Glad to see that the fractionalization and chaos of vibe code builders, ai workflows, business automation and agentic solutions is being addressed.
I am seeing where the common business owner or entrepreneur is commenting profusely to download all these, for example – N8N scenarios, but we all know they'd never get touched or even if they attempt it's just too confusing. It defines a clear gap between seeing and understanding the solution, yet the final outcome and result you want to experience is so very very far away.
McDonald's provided a similar solution; you could see and feel that you want to satiate your hunger, but you didn't want to go to the store, get the ingredients and cook a hamburger.
You can download a .json file, but then what?
So to see all this is encouraging for where we want to take Vibecodecompany.com – the go to company to simplify all things convoluted in the spaghetti bowl of nodes and connectors.
We are also open to creating a library for any developers who are building SAAS, unique work flows, or more in-depth apps that are solving real world problems.
Thank you for this article and the awareness; it's a solid compass for any business owner, manager or entrepreneur to review as we move further into this next exciting adventure of the agentic era in 2026.
The takes on automation in cybersecurity and unstructured data are directionally correct. Arguably the most interesting things about both cases is that the existing incumbents are offering capabilities that address those needs, but either the customers are not able to implement them properly and translate into value, or the tools are a bit of a vaporware.
The obvious winners will be those that "just work" and you can't argue whether or not they deliver value.
Jennifer’s framing is spot on. The constraint is no longer model capability, it’s data entropy.
What we’re seeing in live enterprise environments is that RAG and agentic workflows don’t fail because of reasoning. They fail because the underlying unstructured data is incoherent, stale, duplicated, or ungoverned.
At SageX, we’ve learned the hard way that one-time cleanup doesn’t work. The only systems that deliver durable value behave like a continuous data refinery: ingesting multimodal inputs, extracting context-aware structure, reconciling conflicts across sources, and keeping data machine-legible over time.
The real shift isn’t “better embeddings”, it’s treating unstructured data as infrastructure, not preprocessing.
Strong signal from a16z that this layer is becoming foundational rather than optional.
Glad to see multimodal data showing up here!! We’re working on the problem of customer understanding for product teams using mixed methods research analysis.
This resonates deeply with what we're building at Quantum Mosaic — a judgment OS for institutional decisions.
@Sarah Wang nails it: the real disruption isn't AI replacing systems of record, it's systems of record evolving into systems that anticipate, coordinate, and execute end-to-end workflows.
In private capital, the scarce asset isn't data — it's judgment. The missing layer is a system that sits in the commit path, captures full decision traces (context → deliberation → outcome), and governs both humans and AI agents with the same rails.
That "judgment OS" feels like the counterpart to the data and agent infrastructure described here — and where the next generation of trust and network effects will be built.
(cc @Jennifer Li — your point on unstructured data as "the limiting factor for AI" is exactly what we're seeing in institutional decision-making too)
great read, got me to thinking how i can think from 2036 perspective and create the legal vernacular RAG we are building (I am CTO at vaquill.com, building RAG)
This was a wonderful read. Appreciate you all sharing.
A couple of thoughts, if I may.
Malika Aubakirova's section on agent-native infrastructure highlights the way future network traffic will differ (and grow) but misses the far larger impact it will have on the economy as a whole. Businesses of all types are going to have to re-architect in order to respond to agent-driven purchasing. The speed of business is going to increase.
Regarding the infrastructure, Malika may be right about the 2026 timeline. The broader impact will take a bit longer but will be right at its heels.
Sarah Wang and Seema Amble are highlighting what is the most exciting thing about AI for me, which is pushing back-office work back a layer. I'm thinking about it as "CI Everywhere". The idea of a business fundamentally being something that responds to a request and those requests trigger workflows. Moving to those being orchestrated and calling out to people when a decision or action is needed (rather than people triggering workflows that call AIs) ultimately will result in some businesses that move far faster than others. It won't be all workflows but it will be a lot.
I still think that workflows orchestrated outside of AI are the way to go for this type of work due to the more deterministic nature (shorter runtime for each step, pre-defined context and inputs, ability to test output at each step) but I am getting prepared for a future state where that may all be subsumed by a core model. (That settles transactions internally?)
I liked Jon Lai's vision for story building. Attributable, verifiable contributions so that the the users adding to the product/story are paid fair value for what they contributed. Can be done anonymously via web3 tech. Doesn't have to be super high tech stuff either. We should have platforms for just old school story building where any contributor is paid. The Mad Libs economy.
I think what Jon Lai seems to be describing is the metaverse. And, as a parent that is convinced that the metaverse is here in the form of Roblox, I wonder if a Roblox Creator Tools MCP server would fundamentally fit his vision.
Roblox is interconnected universes across genres (two new ones I saw this week were a GTA clone with ATM smashing and virtually working at a Raising Canes restaurant) where creators earn income crafting assets.
Couldn't agree more with what Jennifer is saying about untangling data. I agree with everything written but I also would argue that human understanding of systems (and our place in those systems) needs to change in addition to the technical solutions landing for untangling that data.
If you're the type of person that doesn't question why you're copying and pasting the same type of data or object from one system in one tab to another system in another tab, then those are the people that need to be educated first on object oriented design and systems level thinking.
Then RAG becomes more reliable. But I would argue that human change management needs to happen as well for Jennifer's thesis to play out. And I hope it does.
Glad to see that the fractionalization and chaos of vibe code builders, ai workflows, business automation and agentic solutions is being addressed.
I am seeing where the common business owner or entrepreneur is commenting profusely to download all these, for example – N8N scenarios, but we all know they'd never get touched or even if they attempt it's just too confusing. It defines a clear gap between seeing and understanding the solution, yet the final outcome and result you want to experience is so very very far away.
