If only we had a name for this - where you - I don't know, kinda map out a flow - lets call it a value stream. And you look at the stations and the queues - I wonder - thinking out loud her - maybe we should call it something like Value Stream Mapping? Of course there would probably be a precursor to this - you know - writing down all the things that actually take the time - kind of making work visible . . .
Just wish we didn't have to come up with all of this stuff from scratch :)
The distinction between "AI bolted on" and "SDLC redesigned" is the right frame. But I think the compounding gap you describe has a deeper root than most people realize.
The teams that are compounding aren't just faster at shipping code. They've done something harder first – they've extracted the tacit organizational knowledge that used to live only in people's heads and made it explicit enough for agents to work with.
That extraction step is the bottleneck. Not the models, not the tokens, not the tooling. It's the organizational specification. What does "good" mean here? What are the constraints? What does our architecture actually assume?
The companies that specified their processes before the agentic wave hit are the ones compounding now. The ones that didn't are discovering that faster agents just compound the ambiguity.
The skyscraper foundation metaphor is right. The foundation isn't the tooling. It's the specification that the tooling executes against.
Agreed 100% and perhaps I should have made this clearer. You can download the prompts from StrongDM, tweak them and have a software factory in a couple of days. Or you can use or build pretty much any pipeline processing system to implement your orchestrator. ALL the work is in capturing and verifying the tacti knowledge for your code base and company and industry and teams and the subsets that do and don't apply to various repos. Good is probably not identical for your React code and your data pipelines or for a random internal admin back end and your primary customer facing mission critical flows.
Thanks for all the examples!
Indeed, it depends on rethinking the value stream, not just improving the slowest part.
If only we had a name for this - where you - I don't know, kinda map out a flow - lets call it a value stream. And you look at the stations and the queues - I wonder - thinking out loud her - maybe we should call it something like Value Stream Mapping? Of course there would probably be a precursor to this - you know - writing down all the things that actually take the time - kind of making work visible . . .
Just wish we didn't have to come up with all of this stuff from scratch :)
Hopefully all of our lean/agile coaching friends will be kept busy for at least a season :)
Maybe you should write a book about it!
smarter people than me already did :) Anyway, my O'Reilly book should be enough to keep me busy this month and next! http://scalingaibook.com/
The distinction between "AI bolted on" and "SDLC redesigned" is the right frame. But I think the compounding gap you describe has a deeper root than most people realize.
The teams that are compounding aren't just faster at shipping code. They've done something harder first – they've extracted the tacit organizational knowledge that used to live only in people's heads and made it explicit enough for agents to work with.
That extraction step is the bottleneck. Not the models, not the tokens, not the tooling. It's the organizational specification. What does "good" mean here? What are the constraints? What does our architecture actually assume?
The companies that specified their processes before the agentic wave hit are the ones compounding now. The ones that didn't are discovering that faster agents just compound the ambiguity.
The skyscraper foundation metaphor is right. The foundation isn't the tooling. It's the specification that the tooling executes against.
Agreed 100% and perhaps I should have made this clearer. You can download the prompts from StrongDM, tweak them and have a software factory in a couple of days. Or you can use or build pretty much any pipeline processing system to implement your orchestrator. ALL the work is in capturing and verifying the tacti knowledge for your code base and company and industry and teams and the subsets that do and don't apply to various repos. Good is probably not identical for your React code and your data pipelines or for a random internal admin back end and your primary customer facing mission critical flows.