Insights
On systems, architecture, and the cost of getting it wrong.
Writing for founders, technical leaders, and investors navigating high-stakes architectural decisions.
Mar 2026Mar 2026Feb 2026Feb 2026Feb 2026Feb 2026Feb 2026Feb 2026Feb 2026Jan 2026Jan 2026Jan 2026Jan 2026
What Investors Should Look For in AI Infrastructure
Infrastructure claims in AI pitches are almost always overstated — not from dishonesty, but because demo conditions look nothing like production conditions at scale.
Technical Diligence for AI Startups
Technical diligence for AI startups requires a different frame: not code quality, but whether the system's claims are structurally supportable beyond demo conditions.
Designing AI Systems with Revocation and Oversight
Most AI systems are designed for the success path and cannot be stopped. Revocation is an architectural property that must be designed in — it cannot be retrofitted.
Why AI Governance Is an Architecture Problem
AI governance implemented as policy fails because policies not enforced structurally are not enforceable. Governance lives in boundaries, not in documents.
Control Surfaces in AI Systems
A control surface is any point where human intent can be expressed or enforced. AI systems erode them by default. Designing them deliberately is a governance prerequisite.
The Difference Between Prompt Engineering and System Architecture
Prompt engineering shapes model behavior within a single interaction. Architecture determines what the system can do regardless of how good the prompts are.
When Should a Startup Hire an Architect?
Architects are not a scaling hire. The decisions that determine whether a system can scale are made in the first months — usually before anyone sees the need.
How to Evaluate the Architecture of an AI Startup
You don't need to read the code to evaluate an AI startup's architecture. You need the right diagnostic questions — ones that reveal maturity, not just capability.
The Hidden Complexity of LLM Agent Architectures
Agent architectures look simple on a whiteboard and accumulate complexity faster than any other pattern because the complexity is behavioral, not structural.
Designing Reliable AI Systems in Production
Reliability in AI systems comes from system design, not model accuracy — explicit failure modes, bounded authority, and separation between inference and execution.
Why Most RAG Architectures Break at Scale
RAG failures at scale are retrieval failures misdiagnosed as model failures. The retrieval layer has no contracts, no quality guarantees, and no first-class design.
Common Architecture Mistakes in AI Startups
AI startups fail for the same reasons all software fails: missing boundaries, absent contracts, deferred structure — hidden longer because AI obscures them.
When Your AI Passes the Demo — But Fails the System
The demo works. The integration doesn't. Here's why that gap is architectural, not technical.