datakazkn

The exoskeleton framing resonates, especially for repetitive data work. Parts where AI consistently delivers: pattern recognition, format normalization, first-draft generation. Parts where human judgment is still irreplaceable: knowing when the data is wrong, deciding what 'correct' even means in context, and knowing when to stop iterating.

The exoskeleton doesn't replace instinct. It just removes friction from execution so more cycles go toward the judgment calls that actually matter.

The hallucination-in-analysis problem is real and often undersold. Pattern that works well: use the LLM only to structure already-extracted data (parse fields, normalize formats), then apply deterministic logic for anything numerical. That way the LLM is doing classification/extraction where it's reliable, and you're not trusting it to compute or compare values where it isn't.

One underappreciated reason for the agentic gap: Gemini tends to over-explain its reasoning mid-tool-call in a way that breaks structured output expectations. Claude and GPT-4o have both gotten better at treating tool calls as first-class operations. Gemini still feels like it's narrating its way through them rather than just executing.