The gap you described in the current stack feels very real. Systems keep moving forward with less and less context attached to them. Over time the code still compiles, the tests still pass, features keep shipping but the system slowly loses its identity.
Interestingly, I think this problem exists in humans as well. Teams carry context informally: conversations, intuition, institutional memory. As people change teams or leave, that context fades. AI just exposes the fragility of this much faster.
Reading the article made me wonder about the shape of the “architectural substrate” you describe. Would it look like a very comprehensive, continuously maintained artifact, something like a living system document that captures architectural decisions, invariants, and boundaries across teams? Or could it evolve into something more dynamic, perhaps a layer of systems or agents that continuously observe the codebase, infer and enforce invariants, and communicate architectural constraints as the system evolves? Or something else?
Really appreciate that, Chandan. This problem needs people who’ve actually felt the drift up close, not just watched the demos. I’m building the substrate exactly for teams like yours, and having thoughtful operators around it early will make it stronger.
I’ll reach out when the first beta group opens. Would be great to have your eyes on it as it takes shape.
Really appreciate this. You’re describing the exact pain that pushed me to start working on the substrate in the first place. The drift isn’t a model problem. It’s the system losing the memory of itself, and AI just removes the human buffer that used to hide it.
I’m deep in this problem right now and building toward something that tackles it at the architectural level, not the tooling level. I’ll be opening it up for early testing soon, and your perspective is exactly the kind of thinking I want around it.
Amazing POV, this resonated a lot.
The gap you described in the current stack feels very real. Systems keep moving forward with less and less context attached to them. Over time the code still compiles, the tests still pass, features keep shipping but the system slowly loses its identity.
Interestingly, I think this problem exists in humans as well. Teams carry context informally: conversations, intuition, institutional memory. As people change teams or leave, that context fades. AI just exposes the fragility of this much faster.
Reading the article made me wonder about the shape of the “architectural substrate” you describe. Would it look like a very comprehensive, continuously maintained artifact, something like a living system document that captures architectural decisions, invariants, and boundaries across teams? Or could it evolve into something more dynamic, perhaps a layer of systems or agents that continuously observe the codebase, infer and enforce invariants, and communicate architectural constraints as the system evolves? Or something else?
Thoughts?
Happy to hear that you're already building something towards it.
If there’s any way I can help feedback, testing, or just sharing thoughts as things evolve. I’d be more than happy to contribute.
Really appreciate that, Chandan. This problem needs people who’ve actually felt the drift up close, not just watched the demos. I’m building the substrate exactly for teams like yours, and having thoughtful operators around it early will make it stronger.
I’ll reach out when the first beta group opens. Would be great to have your eyes on it as it takes shape.
Really appreciate this. You’re describing the exact pain that pushed me to start working on the substrate in the first place. The drift isn’t a model problem. It’s the system losing the memory of itself, and AI just removes the human buffer that used to hide it.
I’m deep in this problem right now and building toward something that tackles it at the architectural level, not the tooling level. I’ll be opening it up for early testing soon, and your perspective is exactly the kind of thinking I want around it.
Happy to keep you in the loop as it gets closer.