My Thinking: Architecture, Software & Operations
The tested engineering principles, mental models, and operating beliefs I use every day to build resilient software, eliminate cloud waste, and lead high-leverage teams.
Over a decade of designing distributed systems across enterprises like Deloitte, Clean Harbors, PayPal (via eClerx), and CAI, followed by co-founding Fraoula, has taught me that the hardest problems in software are rarely about writing more code. They are about choosing the right constraints, keeping architectures simple enough for humans to understand, and respecting the underlying physical hardware.
Physics always wins over software abstractions. Memory is millions of times faster than disk.
Every line of code, database query, and background worker has a recurring monthly dollar cost.
Ten autonomous builders who communicate in plain English beat 100 engineers bogged down in meetings.
If you cannot explain an architecture in plain 5th-grade English, you do not understand it yet.
Foundational Systems & Software Principles
Respect the Computer Memory Hierarchy
CPU cache (L1/L2) takes nanoseconds. Main memory (RAM) takes ~100 nanoseconds. Solid-state disk reads take microseconds. Cross-region network round-trips take hundreds of milliseconds. Most software crawls slowly not because the programming language is slow, but because engineers repeatedly read from disks and make unnecessary network hops. Keep hot working state in RAM.
The Modular Monolith-First Rule
Never build microservices prematurely. A well-organized modular monolith running inside a single container or server process can easily handle tens of thousands of requests per second with zero network latency between services. Microservices add distributed failure modes, network serialization overhead, and operational complexity that small teams struggle to debug.
Ephemeral Data & Zero Disk Leaks
The safest database record is the one that was never written to persistent disk storage. When building compliance or verification software, perform calculations entirely in volatile RAM, check cryptographic proofs, and immediately wipe temporary memory buffers cleanly. This approach eliminates risk under GDPR, HIPAA, and CCPA directives.
Observability Before Optimization
Never guess where software is struggling. Before refactoring a line of code, install distributed tracing, analyze 95th and 99th percentile latency histograms, and inspect query execution plans. Optimization without rigorous measurement is just speculation that often introduces subtle bugs.
Infrastructure Efficiency & FinOps Principles
Zero Idle Compute Policy
Unused server power is a sign of unmeasured code. When teams measure how their software runs and how much memory it actually needs, cloud bills drop dramatically. A server running at 3 percent capacity is cash thrown away. Turn off test environments during nights and weekends.
Mandatory Tagging Governance
Every provisioned virtual machine, storage bucket, and container cluster must have a designated human owner, an environment label, and an expiration date. Enforcing infrastructure-as-code linting prevents zombie databases and unattached storage volumes from silently accumulating millions in waste.
Boring Infrastructure Scaled Well
PostgreSQL, Redis, Linux, and plain HTTP scale reliably to millions of users when configured properly. Avoid adopting proprietary database dialects or complex vendor-specific frameworks that trap your business in closed ecosystems and unpredictable pricing surges.
Engineers Must Understand Unit Economics
Engineers should always know the cost of processing 10,000 requests on their platform. When technical teams understand unit economics, architectural choices automatically balance speed, reliability, and business profitability.
Team Topologies & Operational Leadership
The 10-Person Team Optimum
Communication complexity increases quadratically with every added engineer. A team of 10 people has 45 communication channels; a team of 100 has 4,950 channels. Small, autonomous groups of senior builders make decisions in minutes, ship software daily, and eliminate layers of administrative coordination.
Charge for Outcomes, Not Logins
Per-seat software billing encourages applications to be intentionally complex and slow so employees spend all day clicking buttons. When software charges based on verified work accomplished, the incentive shifts to building quiet, instant, and frictionless tools.
Writing as the Primary Thought Medium
If an engineering proposal cannot be explained in a clear two-page memo using simple 5th-grade English, it is not ready for implementation. Writing forces clarity, exposes logical fallacies, and creates an enduring organizational record that meetings cannot match.
Stoic Calm Under Production Incidents
Outages happen. Network cables get cut. In difficult moments, panic solves nothing. Taking a deep breath, isolating the failure domain, verifying hypotheses with telemetry, and methodically repairing the problem is how world-class engineering teams sustain high availability.
Explore In-Depth Thinking & Case Studies
Follow these direct pathways to read the complete essays, empirical case studies, and research publications behind these principles:
Start Here: Essential Reading
The top 10 foundational essays that define my architectural philosophy.
Open Reading Guide →Saving $12M on Cloud at CAI
A step-by-step account of turning off unused computers to save millions.
Read Case Study →Books That Shaped My Mind
The 10 most influential books on science, engineering, and human nature.
Explore Reading List →Annual Founder Letter
Buffett/Gates-style annual review letter detailing successes and failures.
Read Annual Letter →Apply These Principles to Your Production Systems
I advise technology leaders, CTOs, and founders on eliminating cloud waste and architecting sub-second in-memory systems. You can book a direct 30-minute advisory consultation or commission a complete Speed & Cost Audit.