Thanks for the great article! I fully agree with your points, in fact I just published an article on context engineering with a similar message (https://jannalipenkova.substack.com/p/ai-for-business-14-context-engineering). I think your notion of "process engineering" encompasses the creation of agentic tools as well as how they are integrated into workflows.
As additional food for thought, I think that the moat might not even be in the end result (domain & process knowledge encoded in the AI system) but in the methods and tooling that enable this. In my experience, "offloading" domain knowledge into an AI system is rarely successful as a one-off effort. Rather, we should think of it as a continuous, iterative process that runs in parallel to the daily work of domain experts - this allows them to better reflect their reality and maintain the representations up-to-date. The challenge is in providing them with the right tooling, user experience, and incentives so they stick to it on a continuous basis.
Also as a former distressed debt investor at a large fund, the AI software needs to be 100% right to be trusted if it's something like a financial model. Ai productivity tools that make your IC after you've done all the research or help you think of various sensitivity scenarios etc is probably adopted already.
this is the counter i have heard too, but think this piece articulates why the highest leverage thing for team to do is focus on the outcome and the process to get there.
the tools they leverage likely still continue to be produced by specialists, albeit probably at a much higher level of quality utilizing language models.
The existing provider is best place to do so. They need to develop and improve their existing software to stay relevant and to reduce the desire for the purchaser to go it alone.
The AI threat might be deflationary but if the existing can execute well enough they should win.
We will eventually reach a word where SI can do all this. But we don't know what page we are on or how many pages are in the book before we reach the final chapter.
Great post! It also aligns well with Julien Bec’s recent post on how services are becoming the new software. https://x.com/JulienBek/status/2029680516568600933?s=20
Ilya Sutskever believes in hyper specialised AI which probably align with this post. Really enjoyed this read, thank you
Thanks for the great article! I fully agree with your points, in fact I just published an article on context engineering with a similar message (https://jannalipenkova.substack.com/p/ai-for-business-14-context-engineering). I think your notion of "process engineering" encompasses the creation of agentic tools as well as how they are integrated into workflows.
As additional food for thought, I think that the moat might not even be in the end result (domain & process knowledge encoded in the AI system) but in the methods and tooling that enable this. In my experience, "offloading" domain knowledge into an AI system is rarely successful as a one-off effort. Rather, we should think of it as a continuous, iterative process that runs in parallel to the daily work of domain experts - this allows them to better reflect their reality and maintain the representations up-to-date. The challenge is in providing them with the right tooling, user experience, and incentives so they stick to it on a continuous basis.
Also as a former distressed debt investor at a large fund, the AI software needs to be 100% right to be trusted if it's something like a financial model. Ai productivity tools that make your IC after you've done all the research or help you think of various sensitivity scenarios etc is probably adopted already.
95% Spot on, which for a blog post is a pretty good place to be!
Can’t internal tech team, few people, do this using AI coding tools? Build something easily that is 100% perfect for them
this is the counter i have heard too, but think this piece articulates why the highest leverage thing for team to do is focus on the outcome and the process to get there.
the tools they leverage likely still continue to be produced by specialists, albeit probably at a much higher level of quality utilizing language models.
The existing provider is best place to do so. They need to develop and improve their existing software to stay relevant and to reduce the desire for the purchaser to go it alone.
The AI threat might be deflationary but if the existing can execute well enough they should win.
We will eventually reach a word where SI can do all this. But we don't know what page we are on or how many pages are in the book before we reach the final chapter.