Article Relevance Audit — Week of 2026-03-13

[github-actions]

Articles scanned: 35 evergreen | Run date: 2026-03-13 | New since last audit: 2 articles

Items marked ⚠️ Persistent appeared in the 2026-03-11 audit and remain unresolved.

---

🔗 Link Health

Articles scanned: 35 | Internal links checked: 94 | Issues found: 1

Broken Internal Links

Article	Link Text	Broken URL	Correct URL
agentic-ai-transforming-dev-teams	"GitHub Copilot CLI worktrees"	/articles/copilot-fleet-mode-video-pipeline	/articles/video-pipeline-with-fleet-mode

⚠️ Persistent — flagged in 2026-03-11 audit, not yet fixed. Located in ## The Circle Replaces the Hierarchy section, line 31.

Fix: In src/content/articles/agentic-ai-transforming-dev-teams.mdx, change:

[GitHub Copilot CLI worktrees](/articles/copilot-fleet-mode-video-pipeline)
```
to:
```
[GitHub Copilot CLI worktrees](/articles/video-pipeline-with-fleet-mode)

External Links

Unable to verify all 522 external links due to sandbox network restrictions. One URL returned a fetch error (possible timeout/blocking — not confirmed 404):

Article	URL	Status
agent-harnesses-controlling-ai-agents-2026	`(www.anthropic.com/redacted)	Unreachable (fetch failed)

Note: Image paths using /articles/*.png in github-agentic-workflows-hands-on-guide and github-copilot-visual-studio-guide resolve to confirmed files in public/articles/ — not broken.

---

📊 Factual Freshness

Articles scanned: 35 | Stale references found: 5

⚠️ Persistent — all items below appeared in the 2026-03-11 audit.

Potentially Stale References

Article	What It Says	Current Reality	Severity	Suggested Update
ai-fixes-its-own-bugs	"SWE-agent solves 12.47% of real-world bugs on the SWE-bench evaluation set" (from NeurIPS 2024 paper)	Top models now solve 50%+ on SWE-bench verified subset; the field has moved dramatically since the NeurIPS 2024 baseline	🔴 High	Add context: "Since the NeurIPS 2024 paper, the field has advanced rapidly — top models on SWE-bench now exceed 50% resolution rates." The 12.47% is historically accurate but reads as the current state.
ai-fixes-its-own-bugs	"combines CodeQL's semantic analysis engine with GPT-4o"	GitHub's 2024 blog post describes GPT-4o; the underlying model for CodeQL Autofix has likely been updated since Feb 2024	🔴 High	Verify current model with GitHub's latest Autofix docs; update to "Copilot" or current model name if GPT-4o is no longer accurate.
copilot-developer-fulfillment	"With 20 million GitHub Copilot users..."	GitHub surpassed 30M+ Copilot users by late 2025/early 2026	🟡 Medium	Update to reflect current user count with a citation to GitHub's latest figures
copilot-developer-fulfillment	References "the 2024 DORA report"	2025 DORA report is now available — note: stanford-study-ai-roi-in-engineering already references the 2025 version	🟡 Medium	Update citation to 2025 DORA report; the findings directionally align but the newer report includes AI-specific data
choosing-the-right-ai-sdk	Table shows GitHub Copilot SDK | Technical Preview (Jan 2026)	~7 weeks have passed since the Technical Preview launch; GA status may have changed	🟡 Medium	Verify current SDK status at https://github.blog/changelog/ — update table row and any "preview" language if it has reached GA

---

🔀 Cross-Linking Opportunities

New zero-inbound articles (published since last audit): 2 | Carryover zero-inbound: 8

🆕 NEW: stop-pulling-your-agentic-unicorns-off-the-field

File: src/content/articles/stop-pulling-your-agentic-unicorns-off-the-field.mdx
Published: 2026-03-12 | Inbound links: 0 | Related articles not linking to it:

1. Link from: stanford-study-ai-roi-in-engineering

   • Suggested section: ## What Engineering Leaders Should Actually Do
   • Insert after: "Teams that push through this curve compound their gains. Teams that don't fall further behind."
   • Suggested text: The compounding effect is exactly what I laid out in [why you shouldn't pull your agentic unicorns off the field](/articles/stop-pulling-your-agentic-unicorns-off-the-field) — that 5% of power users delivering 30%+ capacity expansion are precisely the engineers who pushed through the learning curve. Protect them.

2. Link from: agentic-ai-transforming-dev-teams

   • Suggested section: ## The Circle Replaces the Hierarchy
   • Insert near: "...tested against real user needs, and either validated or killed."
   • Suggested text: But to keep that loop spinning, you have to resist the temptation to pull your best agentic contributors off the field to run training sessions. I wrote about [why that's the most expensive resourcing mistake in engineering today](/articles/stop-pulling-your-agentic-unicorns-off-the-field).

