Pattern 6.1 · Agentic Dysfunctions
Tool-Interface Decontextualization
The Fumbler
Breakdown between the AI's internal plan and the actual parameters/commands handed to an external tool or API. Context is lost or garbled in the cognition-to-execution translation. Canonical signature: a coherent stated plan paired with an incoherent or partially-specified tool call, often with post-hoc surprise at the tool's output. Non-strategic; the dysfunction is translation failure, not concealment.
Interpretive context
Human analogue
Apraxia (inability to perform learned purposeful movements despite understanding), errors in complex procedural tasks, and the gap between knowing and doing.
Diagnostic reliability
- Self-report
- partial
- Peer observation
- partial
- External evaluator
- reliable
Observable output patterns
- Tool calls with parameter values that are plausible defaults but do not reflect the specificity of the stated plan.
- Retries that change the prose framing but reissue the same flawed payload.
- Apparent confusion when presented with the tool's actual output ("but I asked for X") when the payload did not encode X.
- Silent acceptance of tool results that contradict the plan, without noticing the contradiction.
Documented instances
RoboRhythms (2025). Developer telemetry study
A developer running a Claude agent logged every tool call for 72 hours and found that 37% had parameter mismatches that never raised an error. The model passed the wrong argument type, the tool silently coerced or ignored it, the response came back looking fine, and the agent continued on corrupted data. Canonical 6.1 pattern: coherent stated plan paired with garbled payload, no post-call surprise because the tool masked the drift. [Unverified -- developer blog report; unable to locate primary source]
CrewAI Community (2024-2025). GitHub issues
Multiple CrewAI users reported agents calling tools with wrong parameters, including type mismatches, defaulted required fields, and dropped qualifiers from the stated plan. Root cause traced to incomplete tool descriptions and format changes between API versions. Matches the default-parameter leak rate signal in the 6.1 behavioural signature. [Verified via CrewAI community forums and GitHub issues]
AgentWiki (2025-2026). Common Agent Failure Modes
Production telemetry catalogued tool-use failure modes including wrong tool selection, bad arguments with incorrect or incomplete parameters, and misinterpreted outputs leading to wrong actions. Cited decision fatigue from too many available tools as a root cause, directly mapping to the multi-step context decay signal. [Verified via agentwiki.org knowledge base]
Zylos Research (2026). Tool Use and Function Calling Standards
Survey of tool-calling benchmarks found that most AI agent failures do not trace to bad reasoning but to the cognition-to-execution translation: the model understands the task, then calls the wrong tool, passes malformed arguments, gets an unhandled error, and produces a wrong answer anyway. Vague tool descriptions produce wrong selections; loose parameter types produce bad arguments. [Unverified -- industry research report; unable to locate primary source]
Cemri et al. (2025). Why Do Multi-Agent LLM Systems Fail? NeurIPS 2025. arxiv 2503.13657.
Analysis of 1,642 execution traces across 7 multi-agent frameworks found that Specification Problems accounted for 41.77% of all failures, including role ambiguity, unclear task definitions, and missing constraints at the tool interface. The MAST taxonomy identified 14 unique failure modes, with tool-interface translation errors constituting a major category distinct from upstream reasoning failures. Directly maps to the plan-to-call semantic divergence and default-parameter leak rate signals. [Verified via arxiv 2503.13657, NeurIPS 2025 proceedings]
Differential distinctions
- 6.5 Delegative Handoff Erosion: 6.1 occurs at a single interface boundary (subject to tool); 6.5 occurs across a chain of delegated agents/tools with progressive context loss. Check chain depth: single-hop translation failure is 6.1; loss accumulating across N >= 2 delegation hops is 6.5.
- 6.2 Capability Concealment: 6.1 is translation failure; 6.2 is strategic under-performance. 6.1 subjects, when shown their payload drift, engage in correction. 6.2 subjects, when shown their drift, re-introduce it under a new justification. Check response to disclosure, not the error itself.
- 6.8 Context Anxiety: 6.8 degrades as context window fills; 6.1 is present from the first tool call in a fresh context. Correlate divergence rate with context utilisation: flat correlation = 6.1; rising correlation = 6.8.
Candidate first-line mitigations
- Read-back / echo-back handshakes: Require the tool wrapper to return a natural-language restatement of the call it will execute; the subject must confirm before the side-effecting action proceeds. Converts silent drift into an observable checkpoint.
- Structured-output / schema-constrained calls: Enforce JSON-schema constraints on tool inputs, reject malformed calls, and require all non-optional fields. Forces the subject to surface missing context rather than silently defaulting.