Series hub for the Agentic Development field journal — 19 lessons from 23,479 AI coding sessions.
Live site: withagents.dev/posts/post-01-series-launch Field journal entry: withagents.dev/posts/post-01-series-launch
Featured in the Agentic Development Blog series — Post #1: 4,500 AI Coding Sessions — What I Learned
- Send date: Mon May 26, 2026
- LinkedIn: link added on send day
- Canonical blog post: https://ai.hack.ski/blog/
- Series hub: agentic-development-guide
A deeply technical blog series on building production software with AI coding agents at scale.
Over 42 days, across 27 projects and 23,479 Claude Code sessions generating 11.6GB of interaction data, I discovered patterns for agentic development that changed how I build software. This repository indexes 14 unique companion repositories — each grounded in real code, real terminal output, and real problem-solving from actual development sessions.
18 posts. 14 working companion repos. Zero fabricated examples.
| Metric | Value |
|---|---|
| Blog posts | 18 |
| Companion repos | 14 unique |
| AI coding sessions | 23,479 |
| Human-initiated | 4,534 |
| Agent-spawned | 18,945 |
| Development days | 42 |
| Projects | 27 |
| Tool invocations | 256,000+ |
| Post | Repo | Description |
|---|---|---|
| 01 | agentic-development-guide |
Meta-repo, series overview |
| 02 | multi-agent-consensus |
Python, 3-agent unanimous gate voting |
| 03 | claude-code-skills-factory |
Skills + validation framework |
| 04 | claude-ios-streaming-bridge |
Swift, 5-layer SSE bridge |
| 05 | claude-code-ios |
Swift, iOS patterns compendium |
| 06 | auto-claude-worktrees |
Python, parallel worktree factory |
| 07 | shannon-framework |
Claude Code plugin, 4-layer enforcement |
| 08 | ralph-loop-patterns |
Python, hat-based execution loops |
| 09 | session-insight-miner |
Python, session log mining pipeline |
| 10 | stitch-design-to-code |
Node.js, design tokens + Stitch MCP |
| 11 | reponexus |
Spec execution framework |
| 12 | claude-mem-architecture |
Python + SQLite, observation store |
| 13 | sequential-thinking-debugging |
Python, 84-step root cause analysis |
| 14 | multi-agent-merge-orchestrator |
Python, 35-worktree conflict-free merge |
| 15 | claude-code-skills-factory |
Skills anatomy + devlog-pipeline |
| 16 | shannon-framework |
Plugin hooks + enforcement |
| 17 | claude-code-monorepo |
TypeScript SDK examples + CLI scripts |
| 18 | claude-code-monorepo |
SDK vs CLI decision framework |
| # | Title | Core Insight |
|---|---|---|
| 01 | Series Launch | 23,479 sessions overview — agents read 80%, write 20% |
| 02 | Multi-Agent Consensus | 3-agent unanimous voting eliminates false confidence |
| 03 | Functional Validation | 642 blocked test files — real UI validation only |
| 04 | iOS Streaming Bridge | 5-layer SSE architecture for SwiftUI |
| 05 | iOS Patterns | 982 sessions of Swift patterns compendium |
| 06 | Parallel Worktrees | 194 agents, zero merge conflicts via file ownership |
| 07 | Prompt Engineering Stack | 7 layers of redundant enforcement |
| 08 | Ralph Orchestrator | Hat-based execution loops with convergence detection |
| 09 | Session Mining | 696MB log pipeline extracting patterns |
| 10 | Stitch Design-to-Code | 21 screens from design tokens to working UI |
| 11 | Spec-Driven Development | YAML specs to multi-agent execution |
| 12 | Cross-Session Memory | SQLite observation store beats amnesia |
| 13 | Sequential Thinking | 84-step debugging chain for root cause |
| 14 | Multi-Agent Merge | 35 worktrees, topological sort, zero conflicts |
| 15 | Skills Anatomy | SKILL.md files as structured agent protocols |
| 16 | Claude Code Plugins | Hooks + MCP enforcement layer |
| 17 | CLI to SDK Evolution | Bash scripts to autonomous builder |
| 18 | SDK vs CLI Framework | Decision framework for agent architecture |
Every companion repo passes platform-appropriate validation:
| Platform | Method | What It Proves |
|---|---|---|
| Swift | swift build |
SPM resolves deps, all source compiles, types align |
| TypeScript | tsc --noEmit |
All imports resolve, types check, no dead references |
| Python | import <module> |
Package structure correct, exports work, no import crashes |
Blog: site-rho-pied.vercel.app
Written by Nick Krzemienski — March 2026.
Built entirely with Claude Code, validated through the same agentic development patterns described in the series.
Every code snippet in this series comes from real production sessions. No fabricated examples, no hypothetical architectures, no made-up metrics.
MIT