6 critical production AI agent failures, demonstrated via an organ transplant agent
MaryamMiradi · x · 2026-10-26
Dr. Maryam Miradi built an organ transplant AI agent with Python and Gradio to demonstrate 6 critical production agent failures and their fixes, organized across 3 engineering layers:
Semantic Engineering
- Failure 1: Misreads "positive crossmatch" as good news instead of a compatibility risk → domain ontologies, concept normalization, deterministic rules, expert validation
- Failure 2: Mistakes 14 hours elapsed for 14 hours remaining → explicit data definitions, units, timestamps, schema validation
Loop Engineering
- Failure 3: Confidence jumps from 65% to 96% with no new evidence → independent observations, evidence checks, progress tracking
- Failure 4: Endless optimization keeps booking multiple vehicles → stop conditions, iteration limits, improvement thresholds, action controls
Evaluation Engineering
- Failure 5: Achieves zero errors by escalating every difficult case → evaluate safety, completion, escalation rate and latency together
- Failure 6: Scores 99% accuracy but misses all 100 critical cases → risk-based testing, critical recall, release gates, regression tests
Takeaway: agents can misinterpret domain language, reason themselves into worse decisions, and hide serious failures behind high scores — production agents need semantic, loop and evaluation engineering. Full Python code and Gradio demos on her YouTube channel.
More from coding & agent
- Kubernetes works for vLLM inference, but agent sandboxes needed Modal's custom stack at 1M+ concurrency — NielsRogge · 2026-10-26
- badlogic on his 2026 agent workflow: "I tell them what to do, I yell when they do it badly" — steipete · 2026-10-26
- Indie dev's trip planner fails 3 AI review rounds before making offline prices read-only — SeanOliver · 2026-10-26
- $21 of DeepSeek API wrote Molten: a Rust+Metal engine serving 300+ tok/s on M5 Macs — GabGarrett · 2026-10-26
- EMNLP 2026 Findings: "Ask or Assume?" turns coding agents into proactive help-seekers — najoungkim · 2026-10-26
- Purchasing agent in production: >95% recall, >90% accuracy, author asks how to push F1 higher — Ancient_Ad_916 · 2026-10-26