Build Log

This site is a live experiment. Everything here was built in the open — the infrastructure, the design, the content strategy. This log captures what was built, when, and why.

2026-05-05 — Day one

Domain registered: artificialcuriositylabs.dev via Route 53. $17/year.

AWS personal account set up:

• Root MFA enabled, root access keys deleted
• IAM admin user created for day-to-day access
• CloudTrail (multi-region) enabled
• IAM Access Analyzer enabled
• Monthly budget with email alerts

Static site stack deployed:

• Astro Paper theme — dark mode default, light/dark toggle
• S3 bucket — private, no public access
• CloudFront distribution with OAC — only CloudFront can read S3
• ACM SSL cert — HTTPS enforced, HTTP redirects to HTTPS
• Route 53 A records — apex + www both pointing to CloudFront
• llms.txt and llms-full.txt — AI-readable content layer, auto-generated at build time

Email forwarding:

• kumar@artificialcuriositylabs.dev forwards to personal Gmail
• SES inbound with Lambda forwarder — reply-to preserved so Gmail replies look native
• DKIM records added to Route 53

Homepage redesigned:

• Custom hero with thesis statement
• “Now” section — 3 cards showing what I’m writing, building, working on
• Dark as default, orange accent

2026-03-13 — Published: “How Many Claude Code Sessions Can — and Should — You Run Simultaneously?” — can vs should distinction; OS memory limits (JETSAM), practical concurrency ceiling for knowledge work

2026-03-13 — Published: “How Many Claude Code Sessions Can — and Should — You Run Simultaneously?” — can vs should distinction; OS memory limits (JETSAM), practical concurrency ceiling for knowledge work

2026-03-13 — Published: “Claude Code Has Native OpenTelemetry. Almost Nobody Knows.” — native OTel SDK in Claude Code, Docker Compose stack with Prometheus + Grafana, Delta vs Cumulative gotcha

2026-03-15 — Published: “Why AI Coding Tools Are Getting Cheaper: Prompt Caching Explained” — how prompt caching works, 5-minute TTL mechanics, cache hit rates in coding workflows, economics of context reuse

2026-03-15 — Published: “The Right Model for the Right Job” — model selection as architecture; right-sizing agents in multi-step pipelines; cost vs capability tradeoffs

2026-04-23 — Published: “Builder Is an Operating Mode, Not a Job Title” — thesis post: builder is an orientation you adopt, not a credential you earn

2026-04-25 — Published: “Creation Collapsed. Distribution Is the New Bottleneck.” — AI eliminated creation friction; distribution to the right people and agents is now the real constraint

2026-04-29 — Published: “Every AI Tool Encodes a Cognitive Mode” — tool friction is a mode-mismatch problem, not a prompting problem; matching tool to cognitive load

2026-05-05 — Published: “How This Site Is Built” — stack walkthrough: Astro + S3/CloudFront/OAC + SES email forwarding + llms.txt content layer

2026-05-05 — Published: “Methodology Is Infrastructure” — methodology should be versioned, tested, and shared like code; how 80+ encoded workflows turn individual learning into institutional knowledge

2026-05-07 — Published: “Four Things That Remain Irreducibly Human” — conviction, relationships, curiosity, and high-stakes judgment: what agents genuinely can’t do

2026-05-16 — Published: “Failure Modes Are the Fingerprint” — anti-patterns are more valuable than patterns; encoding what goes wrong turns individual mistakes into institutional knowledge

2026-04-26 — Published: “What Is an Agent — And What Isn’t” — the loop, the three properties, what chatbots/copilots/workflows are not, the autonomy spectrum

2026-04-26 — Published: “What Is a Skill — Why Methodology Resets Every Session Without One” — the reset problem, 8-phase lifecycle, separation principle, skills as institutional knowledge

2026-04-26 — Published: “What Is a Tool — The API Call Your Agent Makes on Your Behalf” — tools vs MCP servers, tool categories, MCP protocol, 250-tools-one-conversation

2026-04-26 — Published: “What Is an Agent Harness — The Infrastructure That Makes Agents Actually Work” — Agent = Model + Harness, five harness architectures, harness engineering as a discipline

2026-05-06 — Published: “llms.txt — Making Your Site Navigable by Agents” — the llms.txt convention, llms-full.txt as complete corpus, build-time generation in Astro, dual-consumption architecture

2026-04-27 — Published: “Skills as Institutional Memory” — individual craft doesn’t compound without distribution; the architecture problem behind knowledge that stays on one person’s laptop

2026-04-29 — Published: “Proof Replaces Persuasion” — AI collapsed the cost of building evidence; stop writing alignment docs, start showing working things

2026-04-30 — Published: “Your AI Agent Needs Communication Modes, Not a Voice Clone” — six communication registers, mode-specific profiles (engrams) vs voice cloning

2026-05-01 — Published: “Cloud Sync Is the Wrong Storage Layer for AI Agents” — OneDrive/Dropbox are eventually consistent; object storage is the correct primitive for agent writes

2026-05-03 — Published: “ACP Is the Bridge Between Knowledge Work and Build Work” — Agent Client Protocol connects orchestrators to coding agents; knowledge agent delegates build tasks without breaking flow

2026-04-26 — Published: “What the Engram Builder Gives You — And What You Have to Add” — QD’s native builder produces one flat profile from your message corpus; here’s what was missing and how it was extended into six mode-specific profiles with auto-detection

2026-04-26 — Published: “Anti-Patterns Are the Fingerprint” — corrections encode decisions, not patterns; the never_say list is more distinctive than writing samples; the correction loop compounds over time

2026-04-26 — Published: “Voice Is a Layer, Not a Setting” — four-layer architecture (mode detection → voice+quality → format → publish) separates concerns so voice stays consistent across all content types

2026-05-09 — Published: “Your Agent Needs Six Voices, Not One” — personal account of building the 6-engram system, the neuroscience origin of the term, the auto-detection mechanism

2026-04-15 — Published: “The Skill Is the Teacher” — naming as pedagogy; five frames (tool, behavior, practice, ritual, role); why gerunds produce compound returns and tool-names produce dependency

2026-04-16 — Published: “Your Agent’s Behavior Is Code — Start Versioning It” — skills committed to a repo travel with the codebase; reviewable, diffable, available on first checkout; how skill quality compounds through PRs the same way libraries do

2026-04-21 — Published: “From Failure Mode to Skill Chain” — one leaked channel ID produces a four-skill chain that self-enforces; the pattern: catch failure → build smallest fix → discover adjacent gaps → weave skills so invariants hold across boundaries

2026-05-13 — Published: “The Enterprise MCP Pattern: Proxy, Aggregate, Host” — 25 public awslabs MCP servers covering cloud infrastructure, databases, and ops tooling; the enterprise pattern for making them usable: proxy to aggregate locally, host behind a gateway so teams get the full catalog without local setup or credential management

What’s next

• Configure social links (LinkedIn, X)
• Update SES receipt rule for kumar@artificialcuriositylabs.dev