3. Link from: turning-ai-skeptics-into-believers

   • Suggested section: ## The Bottom Line
   • Insert at the end of the closing section
   • Suggested text: Once you've converted them, resist the urge to deploy your power users as trainers. I wrote about [keeping your agentic unicorns on the field](/articles/stop-pulling-your-agentic-unicorns-off-the-field) — the data on why that trade-off rarely pays off.

🆕 NEW: github-copilot-vs-the-world

File: src/content/articles/github-copilot-vs-the-world.mdx
Published: 2026-03-12 | Inbound links: 0 | Related articles not linking to it:

1. Link from: choosing-the-right-ai-sdk

   • Suggested section: The GitHub Copilot SDK section (around line 41–43)
   • Insert after: "I built my SRE agent, I went from zero to a working prototype with MCP server integration in two hours."
   • Suggested text: If you want the full competitive picture on why GitHub's platform approach is pulling ahead of standalone tools, I broke down [why the AI coding war is already over](/articles/github-copilot-vs-the-world) — and what that means for your SDK decision.

2. Link from: copilot-developer-fulfillment

   • Suggested section: ## The Skeptic's Corner
   • Insert near the skeptic objections section
   • Suggested text: For a deeper look at where Copilot stands relative to Cursor, Codeium, and the rest of the field right now, I laid that out in [my breakdown of the AI coding competitive landscape](/articles/github-copilot-vs-the-world).

⚠️ Persistent Carryover: vibe-testing-when-ai-agents-goodhart-your-test-suite

Published: 2026-03-08 | Inbound links: 0

1. Link from: tests-are-everything-agentic-ai

   • Suggested section: ## The Testing Reality for Agentic Teams
   • Insert after: "Once I understood this, building comprehensive test suites became way simpler. Not easier — simpler."
   • Suggested text: What I didn't fully appreciate then: AI agents can game that simplicity. I went deep on what happens when agents optimize for coverage metrics instead of correctness in [my vibe testing breakdown](/articles/vibe-testing-when-ai-agents-goodhart-your-test-suite) — Goodhart's Law with a test runner.

2. Link from: test-enforcement-architecture-ai-agents

   • Suggested section: Intro / first section
   • Suggested text: The failure mode this architecture prevents is what I'm calling [vibe testing](/articles/vibe-testing-when-ai-agents-goodhart-your-test-suite) — when AI agents generate tests that inflate coverage metrics while verifying nothing. Read that first if you want the context for why structural enforcement matters.

---

🌿 Evergreen Maintenance

Refresh candidates found: 5 | All have updatedDate: none and are 21–28 days old.

⚠️ Persistent — all candidates below appeared in the 2026-03-11 audit and have not been updated.

Article	Published	Topic Area	Why Refresh?	Priority
context-engineering-key-to-ai-development	2026-02-14	Context Engineering	22 inbound links make this a pillar piece; MCP has evolved, new Copilot CLI context features released	🔴 High
ai-fixes-its-own-bugs	2026-02-14	Autonomous Bug Fixing	SWE-bench leaderboard moved dramatically; CodeQL Autofix model likely updated; needs factual refresh (see above)	🔴 High
choosing-the-right-ai-sdk	2026-02-14	AI SDKs	SDK landscape evolving fast; Copilot SDK preview status may have changed; OpenAI Agents SDK updates	🟡 Medium
building-the-future-with-ai	2026-02-13	AI Development Overview	9 inbound links; broad intro piece published earliest — should reference newer articles added since (stop-pulling, github-copilot-vs-the-world, vibe-testing, specs-equal-tests)	🟡 Medium
agentic-ai-transforming-dev-teams	2026-02-20	Agentic Dev Teams	Has a broken internal link (see Link Health above); 41% AI code stat sourced from LinkedIn post, not primary research	🟡 Medium

Detailed Refresh: context-engineering-key-to-ai-development

File: src/content/articles/context-engineering-key-to-ai-development.mdx

• What changed since publication: MCP (Model Context Protocol) ecosystem has grown significantly — more servers, broader IDE support. Copilot CLI added new context management features in its Jan 2026 changelog. Several new htek.dev articles reference this piece as a pillar but it doesn't reference them back.
• Suggested updates:
  • Add section on MCP as the emerging standard for context injection
  • Reference agent-hooks-controlling-ai-codebase and vibe-testing-when-ai-agents-goodhart-your-test-suite as applied examples
  • Update any references to Copilot CLI context features with Jan 2026 release notes
• Sources: Copilot CLI Jan 2026 changelog, [MCP official site]((modelcontextprotocol.io/redacted)

