BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.devcon.org//devcon8//talk//JCXENF
BEGIN:VTIMEZONE
TZID:IST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:IST
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devcon8-JCXENF@cfp.devcon.org
DTSTART;TZID=IST:20261105T111500
DTEND;TZID=IST:20261105T113000
DESCRIPTION:Small open models are increasingly capable of powering AI agent
 s locally\, giving developers more control over privacy\, cost\, and deplo
 yment. But most agent demos follow the happy path: tools work correctly\, 
 inputs are predictable\, and the agent successfully completes the task. Re
 al-world systems are rarely that simple.\n\nWhat happens when a tool fails
 \, returns unexpected information\, or when a task requires multiple steps
 ? More importantly\, can a small open model recognize failure\, recover in
 telligently\, and avoid confidently producing an incorrect answer?\n\nIn t
 his talk\, I will explore these questions through a reproducible experimen
 t using AI agents powered by small open models running locally. The agents
  will be tested across increasingly challenging scenarios\, including tool
  selection\, argument generation\, multi-step tasks\, tool failures\, unex
 pected outputs\, and failure recovery.\n\nRather than evaluating only the 
 final response\, we will examine the agent's behavior throughout the task 
 using practical measures such as tool correctness\, task completion\, reco
 very rate\, step efficiency\, and graceful failure.\n\nAttendees will see 
 where small open models perform well\, where agent reliability begins to b
 reak down\, and what developers should consider when moving beyond demos. 
 The experiment\, test scenarios\, evaluation code\, and results will be re
 leased openly for others to reproduce and extend.
DTSTAMP:20261101T122847Z
LOCATION:Community Led Stage
SUMMARY:Can Small Open Models Power Reliable AI Agents? - Kalyan Prasad
URL:https://cfp.devcon.org/devcon8/talk/JCXENF/
END:VEVENT
END:VCALENDAR