Thesis
ChatGPT Astra vs Claude Fable 5.1: They're Not Competing Anymore
Astra and Fable 5.1 have diverged so fundamentally that comparing them on benchmarks misses the point - route by workload, not leaderboard.
Ideas
I write about how enterprise AI gets built.
Thesis pieces, frameworks, build notes, and security briefs on the infrastructure, runtimes, and control layers underneath enterprise AI.
26 ideas Search by title, topic, or series.
Latest publication
Archive
Thesis
Stop Hiring AI Engineers. Design Your AI Team First.
CTOs who hire AI talent without a structural blueprint waste 6-12 months on misaligned teams that can't ship to production.
Framework
AI Engineer Career Path: 5 Skills You Actually Need in 2025
The five skills required for an AI engineer career form a dependency stack, not a checklist, and skipping a layer breaks everything above it.
Framework
AI Agent Hallucinations: A Practitioner's Guide to Detection and Fixes
Agent hallucinations are systematic failures with structure, and structure means you can build specific guardrails, evaluation pipelines, and fixes against them.
Framework
Context Layer for AI Agents: The Architecture That Makes Agents Useful
The context layer is the architectural component that separates production AI agents from stateless chatbots, and it deserves more engineering effort than model selection.
Thesis
Large Language Models Explained: ChatGPT Is Math, Not Magic
Understanding that LLMs are prediction machines with no truth checker is the single most important insight for using AI effectively.
Thesis
Agents vs Workflows: The One-Question Framework for Every Build
If the path through the problem is known before execution begins, build a workflow; if the system has to discover the path itself, that's when you reach for an agent.
Thesis
Semantic Layers Beat Model Upgrades for AI Agents
The enterprise AI bottleneck isn't model quality - it's machine-interpretable meaning, and semantic layers are the infrastructure that closes the gap.
Thesis
OpenAI's Deployment Company Is a Lock-In Machine, Not a Consulting Firm
DeployCo solves OpenAI's commoditization problem, not your deployment problem, and the lock-in is by design.
Thesis
Why Salesforce's Agent Platform Is Actually Genius Lock-In
Headless 360 shifts lock-in from the UI to the infrastructure — enterprises adopting it without abstraction layers trade visible constraints for invisible ones.
Thesis
The "Queryable Company" Is a Data Architecture Problem, Not a Startup Superpower
The competitive edge in building AI-native isn't being a startup, it's having intentional data architecture from day one, a discipline most founders lack.
Contrarian
Stop Calling Every Workflow an Agent
Autonomy, recovery, and access boundaries matter. Without them, it is automation wearing new language.
Security Brief
Why MCP Is a Trust-Boundary Problem
The problem is not connector count. It is identity, scope, trust boundaries, and what the runtime is allowed to do.
Build Note
What Every Agent Runtime Should Share
Triggers, validation, recovery, and auditability belong in the scaffold, not re-invented agent by agent.
Thesis
The AI-Ready Enterprise Stack
Most enterprise AI conversations start at the model layer. They should start several layers lower.
Framework
Decision Automation Needs Three Layers, Not One
Rules, ML signals, and LLM reasoning each have different jobs. Treating them as one layer creates brittle systems.
Visual
Why MCP Tool Access Does Not Replace Runtime Identity
MCP can narrow which tools an agent may call, but it does not replace runtime identity, delegated user access, or downstream system permissions.
Framework
Agent Identity Patterns: Which One to Use, and When
The real choice is not whether an agent has credentials. It is which identity pattern fits the ownership boundary around the action.
Thesis
Google Just Shipped the Agent Mesh. Most People Missed It.
The real story in Google's Agent Platform isn't Gemini models, it's three governance primitives that form an agent mesh solving why enterprises can't get agents past POC.
Visual Security Boundaries for Enterprise AI
Why MCP Tool Access Does Not Replace Runtime Identity
MCP can narrow which tools an agent may call, but it does not replace runtime identity, delegated user access, or downstream system permissions.
Framework Security Boundaries for Enterprise AI
Agent Identity Patterns: Which One to Use, and When
The real choice is not whether an agent has credentials. It is which identity pattern fits the ownership boundary around the action.
Thesis The AI-Ready Enterprise Stack
The AI-Ready Enterprise Stack
Most enterprise AI conversations start at the model layer. They should start several layers lower.
Build Note Inside the Agent Runtime
What Every Agent Runtime Should Share
Triggers, validation, recovery, and auditability belong in the scaffold, not re-invented agent by agent.
Security Brief Security Boundaries for Enterprise AI
Why MCP Is a Trust-Boundary Problem
The problem is not connector count. It is identity, scope, trust boundaries, and what the runtime is allowed to do.
Framework
Decision Automation Needs Three Layers, Not One
Rules, ML signals, and LLM reasoning each have different jobs. Treating them as one layer creates brittle systems.
Contrarian
Stop Calling Every Workflow an Agent
Autonomy, recovery, and access boundaries matter. Without them, it is automation wearing new language.
No ideas match the current filters
Try a different topic, format, or search term.