Design Your Solo SaaS for Operational Continuity, Not Faster Code
AI speeds development, but solo SaaS fails on coverage. Design for migrations, incidents, succession, and a bus factor greater than one.
hemju — calm notes on software and AI
Personal notes on software architecture and applied AI, with a focus on reliability, responsibility, and long-term operability.
Latest posts
Do Not Run Bedrock Guardrail Checks on Every Agent Turn
AWS's resourceless InvokeGuardrailChecks API moves guardrail policy into code. Why blanket per-turn checks break your quota and miss real trust boundaries.
The METR Study Flipping Sign Is Not Permission to Raise Your Velocity Targets
A sign-flipping AI productivity study cannot justify higher velocity targets; CTOs should measure review pressure, failures, and recovery instead.
Pin Your GenAI Telemetry Schema or Your LLM Release Gates Will Lie
Why OpenTelemetry's GenAI conventions are an unstable upstream dependency, and how to pin, normalize, and version-test them so a renamed attribute can't quietly approve a...
Stop Giving Your Best Engineer Both the Architecture and the Deadline
Splitting deadline accountability from architecture authority can surface costly tradeoffs before they harden into hidden technical debt.
Use Aurora DSQL CDC to Transport an Outbox, Not to Invent Domain Events
Aurora DSQL's managed CDC removes the connector, not the outbox. Why intentional, versioned outbox rows beat exporting raw row mutations as domain events.
Your RAG Pipeline Doesn't Need a Vector Database, It Needs a Cache Invalidation Strategy
Why production RAG failures are almost always cache-invalidation and versioning problems—not embedding choices—and how to design chunk identity, tombstones, and staleness windows.
Your Domain Model Is a Political Document, Not an Engineering One
Why bounded contexts rot into distributed monoliths: the boundary that matters is data ownership and change cadence, not the org chart you quietly drew it around.
Localization Is a Cache Invalidation Problem, Not a Translation Problem
At production scale, localization breaks as stale reads across caches, not bad translations. Treat locale strings as versioned distributed state, not static assets.
PromptOps: Configuration Management for Probabilistic Systems
Stop hardcoding prompts. Learn how PromptOps brings versioning, automated testing, and observability to your AI infrastructure in 2026.
The Prompt Engineering Playbook: From Magic Spells to Prompt-as-Code
In 2026, 'Vibe Coding' is out. Learn how to treat prompts as code with structured frameworks, version control, and model-specific logic.
The AI OSPO: Governance for the Open Source Era
As AI models proliferate, companies need governance beyond code. Learn why an AI OSPO is becoming essential for CTOs and engineering leaders.
From Prompts to Agents: A Holistic View of System Design
Stop designing prompts; start designing agents. Explore the 2026 architectural shift from conversational AI to autonomous, self-correcting agentic workflows.
Sovereign AI: Why Open Source Is a Compliance Requirement
In the era of the EU AI Act and DORA, 'Sovereign AI' is no longer optional. Learn why open-source models are essential for regulatory defensibility.
Open Source vs. Proprietary: The 2026 Strategic Hybrid
Stop choosing sides. In 2026, the winning AI strategy isn't Open Source vs. Proprietary—it's building a hybrid architecture that blends the best of both.
The Holistic AI Audit: Assessing 360° Readiness
AI readiness isn't a technical checklist; it's an organizational stress test. Learn how to audit your data, infrastructure, and culture for 2026.
When NOT to Use a Vector Database
Vector databases are the cargo cult of 2026. Learn when to skip the complexity and stick with simpler, faster, and cheaper architectural alternatives.
The 2026 CTO: From Code Overseer to Inference Architect
The CTO role has shifted from managing code quality to architecting inference. Discover the new LAMA stack, critical AI KPIs, and leadership strategies for 2026.
The Cultural Hallucination: Why LLMs Struggle Globally
LLMs speak 100+ languages but understand zero cultures. Learn how to architect AI systems that respect global intent and avoid the 'WEIRD' data bias.
State vs. Statistically Likely: Mixing Databases with LLMs
Modern systems are hybrids of deterministic databases and probabilistic LLMs. Learn the architectural patterns to keep your data integrity intact in 2026.
Beyond the Chatbot: Finding the Invisible AI ROI
Stop wasting AI budgets on flashy chatbots. Learn how to find the 'invisible' ROI in backend infrastructure, data hygiene, and engineering velocity.
AI Adoption Without Regret: A Decision Framework for CTOs and Founders
A pragmatic framework for CTOs and founders to evaluate AI initiatives. Move past the hype and make adoption decisions based on risk, ROI, and long-term health.
Where LLMs Break: Architectural Patterns for Safe AI Integration
Stop treating LLMs like deterministic utilities. Learn the architectural patterns for building resilient, production-ready AI systems that fail gracefully.
Architecting for Uncertainty: The New Paradigm of Software Design in the AI Era
As AI moves into the core of the stack, determinism is fading. Learn how to architect resilient systems that treat uncertainty as a first-class property.