Thought provoking post. A few considerations it generated for me:
1) Even though unstructured data can be stored unstructured, eventually, it will be cheaper and more efficient to store it in a structured way. The API costs and query response of a non-LLM system is going to be lower and faster. Keepings thing unstructured is great when the focus is on product experience exploration, but once we converge on a generally accepted user pattern, the work will begin to make it cheap and fast. It doesn't hurt that the text is converted into numbers for LLM use anyway.
2) I can imagine new vendors will emerge that serve companies that are too small for current incumbents, and some of these vendors will grow into larger companies with their clients (Deel is an example of this pattern). But I wonder if there is truly a use case wedge that enables sales into enterprise clients, today, in the HR stack. Incumbents were investing in AI before ChatGPT, and several helped fund the Gen AI wave. They've been relatively quick to experiment with Gen AI capabilities.
3) In autonomous robotics systems, it seems like 'mixtures' of systems that have more deterministic rules + inferred rules outperform purely learned rules. If this applies, I can envision a short and medium term system that has rules-based along side agentic.
It's just a thought experiment, rather than a definitive prediction.
1. Yes. I can imagine that AI tooling builds the structured data in new ways that suit AI. What we mean by structured data may change.
2. The incumbents have been v resilient. ADP has been around since 1949, and even Workday is 20 years old now. The incumbents have done a better job on AI than they have on cloud adoption, but deel is positioned well.
3. YES. This is what we see with our investments in Techwolf, PeopleReign and DatascaleHR. a blend is a thing.
A while ago I wrote an article, not nearly as erudite as this. But in it I put forward a thought experiment where a transaction was viewed as a (badly explained) quantum wave, and the process of driving the conclusion of the transaction was brought about by the iterative collapse of the wave form into an approved or unapproved state. Essentially I was making the case for doing away with fixed processes and using the ability of complex systems (ML) to find the best route for a transaction to follow based on the constraints placed on the transaction type. (https://lylecooperblog.wordpress.com/2020/01/14/taking-hr-processes-beyond-workflow/)
The rose metaphor is a much better analogy to use, for what it is worth, I think the thorns will always be the stubborn humans who will not give up the 'old ways'.
I am stealing your flower metaphor and love the "grammatical abomination" that is the newest buzzword "Agentic". The thorns on the roses are what I am most worried about as a Payroll and Time enthusiast and when it comes to auditing system of record transactions this could very well be the undoing as governments and other compliance creators and regulators will likely not keep up with the way AI will process data and make decisions. It is an interesting time for sure.
Thought provoking post. A few considerations it generated for me:
1) Even though unstructured data can be stored unstructured, eventually, it will be cheaper and more efficient to store it in a structured way. The API costs and query response of a non-LLM system is going to be lower and faster. Keepings thing unstructured is great when the focus is on product experience exploration, but once we converge on a generally accepted user pattern, the work will begin to make it cheap and fast. It doesn't hurt that the text is converted into numbers for LLM use anyway.
2) I can imagine new vendors will emerge that serve companies that are too small for current incumbents, and some of these vendors will grow into larger companies with their clients (Deel is an example of this pattern). But I wonder if there is truly a use case wedge that enables sales into enterprise clients, today, in the HR stack. Incumbents were investing in AI before ChatGPT, and several helped fund the Gen AI wave. They've been relatively quick to experiment with Gen AI capabilities.
3) In autonomous robotics systems, it seems like 'mixtures' of systems that have more deterministic rules + inferred rules outperform purely learned rules. If this applies, I can envision a short and medium term system that has rules-based along side agentic.
It's just a thought experiment, rather than a definitive prediction.
1. Yes. I can imagine that AI tooling builds the structured data in new ways that suit AI. What we mean by structured data may change.
2. The incumbents have been v resilient. ADP has been around since 1949, and even Workday is 20 years old now. The incumbents have done a better job on AI than they have on cloud adoption, but deel is positioned well.
3. YES. This is what we see with our investments in Techwolf, PeopleReign and DatascaleHR. a blend is a thing.
A while ago I wrote an article, not nearly as erudite as this. But in it I put forward a thought experiment where a transaction was viewed as a (badly explained) quantum wave, and the process of driving the conclusion of the transaction was brought about by the iterative collapse of the wave form into an approved or unapproved state. Essentially I was making the case for doing away with fixed processes and using the ability of complex systems (ML) to find the best route for a transaction to follow based on the constraints placed on the transaction type. (https://lylecooperblog.wordpress.com/2020/01/14/taking-hr-processes-beyond-workflow/)
The rose metaphor is a much better analogy to use, for what it is worth, I think the thorns will always be the stubborn humans who will not give up the 'old ways'.
I am stealing your flower metaphor and love the "grammatical abomination" that is the newest buzzword "Agentic". The thorns on the roses are what I am most worried about as a Payroll and Time enthusiast and when it comes to auditing system of record transactions this could very well be the undoing as governments and other compliance creators and regulators will likely not keep up with the way AI will process data and make decisions. It is an interesting time for sure.