video1.0<iframe src="https://www.loom.com/embed/4c87bb614e904f34b92d2126b9e17e8c" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/os/loom-media-content/loom/24e9bc0a-c0ba-45f5-8a0a-e6a7263bc9a1/38365f2e-f613-49ac-a7ab-27c581b02381/sessions-thumbnails-4c87bb614e904f34b92d2126b9e17e8c-bbfb569c6644abe3.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS84dW03YzdlcWJybWdhNnB1IiwiYWxnIjoiUlMyNTYifQ.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.EaVM0vKVfWmmUgkMbGFCXEuXoZiGzY-YFwHkl8xVvDOPA2Ia32TLKMchoUPBjWEmOwj-o-bHtqJK6cKX45KDreoQYExJS76YE4HrO-LKpOguP03LR0wdEL0qv-FK3KI2Tm1t1FNmANIq6aaW93r33wuGwgl9a_qytBRqh8I6dpcASfUpaMRSkXNpIfU4ki5uRVg2QRYPBDPpKWxzzIxQZOFj0OnRZWRjE8jyysgFSvsXBWg1vJJMwhhoIp3FGCC60Pn8rxI635VZbecmWfTA-Gl3FVI5zNTpPnl31BcMsrcRCZqdnCmtUXAC92sgrORx_fM_d24YVB5wNIoZwhHafw&Expires=1794735989&Signature=i2YHcQHFhcno1BUFP5MqkM5pzUFQ3v0K4Xu2Ol3wMeY52WyPeJvhYEqW2ndAi0scf7CjH7q%7ETD6fY89CkMdiXYg8puA2x9Jcz6ENgiSHgVnsdUqSIC7FbxfqjchVdw9v9-rx6AtDawHiM0yqMStH6cd-aXUy0CXkyIQVWU5%7EC8TNO7GE3-s5Sok%7EvocWhNbaTKLR0Rh-w2CvRBaFTWSn-4mLpApcPXAxfOaBYjNjAUPMViJ5c8nWQk%7EcqX-UmXX0CyLyTftryBslHdNnZXj69GRzioOeL95utBJi-IVfjqL3Lp1HD9MowZZju%7EJC3lTz3XY3WHZGRg52w9YUx50slA__&Key-Pair-Id=KQOSYIR44AIC0426.347Debugging Agentic Systems with AGDebugger This Loom explains how to debug agentic systems using a checkpointing framework called KGDebugger. It argues that standard unit tests and logs are not sufficient because agent steps are probabilistic and issues may originate in earlier stages even when a later step fails, such as a parsing error in step 5. The framework provides a developer view showing inputs and outputs at each stage, supports step-level parameter overwrites for what-if testing, and includes an RCA agent that identifies likely faulty checkpoints and suggests or applies fixes. In the example, the failure was due to converting an amount from USD to INR, after which the workflow and final response worked correctly.