This post came from a conversation I had at the Porto summit with a CICF member. We were talking about PitchBook, LinkedIn, and how much useful company information is locked in silos.
Here’s the thing. If we want an agentic future, we need company information that is machine-readable, agent-friendly, and split into layers.
We need:
My argument is simple. Websites should expose structured, agent-friendly company data, and then multi-company platforms (MCPs) can build on top of that data.
If you have no token, you only see the public information. If you are trusted and authenticated, you gain access to more.
This is about breaking down PitchBook and LinkedIn-style silos, making information more open, and enabling both public and private MCP layers for company intelligence.
Frankly, we should not have to pay huge sums simply to know who to invest in, or where the information came from. It’s time to rethink how we share and access company data.
If you are leading a business, you can feel the tension.
What I discovered surprised me when I started mapping this out with founders and operators.
The biggest blocker is not the AI model. It is the mess of disconnected systems, inconsistent naming, and unclear ownership of truth.
Michel Tricot put it bluntly:
"The bottleneck for production agents was never retrieval speed. It was whether your agent could discover what data exists across disconnected systems, resolve entities across silos, and actually do something with what it found. RAG solves the easy part. The hard part is everything else."
Most leadership teams still treat “company data” like a marketing output.
That works for humans skimming. It fails for agentic workflows, where software agents need to:
A-Team Insight framed the same challenge from the enterprise angle, asking how firms break down silos and build a semantic layer that makes data usable:
"How can AI help firms break down data silos and integrate data and where are firms on their journey? Unlocking value from unstructured data opens the keys to the kingdom. Beyond storage, what specific AI/ML models are most effective for automatically classifying, tagging, and extracting structured insights from unstructured data at scale? How can GenAI and RAG (Retrieval-Augmented Generation) be used to automatically map, model, and generate the enterprise semantic layer, drastically reducing the manual effort required?"
The leadership takeaway is not “buy more AI”.
It is:
This stuff is genuinely hard, because silos are often organisational.
Rakan Albazaie points to the uncomfortable reality that silos persist due to structure, legacy tech, and reluctance to collaborate:
"Disconnected data and silos in marketing teams are still a thing because of how departments are set up, old tech, and a general reluctance to work together. Even with the promise of AI, these barriers make it tough to share and integrate data seamlessly. To get past this, companies need a shift in mindset, seeing data as a valuable asset."
So if you want your website to become a proper “AI-readable data room”, you are not starting with schema.
You are starting with governance.
A leadership-friendly approach is to design your company information in layers.
This is not about giving away the crown jewels.
It is about reducing friction for the information you already share, and making it usable by both humans and agents.
There is a reasonable question behind the buzz.
Can agents help you operate across silos without forcing a rip-and-replace programme?
Chirag Agrawal argues that agentic AI can reduce manual merging by working across silos as needed:
"Can agentic ai break data silos? Agentic AI can revolutionize data management by autonomously selecting datasets, engaging relevant agents, and generating real-time analytics. This innovation eliminates the need for large teams to manually merge data, significantly saving on computational and storage resources. By tapping into data silos as needed, Agentic AI transforms legacy systems into dynamic analytics powerhouses."
I would add a pragmatic counterpoint.
If you want multi-company platforms (MCPs) to be healthy, we need interoperability.
Otherwise, we repeat the same pattern:
Thilakasiri describes why standards matter for identity, authorisation, and secure interoperability:
"Before modern standards, AI systems existed as isolated silos, often trapped within the confines of single LLM providers and unable to collaborate effectively. Integrating these agents with external tools and APIs relied on insecure, custom connectors. Consequently, critical enterprise tasks, such as secure payments or resource management, remained manual and error-prone because agents lacked verifiable identity and secure authorization. This fragmented environment made scaling, building trust, and enabling truly autonomous enterprise action virtually impossible, highlighting an urgent need for a standardized, secure communication layer. Modern standards are now shaping a connected, interoperable AI ecosystem: MCP standardizes secure connections to tools and data; A2A enables agents to discover and collaborate with each other; and AP2 adds secure, auditable payment capabilities. Together, these standards pave the way for true AI synergy."
https://www.linkedin.com/pulse/from-silos-synergy-how-open-standards-shaping-ai-thilakasiri-vkf7c
Whether you agree with every detail or not, the direction is clear.
If you want to move from “website as brochure” to “website as machine-readable data room”, here is a low-drama way to start.
1) Decide Your Public Data Contract
2) Define The Trusted Layer Access Rules
3) Add Provenance As A First-Class Requirement
4) Pick One Agentic Workflow To Prove Value
Choose something that is boring but frequent, like:
Then measure:
Platforms have value. Distribution matters. Network effects are real.
But it is risky when basic company intelligence becomes paywalled, unverifiable, or hard to reuse.
If we want an agentic future that benefits more than a handful of gatekeepers, we need:
If you are a leader reading this, your next step is simple.
Pick one slice of company information and make it agent-friendly, end to end. Then iterate.