Boldin / Projection Lab. There is also a feature to import your data from Boldin and Projection Lab into CoastMath if you wish to compare the results yourself.
There is also a SimpleFIN integration, if you wish to pull in your own transaction data and see how your financial future changes based on your monthly spending habits (e.g. when I will get to retire if I spend like I did in June vs. July?).
I'm sure there are other features I don't yet know about. Again, I've barely used the app.
Anyway, feel free to use / roast / steal. I don't think that any moat can exist in this class of products in the future (where the rules are laws that are visible to everyone) and I don't expect to turn a profit.
Estimated cost: maybe $2000 if done over the API? I did most of this using the $200 Max subscription.