Watch
Briefings, walkthroughs, and spoken versions of the work.
The video layer of the site. Same ideas, same engineering lens, delivered in a format better suited to walkthroughs and explanations.
Why this page exists
Watch is for ideas that need to be walked through, not just written down.
Some arguments land best in text. Others are easier to understand when the stack, the diagram, or the operating model is explained out loud. This page is for those pieces.
What to expect
Briefings
Short, direct explanations of what is changing in enterprise AI and why it matters operationally.
Walkthroughs
Architecture-led explanations that work better spoken than written: diagrams, flows, and layered models.
Companion pieces
Videos that extend or translate the writing, rather than duplicating it or acting like a separate content universe.
Latest briefing
0:00 Featured
OpenAI's Deployment Company Is a Lock-In Machine, Not a Consulting Firm
OpenAI's $4B Deployment Company isn't consulting, it's the most aggressive vertical integration play in enterprise AI history. Here's what architects need to do before it's too late.
Archive
More briefings from the same body of work.
Architecture, operations, and execution explained in shorter spoken form.
Why Salesforce's Agent Platform Is Actually Genius Lock-In
Salesforce's Headless 360 is architecturally sound and directionally correct. It's also a masterclass in making your platform the one thing agents can never leave. Here's what enterprise architects need to do about it.
The "Queryable Company" Is a Data Architecture Problem, Not a Startup Superpower
Diana Hu's AI-native company vision is the right destination, but the map runs through enterprise data architecture patterns, event sourcing, semantic layers, data contracts, that the startup world chronically underestimates.
Agent Identity Patterns: Which One to Use, and When
I use a simple rule. Runtime identity for platform work, delegated access for user-owned systems, hybrid when one workflow crosses both, and shared keys as the shortcut to avoid.
Decision Automation Needs Three Layers, Not One
Enterprise decision automation works best when deterministic logic, probabilistic signals, and language reasoning each do the part of the job they are actually good at.
Why MCP Is a Trust-Boundary Problem
MCP makes tool use easier, but that does not make it neutral. Once an agent can discover tools, call systems, and move data across boundaries, access design becomes the real problem.
What Every Agent Runtime Should Share
Teams often build their first agents one by one, then discover too late that every flow needs the same operational scaffolding. The runtime is where that shared foundation belongs.
Why MCP Tool Access Does Not Replace Runtime Identity
I think of these as separate control layers. Runtime identity defines the workload, MCP defines the tool surface, delegated access handles user-owned systems, and target systems still enforce their own permissions.
Stop Calling Every Workflow an Agent
A workflow that calls an LLM is not automatically an agent. The term should mean something operationally real, not just technically fashionable.
Stop Calling Every Workflow an Agent
A workflow that calls an LLM is not automatically an agent. The term should mean something operationally real, not just technically fashionable.
Google Just Shipped the Agent Mesh. Most People Missed It.
Google's Gemini Enterprise Agent Platform isn't a rebrand, it's the first agent mesh for enterprise AI. Agent Identity, Registry, and Gateway solve the governance gap keeping agents out of production. The real story isn't the model. It's the infrastructure.
Inside the AI-Ready Enterprise Stack
A layer-by-layer walkthrough of the stack that has to exist before enterprise AI can do real work.
Designing a Reusable Agent Runtime
The shared scaffold for triggers, reasoning, validation, audit logging, and operational recovery.
Why Governance Breaks Before the Model Does
Identity, approvals, audit trails, and access boundaries are usually what collapse first in production.