Scaling BMad

[jasonvanwyk]

Brian and team great job on V4! Really!

I'm starting to wonder how we can scale BMad (I use use Claude Code in VSCode) to run multiple instances - or to get multiple agents (prob dev agents) to work on multiple parts of the code base at the same time. It's nice that Claude Code can connect to GitHub and create/read/solve issues, and I've tried using Claude Code with Git WorkTrees (Which i properly stuffed up - and dont have the time in my current project to try again).

AI (not going to replacing us devs), But as I see it, definitely makes the barrier to entry a lot lower (not withstanding quality issues), and therefore the competition a lot stiffer. So I guess the devs/engineers who learn to yield its power the quickest/best are the ones who stay in the race. i.e. We can now run multiple projects, more easily, using AI - but how best to do it, for me is the question.

For a solopreneur, I almost envision a sort of "ops room" with like say 4 or more standup desks, each with it's own terminals connected to either its own dev server or a central big dev server, with a human operator moving between each terminal, checking, directing, redirecting, instructing the AI's in there respective projects. This could be extrapolated to run these multiple instances on cloud servers, or as multiple human devs working remotely as a team (which we already do anyways), but I guess the point i'm making as with all disruptive technologies, is how do we smaller players take advantage of getting into the game and also having our little space under the sun and not let the big guys always have it all.

[bmadcode]

Thanks for the feedback on V4 and also on the video :) It's something I think about constantly, and will bring up over in the discord also, interesting conversation to be had!

The "ops room" idea is a great way to put it - and its already happening in a sense - I know many that right now work this way even from a single machine with multiple instances and VMs all running from the same machine. I totally see the dev's job turning into that: being the orchestrator. We set the game plan and guide a team of specialized AI agents to handle different parts of the project. Already we have cloud agents that do small simple things that ping us for reviews when ready. This will only get better.

You hit on the main problem right now, though. The tools aren't seamless enough for that kind of teamwork yet, and your struggle with Git WorkTrees is a perfect example. I think the way forward is evolving the BMAD method to manage a "team" of these AIs. We're already seeing this with bots for code review (like Code Rabbit), security, or UI generation. The hard part is getting them all to play nice together.

And that's exactly how smaller players can win. Big companies can throw money at problems, but they can't easily scale a unique vision. For a solo dev or small team, the edge comes from being a great "human-in-the-loop" director. If you can guide your AI team with a clear, cool idea, you can build stuff that stands out from all the generic AI slop out there. That's way more powerful than just being fast.

And you're right about the barrier to entry being lower and devs not being unnecessary. But I think that makes learning the fundamentals even more critical. When anyone can get started, the people who actually know their craft—the ones who can guide the AI, catch subtle bugs, and make smart architectural calls—are the ones who will really stand out. That combo is the real game-changer.

[jasonvanwyk]

For sure Brian - the fundamentals have always been the key! And probably more relevant now than ever!

The nice thing is that the fundamentals are the easiest things about software dev to learn (mastering them is different kettle of fish altogether), but the point is, that it's never been easier to learn programming than it is now.

Now of course the warning is, that as easy it is to learn programming now, it's also a double edge sword - AI aids devs, and even aids learning (this point is often overlooked), but one still needs to know the larger frame work and nuances of software development - which also comes with experience as well.

It's exciting - there's opportunities - we just need to know and master the tools of our trade.

Yes - I think this topic would do better on discord - looking forward to participating in the conversation.

[yesid-bocanegra]

—the ones who can guide the AI, catch subtle bugs, and make smart architectural calls— This my friend, this is the whole point of the disucssion. I fear that at some point we will need regulations about who can use AI, much like kids and social media restrictions.
AI offers so much and the power we can harness now is so much more than we have had until now. Just this week I was able to create a library for my company usin SDD, and inspired by BMAD and Spec Kit I was able to create a codebase of 13k lines of code with test and in my opinon an acceptable degree of quality.
But when you start to actually review the generated you have be incredibly skilled to see through the cracks. I had to rephrase my specification and add some guard rails to prevent the agent to go down paths that it shouldn't, eventually I asked for a refactor that burned 5k lines of code and added another 6k lines of code, now actually doing what I was expecting. Getting to that level requires many years of breaking things to actually know that it will fail.

[yesid-bocanegra]

@bmadcode just because I've touched the subject here, not becuase it belongs to this thread. I hope you read and take it into consideration. We have used test reporting tools like Cucumber, junit html reports and jacoco coverage reports. I think it would be very positive for the agents to actually generate a similar report from the specification to be implemented. That was one of the things that helped me power through thousands of line of code, just let the agent do its magic, validate the AC, FR and Test reports. then do a cautionary review of the test code and that should do it.

[getpat]

I would like to add the salt to this conversation. It is debugging. I have used the agents in a large codebase now for months, the errors are deep and elusive, creating a workflow that includes debugging tools is where the code moves from a gain to valuable. However, your determined effort and tool are exceptional - congratulations. Patrick Bishop *The day science begins to study non-physical phenomena, it will make more progress in one decade than in all the previous centuries of its existence. Nikola Tesla* Rethink your investment objectives, consider hydration as simply the key to your health and well being. Invest in high quality water. PB Confidentiality Notice This message, including any attachments to it, is intended exclusively for the recipient(s) named above. This communication may contain information that is proprietary, privileged or confidential or otherwise legally exempt from disclosure. If you are not the named addressee, you are prohibited to read, print, retain, copy or disseminate this message or any part of it. If you have received this message in error, please notify the sender immediately by e-mail and delete all copies of the message. WARNING: If anyone other than the addressee of this e-mail is reading it, you are in violation of the 1st & 4th Amendments to the Constitution of the United States of America.

…

On Sun, Jun 22, 2025 at 1:46 AM Jason ***@***.***> wrote: For sure Brian - the fundamentals have always been the key! And probably more relevant now than ever! The nice thing is that the fundamentals are the easiest things about software dev to learn (mastering them is different kettle of fish altogether), but the point is, that it's never been easier to learn programming than it is now. Now of course the warning is, that as easy it is to learn programming now, it's also a double edge sword - AI aids devs, and even aids learning (this point is often overlooked), but one still needs to know the larger frame work and nuances of software development - which also comes with experience as well. It's exciting - there's opportunities - we just need to know and master the tools of our trade. Yes - I think this topic would do better on discord - looking forward to participating in the conversation. — Reply to this email directly, view it on GitHub <#255 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACPEXIVBWXC7D4VNJ2KH52D3EZGN5AVCNFSM6AAAAAB7Z4PZISVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGNJUGEYDGMQ> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.*** com>

[jasonvanwyk]

Yes - debugging is the reason we still have jobs. AI is a bit of a double edge sword here, because it's trained on stackoverflow (and others), and although it does a lot of things well, it unfortunately also picks up the slop out there (not saying stackoverflow is slop). As with the nature of tech - it eveolves quickly and the training cutoff dates for LLM's often are months behind newer versions or frameworks. I did see an MPC server the other day that gets LLMs to fetch the freshest version of a tech (I cant remeber where). I have to add though, that AI's have definity helped me write tests and find bugs quicker. When i'm building - I test often and commit often. I also write learnings.md files (and update them after each iteration of a sprint or epic) sometimes to help the AI to remember and not repeat mistakes - as it often can. Let us know if you come across any good workflows or frameworks that we can all benefit from too!