The Data Silo Problem Blocking Infrastructure Intelligence

▶ Watch (01:48)

HashiCorp runs eight products (Terraform, Vault, Boundary, Packer, Waypoint, and Vault Radar among them). Each is a system of record. None talks to the others. The largest financial customers manage roughly a billion assets, with tens of billions of relationships between them. Ask a general-purpose LLM how many resources run log4j and it can’t answer; it doesn’t have that context. That gap, between data that exists and intelligence that doesn’t, is what Project Infograph was built to close.

Project Infograph: A Knowledge Graph of Your Entire Infrastructure

▶ Watch (07:12)

Infograph is a graph database of every infrastructure asset, modeled as nodes and edges. A Java application on a VM connects to the server image, which connects to installed packages, which connects to environment details and PCI compliance status. Those permutations reach into the billions at scale. With that graph in place, a generative AI tool doesn’t just suggest rotating an image; it knows which image, why, and how to do it using Packer and Terraform. Infograph also opens an API so third-party tools can pull context directly.

Governance by Design in an Agentic World

▶ Watch (11:15)

Companies that win at scale, ABN Amro, Canva, Providence Health, built governance into their architecture from day one. Most organizations treat it as a second step and pay later. Over the next few years, Bruno Aziza estimates over a billion applications will be built, with roughly a third being agents. Each agent needs an identity, access controls, and auditable outputs. A request can travel through 10 to 15 agent hops before reaching a database, but the access question is still the same traditional one.

“As defenders we have to be right all the time every day.” — Tyler Lynch

“Attackers need to be right once.” — Tyler Lynch

What the IBM Combination Unlocks

▶ Watch (16:09)

When HashiCorp was acquired, IBM Research had 3,000 people. That’s larger than all of HashiCorp at acquisition. That division is now working on problems HashiCorp couldn’t fund: small language models tuned for Terraform, graph transformers to predict missing configuration or right-sizing recommendations from the Infograph dataset, and formal methods to mathematically prove software correctness. IBM’s Z-series hardware delivers up to eight nines of availability on a single device. Vault can now run on it. For customers with high-throughput secret management requirements, that changes the architecture conversation.

Notable Quotes

they’ve built with governance by design. Bruno Aziza · ▶ 11:55

As defenders we have to be right all the time every day. Tyler Lynch · ▶ 14:39

Attackers need to be right once. Tyler Lynch · ▶ 14:41

experimentation’s over. Bruno Aziza · ▶ 25:24

Key Takeaways

  • Project Infograph maps up to a billion assets and their tens-of-billions of relationships so AI tools have real context.
  • Governance built after the fact creates cascading failures; one-third of future apps are expected to be agents.
  • IBM Research’s formal methods and Z-series hardware give HashiCorp Vault up to eight nines of availability.