McDonald's provided a similar solution; you could see and feel that you want to satiate your hunger, but you didn't want to go to the store, get the ingredients and cook a hamburger.
You can download a .json file, but then what?
So to see all this is encouraging for where we want to take Vibecodecompany.com – the go to company to simplify all things convoluted in the spaghetti bowl of nodes and connectors.
We are also open to creating a library for any developers who are building SAAS, unique work flows, or more in-depth apps that are solving real world problems.
Thank you for this article and the awareness; it's a solid compass for any business owner, manager or entrepreneur to review as we move further into this next exciting adventure of the agentic era in 2026.
The takes on automation in cybersecurity and unstructured data are directionally correct. Arguably the most interesting things about both cases is that the existing incumbents are offering capabilities that address those needs, but either the customers are not able to implement them properly and translate into value, or the tools are a bit of a vaporware.
The obvious winners will be those that "just work" and you can't argue whether or not they deliver value.
This is Pretty awesome. And I can definitely resonate with most of them and really excited for the future. Cheers.
Jennifer’s framing is spot on. The constraint is no longer model capability, it’s data entropy.
What we’re seeing in live enterprise environments is that RAG and agentic workflows don’t fail because of reasoning. They fail because the underlying unstructured data is incoherent, stale, duplicated, or ungoverned.
At SageX, we’ve learned the hard way that one-time cleanup doesn’t work. The only systems that deliver durable value behave like a continuous data refinery: ingesting multimodal inputs, extracting context-aware structure, reconciling conflicts across sources, and keeping data machine-legible over time.
The real shift isn’t “better embeddings”, it’s treating unstructured data as infrastructure, not preprocessing.
Strong signal from a16z that this layer is becoming foundational rather than optional.
I think regulated tokenization infrastructure / bringing real world assets onchain is going to be a huge “idea” in 2026 and for the next decade…
Big takeaway:
The biggest companies of the last century won by finding the average consumer.
The biggest companies of the next century will win by finding the individual inside the average.
2026 is the year the world stops optimizing for everyone and starts optimizing for you.
I vote for the untangling of unstructured & multimodal data from pdfs by Jennifer Li .. 🙏
We agree on much, but not everything. https://open.substack.com/pub/patrickhenryapseudonym/p/what-i-think-might-happen-between?r=3x7typ&utm_medium=ios&shareImageVariant=overlay
Glad to see multimodal data showing up here!! We’re working on the problem of customer understanding for product teams using mixed methods research analysis.
This resonates deeply with what we're building at Quantum Mosaic — a judgment OS for institutional decisions.
@Sarah Wang nails it: the real disruption isn't AI replacing systems of record, it's systems of record evolving into systems that anticipate, coordinate, and execute end-to-end workflows.
In private capital, the scarce asset isn't data — it's judgment. The missing layer is a system that sits in the commit path, captures full decision traces (context → deliberation → outcome), and governs both humans and AI agents with the same rails.
That "judgment OS" feels like the counterpart to the data and agent infrastructure described here — and where the next generation of trust and network effects will be built.
(cc @Jennifer Li — your point on unstructured data as "the limiting factor for AI" is exactly what we're seeing in institutional decision-making too)
Love the creating for agents take!
Check out www.datasprint.us and see how we are changing the face of the traditional data warehouse - faster, ingest more and cost less
great read, got me to thinking how i can think from 2036 perspective and create the legal vernacular RAG we are building (I am CTO at vaquill.com, building RAG)
This was a wonderful read. Appreciate you all sharing.
A couple of thoughts, if I may.
Malika Aubakirova's section on agent-native infrastructure highlights the way future network traffic will differ (and grow) but misses the far larger impact it will have on the economy as a whole. Businesses of all types are going to have to re-architect in order to respond to agent-driven purchasing. The speed of business is going to increase.
Regarding the infrastructure, Malika may be right about the 2026 timeline. The broader impact will take a bit longer but will be right at its heels.
Sarah Wang and Seema Amble are highlighting what is the most exciting thing about AI for me, which is pushing back-office work back a layer. I'm thinking about it as "CI Everywhere". The idea of a business fundamentally being something that responds to a request and those requests trigger workflows. Moving to those being orchestrated and calling out to people when a decision or action is needed (rather than people triggering workflows that call AIs) ultimately will result in some businesses that move far faster than others. It won't be all workflows but it will be a lot.
I still think that workflows orchestrated outside of AI are the way to go for this type of work due to the more deterministic nature (shorter runtime for each step, pre-defined context and inputs, ability to test output at each step) but I am getting prepared for a future state where that may all be subsumed by a core model. (That settles transactions internally?)
I liked Jon Lai's vision for story building. Attributable, verifiable contributions so that the the users adding to the product/story are paid fair value for what they contributed. Can be done anonymously via web3 tech. Doesn't have to be super high tech stuff either. We should have platforms for just old school story building where any contributor is paid. The Mad Libs economy.
I think what Jon Lai seems to be describing is the metaverse. And, as a parent that is convinced that the metaverse is here in the form of Roblox, I wonder if a Roblox Creator Tools MCP server would fundamentally fit his vision.
Roblox is interconnected universes across genres (two new ones I saw this week were a GTA clone with ATM smashing and virtually working at a Raising Canes restaurant) where creators earn income crafting assets.
Couldn't agree more with what Jennifer is saying about untangling data. I agree with everything written but I also would argue that human understanding of systems (and our place in those systems) needs to change in addition to the technical solutions landing for untangling that data.
If you're the type of person that doesn't question why you're copying and pasting the same type of data or object from one system in one tab to another system in another tab, then those are the people that need to be educated first on object oriented design and systems level thinking.
Then RAG becomes more reliable. But I would argue that human change management needs to happen as well for Jennifer's thesis to play out. And I hope it does.