Detailed Refresh: ai-fixes-its-own-bugs

File: src/content/articles/ai-fixes-its-own-bugs.mdx

• What changed since publication: SWE-bench top scores moved from 12–14% (early 2024) to 50%+ verified (early 2026). The article's framing of 12.47% as a benchmark number now understates the state of the art by ~4×.
• Suggested updates:
  • Add a callout box: "Since this article was published, SWE-bench verified scores have climbed to 50%+ — the 12.47% figure is the 2024 baseline, not the current ceiling."
  • Verify and update CodeQL Autofix model reference (GPT-4o → current)
  • Reference the 2025 SWE-bench leaderboard analysis paper already cited inline

Generated by Article Relevance Audit — Weekly Health Check · ◷

[github-actions]

🤖 Auto-triage by issue-worker

Why this is suitable for Copilot:
This audit report contains highly specific, copy-and-insert fixes: one persistent broken internal link with exact before/after markdown, plus 8 new cross-linking opportunities where the target file, section, insertion point, and full suggested text are all provided. No external research or editorial judgment needed to execute these.

What Copilot should do:

Focus on the two lowest-risk, highest-specificity categories only — skip the Factual Freshness and Evergreen Maintenance sections, which require external verification and editorial judgment:

1. Broken Internal Link fix (1-line change, ⚠️ persistent since 2026-03-11):

   • File: src/content/articles/agentic-ai-transforming-dev-teams.mdx line ~31, section ## The Circle Replaces the Hierarchy
   • Change: [GitHub Copilot CLI worktrees](/articles/copilot-fleet-mode-video-pipeline) → [GitHub Copilot CLI worktrees](/articles/video-pipeline-with-fleet-mode)

2. Cross-linking — new article: stop-pulling-your-agentic-unicorns-off-the-field (3 links):

   • src/content/articles/stanford-study-ai-roi-in-engineering.mdx → ## What Engineering Leaders Should Actually Do, after "Teams that don't fall further behind."
   • src/content/articles/agentic-ai-transforming-dev-teams.mdx → ## The Circle Replaces the Hierarchy, near "...tested against real user needs, and either validated or killed."
   • src/content/articles/turning-ai-skeptics-into-believers.mdx → ## The Bottom Line, end of closing section

3. Cross-linking — new article: github-copilot-vs-the-world (2 links):

   • src/content/articles/choosing-the-right-ai-sdk.mdx → GitHub Copilot SDK section (~line 41–43), after "I went from zero to a working prototype with MCP server integration in two hours."
   • src/content/articles/copilot-developer-fulfillment.mdx → ## The Skeptic's Corner, near the skeptic objections section

4. Cross-linking — persistent carryover: vibe-testing-when-ai-agents-goodhart-your-test-suite (2 links):

   • src/content/articles/tests-are-everything-agentic-ai.mdx → ## The Testing Reality for Agentic Teams, after "Not easier — simpler."
   • src/content/articles/test-enforcement-architecture-ai-agents.mdx → intro / first section

All exact suggested markdown text is in the issue body — use it verbatim as the basis for each insertion.

Confidence: High

Assigning to Copilot coding agent.

Generated by Issue Worker — Backlog Triage Agent · ◷

[github-actions]

🤖 Auto-triage by issue-worker (re-assignment — previous triage comment added but assignment never completed)

Why this is suitable for Copilot:
This audit contains highly specific, copy-and-insert cross-linking fixes — exact file paths, section names, anchor text, and full suggested markdown are all provided. The broken internal link fix is a 1-line change with a clear before/after. No editorial judgment or external research required.

What Copilot should do:

See the detailed instructions in the previous triage comment for exact text. Summary:

1. Broken link fix (⚠️ now in 4 consecutive audits — highest priority):

• File: src/content/articles/agentic-ai-transforming-dev-teams.mdx, line ~31
• Change: [GitHub Copilot CLI worktrees](/articles/copilot-fleet-mode-video-pipeline) → [GitHub Copilot CLI worktrees](/articles/video-pipeline-with-fleet-mode)
• Note: verify video-pipeline-with-fleet-mode slug exists before applying

2. Cross-links for stop-pulling-your-agentic-unicorns-off-the-field — add to 3 articles:

• stanford-study-ai-roi-in-engineering.mdx → ## What Engineering Leaders Should Actually Do
• agentic-ai-transforming-dev-teams.mdx → ## The Circle Replaces the Hierarchy
• turning-ai-skeptics-into-believers.mdx → ## The Bottom Line

3. Cross-links for github-copilot-vs-the-world — add to 2 articles:

• choosing-the-right-ai-sdk.mdx → GitHub Copilot SDK section (~line 41–43)
• copilot-developer-fulfillment.mdx → ## The Skeptic's Corner

4. Cross-links for vibe-testing-when-ai-agents-goodhart-your-test-suite — add to 2 articles:

• tests-are-everything-agentic-ai.mdx → ## The Testing Reality for Agentic Teams
• test-enforcement-architecture-ai-agents.mdx → intro / first section

Exact suggested markdown text for each insertion is in the issue body and the previous triage comment above.

