Agentic Coding

ACI-013: The Methodology Desert

Thesis

The agentic coding ecosystem has abundant tools and almost no reusable methodology frameworks. Every major AI company has shipped a coding agent. No one has shipped a framework for working with one over months of real development. The gap between "how to use the tool" and "how to build a practice" is largely unaddressed.

Story

Search for "agentic coding" and you find tools: Claude Code, Cursor, Aider, Windsurf, Copilot, Devin, Cline, Continue, Codex CLI. Each has documentation explaining features, configuration, and basic usage patterns. Anthropic tells you to use CLAUDE.md for project rules. Cursor tells you to use .cursorrules. Aider tells you to use .aider/conventions/.

Search for "agentic coding methodology" and you find almost nothing. No framework tells you how to evolve your CLAUDE.md over time (ACI-002). No guide explains what happens when your agent cross-contaminates between projects (ACI-007). No standard exists for session lifecycle management -- how to start a session, orient the agent, verify its work, finish cleanly, and hand off context to the next session.

The closest things to methodology in the ecosystem are: SWE-bench (a benchmark, not a practice), ChatDev and MetaGPT (multi-agent research frameworks, not practitioner methodology), and individual blog posts describing personal workflows that are useful but not reusable.

Intent's system -- steel threads, work packages, 22 skills covering session lifecycle, 5 specialized subagents, autopsy for correction mining, TCA for codebase auditing, the Highlander Rule -- is probably the most developed agentic development methodology that exists as a reusable framework. It reached this point because almost nobody else is building reusable frameworks for this problem. The ecosystem optimizes tool capabilities while assuming practitioners will figure out the practice on their own.

The result: most developers stay at the autocomplete level and never progress to agentic workflows. Enterprise teams deploy tools without methodology and measure the wrong things.

Evidence

What Exists

Category Examples Methodology content
Coding agents Claude Code, Cursor, Aider, Windsurf, Copilot Tool documentation only
Benchmarks SWE-bench, Aider leaderboard Measures capability, not practice
Multi-agent research ChatDev, MetaGPT, OpenHands Agent-agent frameworks, not human-agent methodology
Blog posts Thorsten Ball, Simon Willison, swyx Individual workflows, not reusable frameworks
Enterprise guidance McKinsey, ThoughtWorks Radar Adoption patterns, not development methodology

What Does NOT Exist (externally)

  • Standard for evolving CLAUDE.md / project rules over time
  • Session lifecycle framework (orient, plan, execute, verify, finish, handoff)
  • Correction mining or learning from agent mistakes
  • Verification methodology beyond "run the tests"
  • Debugging methodology specific to agent-generated code
  • Framework for multi-repo, long-duration agentic development
  • Standard for when to delegate vs. when to direct-code

What Intent Provides (filling the gap)

Gap Intent solution Evidence
Session lifecycle 7 lifecycle skills (in-start through in-handoff) 22 skills in production
Rule evolution CLAUDE.md as living architecture (ACI-002) 28 edits over 10 months
Correction learning Autopsy script (889 lines, Lens 3 reuse) Used across 3 repos
Code audit TCA v3.0 (5 skills, 3 scripts) ST0028
Architecture governance Highlander Rule + MODULES.md ST0025, 25 violations found

Application

  • When evaluating AI coding tools, evaluate the methodology ecosystem around the tool, not just the tool itself. A tool with mature methodology (Claude Code + Intent skills) delivers more value than a more capable tool with no methodology (hypothetical perfect agent + no session management).
  • When building an AI-assisted development practice, budget as much time for methodology as for tool selection. The tool choice matters less than you think. The practice matters more.
  • Look for practitioners who share reusable frameworks, not just anecdotes. A blog post saying "I used AI to build X" is useful but not transferable. A framework saying "here is how to structure sessions, verify output, and learn from mistakes" is transferable.

Anti-Pattern

"Tool Worship": An organization spends six months evaluating whether to use Copilot, Cursor, or Claude Code. They run benchmarks, compare completion quality, measure token costs, and read every blog post comparing features. They finally choose a tool and deploy it to 200 developers. Six months later, usage data shows that 80% of developers use it as autocomplete and 20% have stopped using it entirely. The tool selection was meticulous. The methodology was non-existent. No one taught developers how to move from autocomplete to agentic workflows, how to verify agent output, or how to build durable project rules. The tool is capable; the practice is absent.