A developer-journalist reflects on lessons from Salesforce TDX 2026 about keeping AI agents healthy after initial deployment — the 'day two problem'. Three key insights: Salesforce's Falabella case shows capturing failures and feeding them back into agent instructions creates a self-healing loop; Adobe Population Health's CIO turned compliance audits directly into LLM eval specs at production scale; MuleSoft's SVP argues that existing IT infrastructure (data governance, audit trails, compliance controls) is the enabler — not the blocker — for agentic systems. The post blends conference reporting with personal agent-building experience and ends with a promotional workshop invite.
Table of contents
Jayesh’s “day two” problem for agentsAlex turning the audit into the evalComstock reminds us all to be learners againHeadless is the way I work every dayThe invitationSort: