Hi Prukalpa, thanks for writing this down. Sounds sensible and echoes my own work. However I am having a hard time translating your article into something useful.
Let me take this first at a personal level. For my own work with AI agents I have elements of a “context layer” everywhere. I use Karpathy’s LLM wiki for knowledge management, I use Graphify to create graphs of my source code, I use mempalace, beads, … any number of tools to create the context layer I need. It’s hard for me, so how can an enterprise do it? They are many times more complex than my simple needs. They have data catalogs, databases, systems of record everywhere. It sounds like a nice term but is context layer really a practical thing for enterprises to implement?
This is very interesting and similar to a concept that I’ve been researching - the semantic core - a governed model of
organisational intent that occupies a specific and previously empty position between strategic intent and operational execution. It argues that the absence of this artefact is the structural cause of enterprise transformation failure, and that making it explicit changes what is possible for any organisation attempting to evolve. Looking forward to hearing more.
It is very clear to me that you are building this--the level of clarity in this article proves it. As I was reading, each question forming in my mind was answered in the following paragraphs.
One piece that is not included is the security part. The surface area for attack is huge. The agentic paradigm and the chosen units of consumption for agents are vulnerable to new classes of attack and traps.
Just as you said, legacy components were built for humans, and the same is true for security. I've proved how various components can be exploited--because the underlying low-level unit is an open surface attack area.
I'd be happy to tell you more about it. I'll reach out to you on LinkedIn.
Great job 🫡👍
Hi Prukalpa, thanks for writing this down. Sounds sensible and echoes my own work. However I am having a hard time translating your article into something useful.
Let me take this first at a personal level. For my own work with AI agents I have elements of a “context layer” everywhere. I use Karpathy’s LLM wiki for knowledge management, I use Graphify to create graphs of my source code, I use mempalace, beads, … any number of tools to create the context layer I need. It’s hard for me, so how can an enterprise do it? They are many times more complex than my simple needs. They have data catalogs, databases, systems of record everywhere. It sounds like a nice term but is context layer really a practical thing for enterprises to implement?
This is very interesting and similar to a concept that I’ve been researching - the semantic core - a governed model of
organisational intent that occupies a specific and previously empty position between strategic intent and operational execution. It argues that the absence of this artefact is the structural cause of enterprise transformation failure, and that making it explicit changes what is possible for any organisation attempting to evolve. Looking forward to hearing more.
It is very clear to me that you are building this--the level of clarity in this article proves it. As I was reading, each question forming in my mind was answered in the following paragraphs.
One piece that is not included is the security part. The surface area for attack is huge. The agentic paradigm and the chosen units of consumption for agents are vulnerable to new classes of attack and traps.
Just as you said, legacy components were built for humans, and the same is true for security. I've proved how various components can be exploited--because the underlying low-level unit is an open surface attack area.
I'd be happy to tell you more about it. I'll reach out to you on LinkedIn.