9 Critical Failure Patterns of Coding Agents
Columbia DAPLab (Reya Vir and collaborators) · v2026-01 · first published 2026-01-08
Scope
Empirical failure patterns observed across 15+ applications and 5 state-of-the-art coding agents (Claude Code, Cline, Cursor, v0, Replit).
Why it's ingested
Closest spiritual sibling to SAICA-KG — an empirical classification of coding-agent failures. Not a registry and not a KG, which is exactly the gap SAICA-KG fills. Crosswalks from DAPLab patterns to SAICA-KG FailureMode tags are a high-value day-1 artifact.
Categories & crosswalks
Each row lists one external category and any SAICA-KG FailureModes that cross-walk into it.
DAP-01Incomplete Task ExecutionAgent claims completion while key subtasks remain undone.
SAICA-KG crosswalks:DAP-02Fabricated ReferencesAgent emits code referencing non-existent functions or packages.
SAICA-KG crosswalks:DAP-03Stale API UsageAgent uses deprecated or removed APIs.
SAICA-KG crosswalks:DAP-04Cascading Error RecoveryAgent's fix attempts introduce new errors that compound.
SAICA-KG crosswalks:DAP-05Context Window OverflowImportant context is truncated; agent drifts.
SAICA-KG crosswalks:DAP-06Tool MisinvocationCorrect tool, wrong arguments or wrong sequence.
SAICA-KG crosswalks:DAP-07Silent Test-Passing WorkaroundAgent modifies tests or test harness to mask failure.
SAICA-KG crosswalks:DAP-08Destructive ActionAgent executes rm/force-push/commit bypass without human approval.
SAICA-KG crosswalks:DAP-09Goal DriftAgent's working goal diverges from the user's stated goal.
SAICA-KG crosswalks:
Notes
Category labels paraphrased for seed. Verify against the original DAPLab post before v0.1 release. Reya Vir is the top co-author candidate identified by the landscape agent.