@sd141414: Everyone has the same models. The gap shows up in the loop. If every retry needs you, you’re the bottleneck. If the system can evaluate itself, the leverage compounds. This is the emerging paradigm. #SystemsThinking #SoftwareArchitecture #AICoding #AgenticSystems #EngineeringJudgment
naw. still uses llm heavy, linear context rehydration, no matter how you slice it. need to shift to state first.. hit me up. 😎
2026-05-19 19:38:00
0
Jimmy CrackCrn :
Exactly! I use this to stop drift too. https://github.com/jccidc/.plan-enforcer
2026-04-24 01:12:42
3
Daniel Allen :
What you're describing matches how I've been doing a major system migration. I started with a legacy system and I defined specific workflows and business outcomes, that the existing system accomplishes, then I tell the AI first to spec out the flow, then to rebuild the logic in a new tech stack, then build unit tests, then do a gap analysis, each time through the cycle it catches the issues and corrects them.
2026-03-18 04:26:53
5
heroiklife :
You still have to judge the result. That’s just advanced, prompting not even advanced. Most genic architectures still screw up 1/3 of the time. This is well documented in White papers. Human judgment still required in fact, more advanced human judgment.
2026-05-05 19:05:33
0
meanmode :
I would take it a step further and say “why” instead of what. First-principle manifests allow for the model to explore at a conceptual level “what” could mean, if you provide meaningful sources.
2026-04-25 23:30:37
0
George :
The sweet spot is in the middle for me. I don't want the waterfall-like aspects of spec driven development. But I also don't want to have to approve every single thing. I think there's still value to me being in the loop, but the goal should be to limit the number of loops and each pass of the loop should have a right sized scope.
2026-04-25 01:39:59
2
Robert Morrissey Artist :
It’s also the same in imagery.
2026-04-24 00:00:20
0
Troy Thompson :
u have shown me so much when I make it I'll shout out 100%. thank u so much !
2026-03-17 16:22:10
3
iamjasonic :
Finditsales.com was built this way. Builder/watcher/qa. And a bug finder.
2026-03-18 01:12:01
2
Bobby :
That sounds cool in theory but what do you do when the system determines you need a new api or you need to sign up for a new or higher tier git or repo or dns manager or deployer like railway? I would genuinely love your opinion. You’ve been extremely helpful in turning my micro saas into something that can make money and is safe.
2026-04-23 08:34:16
1
mlab14 :
This is tough to do with accounting automation where it’s integrated with payment systems and dealing with real $. It’s a challenge.
2026-03-17 19:09:20
3
dbp709 🇨🇦 :
Two ccode instances on Debian. Each in its own work tree. Each in its own tmux session. One architect/auditor. One implementer. You talk to the architect. Queue.json file. Architect queues, implementer does the work. Loop.
2026-03-17 18:46:39
3
Jan-Olof Axelsson339 :
I found a couple of problems using AI to develop code. In case there is an insight you need to go back and make a correction because you missed something earlier, you can end up in a situation where there are some code parts left that should be removed since the revised code may introduce new additional code that make some parts of the old code obsolete and if it stays there is a risc it may cause a potential quality issue. There are a couple of similar situations like this that may be problematic in different ways if you ignore this problems.
2026-03-24 22:51:53
1
deblack.official :
I had the same thought, result first prompting.
2026-03-17 17:30:18
3
David Ndungu :
I am finding that I have to tweak my Claude Code skills almost daily to improve the development cycle.
2026-03-18 22:58:37
0
J :
one Claude code, one Claude prompt assistant me overseeing the two off a prd, or road map or bug
2026-03-19 22:33:02
0
Mumtaz Ahmad :
objective alone cannot create the desired workflow
2026-03-18 15:23:32
0
continuouslee :
If only they didn’t hallucinate and think they actually accomplished the goal. They evaluate goals at the syntax level, not at the systems level.
2026-03-17 19:51:32
2
heimdall223 :
Purchased your stuff simply to support this account but also because everything is highly valuable. But i also have a hidden reason tgat wont be that hidden anymore: supporting you might encourage others here to do the same so that you dont need to increase the number of videos and decrease the quality so that you get more followers and so this remains our little secrets that normies won’t know about.
2026-03-17 16:04:51
1
Pray2Play13 :
Create the environment for the Agents to succeed...
2026-03-18 03:01:07
4
mohamedibrahim8616 :
The issue is when you scale, you find these many gaps that quickly stack up to the point you revert to option 1.
it builds very quickly, but not great in the touch ups.
2026-03-17 19:27:49
2
CreativeBear :
i haven't seen ANYTHING in production actually built like this.
2026-03-17 19:00:55
1
monatemediaofficial :
I literally just realised this 🙄 Thanks for being real 👍
2026-03-17 14:41:00
1
White Raven :
I built something beyond ai
2026-04-03 10:21:55
1
kaggajoseph :
thank for your great advice, I'm looking forward to building something awesome with you
2026-04-27 06:57:13
0
Mark Hinderliter :
I would, but it’s a voice agent that calls me, so I have to be in the loop. But when I can, I make it do what you’ve said.
2026-03-25 04:04:08
0
Moving Degen :
ai creates a lot of unnecessary code.
2026-03-20 07:32:23
0
To see more videos from user @sd141414, please go to the Tikwm
homepage.