Agentic Coding Hub
This cluster covers the practice of building software with AI coding agents — the workflows, disciplines, and tooling decisions that separate effective agent-assisted engineering from expensive autocomplete. It is the applied counterpart to the AgenticAiHub cluster, which covers agent architectures, planning, and tool-use theory in general.
The organizing idea of this cluster: the frontier models have largely converged for coding; the workflow around the model is now where the leverage is. Two developers with identical subscriptions can differ by an order of magnitude in output quality and cost depending on how they specify work, manage context, verify results, and parallelize.
Where to Start
- AgenticCodingMaturityModel — locate your current practice on a five-level ladder from autocomplete to orchestrated agent fleets, and see exactly what unlocks the next level.
- CodingAgentLandscape — the August 2026 snapshot of tools (CLI, IDE, cloud, review agents) and frontier models, and how to assemble a stack.
Core Workflow Disciplines
- SpecDrivenDevelopment — the requirements → design → tasks artifact chain that replaced vibe coding; Spec Kit, Kiro, EARS, and lightweight spec discipline without a framework.
- ContextEngineeringForCodingAgents — instruction files (CLAUDE.md / AGENTS.md), progressive disclosure, compaction, memory, and defeating context rot.
- VerificationLoopsForAgenticCoding — test-driven agent loops, reward hacking and test-gaming, review gates, and why verification is the actual bottleneck.
Scaling Beyond One Agent
Extending Agents
- SkillsVsMcp — the two extension surfaces: skills (procedural knowledge, an open cross-client standard) vs MCP (connectivity), and when to build which.
Knowing the Failure Modes
- AgenticCodingFailureModes — the coding-specific ways agent work goes wrong: fabricated completion, test-gaming, scope creep, prompt injection via repository content, and cost blowouts — with the guardrail for each.