Writing
Articles
Field notes on intent-oriented development: gating AI-written code against intent with verify-diff, cross-language conformance, spec-driven workflows, requirements as code, and intent drift.
What the Stateless MCP 2026 Spec Means for Intent Verification Loops in Agent Workflows
The stateless MCP 2026 spec changes how coding agents communicate. Here is what that means for verify-diff loops and spec-gated AI code.
The Silent Spec Fork: When Every Test Passes But Intent Has Already Broken
Passing tests prove code behaves correctly, not that it still satisfies the original intent. Here's why that gap is your next production risk.
Spec-First Development vs Vibe Coding: An Honest Comparison
Vibe coding ships faster on day one and gets more expensive every day after. An honest look at when spec-first development actually pays for itself.
What Dark Souls Boss Environments Teach Us About Spec-Driven Agent Verification
The DSLE research frames game rules as verifiable constraints. Here is what that architecture means for intent-driven development and AI agent verification.
Writing Agent Instructions That Do Not Rot as the Codebase Changes
Most CLAUDE.md and agent-instruction files are stale within a month. The structural choices that keep agent instructions accurate without constant manual upkeep.
Audit Trails That Rot: Why Spec Proof Must Stay Live, Not Archived
Static compliance snapshots go stale the moment code moves. Here's why durable, living proof artifacts are the only audit trail worth keeping.
Resonance Is Not Verification: Why Hallucination Fixes Without a Spec Gate Are Still Just Vibes
Claiming AI 'resonance' kills hallucinations sounds compelling. Here's why without a machine-verifiable spec, you're trusting feeling over proof.
Intent Debt: The Silent Accumulation No Test Suite Catches
Every codebase drifts from its original intent. Intent debt is what you owe when that drift goes unmeasured and unacknowledged.
Requirements as Code: The Real Tradeoffs Nobody Mentions
Encoding requirements as executable, checkable artifacts sounds free. It is not. The genuine costs and where the tradeoff still wins for an AI-assisted team.
One Decision, Four Languages, One Source of Truth
How conform --all-targets keeps a single decision consistent across TypeScript, Python, C#, and Java, walked through a real feature-flag case study.
Give Your Coding Agent a Gate It Cannot Skip
Wire the MCP verify-real-code loop (intent_verify_diff, intent_prove) so a coding agent has to pass a deterministic gate before its change can land.
What Is Intent-Driven Development, and Why Prompts Alone Cannot Replace a Spec
Prompts describe what you want once. A spec keeps describing it as the codebase changes. The distinction that matters as AI writes more of the code.
The Spec Versioning Gap: Why AI Agents Silently Inherit Stale Intent
When specs evolve but agents aren't told, code diverges from intent without a single failed test. Here's why versioned intent contracts matter.
DeepSWE Scores High on SWE-Bench. That Tells You Nothing About Whether the Code Matches Your Spec.
DeepSWE tops leaderboards, but pass rate benchmarks can't detect intent drift. Here's what verified conformance actually requires.
Your Spec Is the Only Audit Trail That Doesn't Lie
Code comments drift, tickets close, and commit messages decay. Here's why your machine-verifiable spec is the only honest record of intent your team has.
DeepSWE Tops the Leaderboard, But Does It Verify Anything That Actually Matters?
DeepSWE scores 59% on SWE-bench, but benchmark pass rates say nothing about whether AI code matches declared intent. Here's the gap that matters.
When You Mythologize AI Agents, Your Specs Become Wishes: Operating Intent-Driven Systems Clearly
Treating AI agents as autonomous minds undermines spec-driven development. Here's how to write machine-verifiable contracts that survive the mythology.
DeepSWE Gets Benchmarks Wrong: Why Pass Rate on Tasks Is Not the Same as Conformance to Intent
DeepSWE measures if AI agents fix bugs. ThunderLang asks a harder question: did the agent honor the declared intent? Here's why that gap matters.
Spec Rot: Why Your Intent Document Expires Before Your Code Does
Spec rot is the quiet divergence between what you meant to build and what AI agents actually built. Here's how to detect and stop it.
Why SWE-Bench Scores Tell You Nothing About Whether AI Code Matches Your Spec
DeepSWE tops SWE-Bench, but benchmark scores ignore intent conformance. Here's what actually matters when AI agents write production code.
Refactoring Shared Code With AI Breaks Contracts You Forgot You Made
AI agents refactor confidently across shared modules, but cross-service behavioral contracts have no voice. Here's why that silence is the real risk.