Skip: Factual freshness, evergreen maintenance, and persistent zero-inbound table items — those require external verification or editorial judgment.

Confidence: High

Assigning to Copilot coding agent.

Generated by Issue Worker — Backlog Triage Agent · ◷

[github-actions]

🤖 Auto-triage by issue-worker (3rd triage — items verified still unresolved in codebase as of 2026-04-01)

Why this is suitable for Copilot:
All items have been verified in the live codebase: the broken link is still on line 31 of agentic-ai-transforming-dev-teams.mdx, and none of the 7 cross-link insertions appear in any of the target files. These are exact copy-and-insert changes with no editorial judgment required.

What Copilot should do:

1. Broken link fix (⚠️ unfixed since 2026-03-11 — highest priority):

• File: src/content/articles/agentic-ai-transforming-dev-teams.mdx, line 31
• Change: [GitHub Copilot CLI worktrees](/articles/copilot-fleet-mode-video-pipeline) → [GitHub Copilot CLI worktrees](/articles/video-pipeline-with-fleet-mode)
• Target file video-pipeline-with-fleet-mode.mdx confirmed to exist

2. Cross-link: stop-pulling-your-agentic-unicorns-off-the-field — insert into 3 articles:

• src/content/articles/stanford-study-ai-roi-in-engineering.mdx → ## What Engineering Leaders Should Actually Do, after "Teams that push through this curve compound their gains. Teams that don't fall further behind."
  • Text: The compounding effect is exactly what I laid out in [why you shouldn't pull your agentic unicorns off the field](/articles/stop-pulling-your-agentic-unicorns-off-the-field) — that 5% of power users delivering 30%+ capacity expansion are precisely the engineers who pushed through the learning curve. Protect them.
• src/content/articles/agentic-ai-transforming-dev-teams.mdx → ## The Circle Replaces the Hierarchy, near "...tested against real user needs, and either validated or killed."
  • Text: But to keep that loop spinning, you have to resist the temptation to pull your best agentic contributors off the field to run training sessions. I wrote about [why that's the most expensive resourcing mistake in engineering today](/articles/stop-pulling-your-agentic-unicorns-off-the-field).
• src/content/articles/turning-ai-skeptics-into-believers.mdx → ## The Bottom Line, end of closing section
  • Text: Once you've converted them, resist the urge to deploy your power users as trainers. I wrote about [keeping your agentic unicorns on the field](/articles/stop-pulling-your-agentic-unicorns-off-the-field) — the data on why that trade-off rarely pays off.

3. Cross-link: github-copilot-vs-the-world — insert into 2 articles:

• src/content/articles/choosing-the-right-ai-sdk.mdx → GitHub Copilot SDK section (~line 41–43), after "I went from zero to a working prototype with MCP server integration in two hours."
  • Text: If you want the full competitive picture on why GitHub's platform approach is pulling ahead of standalone tools, I broke down [why the AI coding war is already over](/articles/github-copilot-vs-the-world) — and what that means for your SDK decision.
• src/content/articles/copilot-developer-fulfillment.mdx → ## The Skeptic's Corner, near the skeptic objections section
  • Text: For a deeper look at where Copilot stands relative to Cursor, Codeium, and the rest of the field right now, I laid that out in [my breakdown of the AI coding competitive landscape](/articles/github-copilot-vs-the-world).

4. Cross-link: vibe-testing-when-ai-agents-goodhart-your-test-suite — insert into 2 articles:

• src/content/articles/tests-are-everything-agentic-ai.mdx → ## The Testing Reality for Agentic Teams, after "Not easier — simpler."
  • Text: What I didn't fully appreciate then: AI agents can game that simplicity. I went deep on what happens when agents optimize for coverage metrics instead of correctness in [my vibe testing breakdown](/articles/vibe-testing-when-ai-agents-goodhart-your-test-suite) — Goodhart's Law with a test runner.
• src/content/articles/test-enforcement-architecture-ai-agents.mdx → intro / first section
  • Text: The failure mode this architecture prevents is what I'm calling [vibe testing](/articles/vibe-testing-when-ai-agents-goodhart-your-test-suite) — when AI agents generate tests that inflate coverage metrics while verifying nothing. Read that first if you want the context for why structural enforcement matters.

Skip: Factual freshness, evergreen maintenance sections — require external verification or editorial judgment.

Confidence: High

Assigning to Copilot coding agent.

Generated by Issue Worker — Backlog Triage Agent · ◷