The Fifth Elephant 2026 Annual Conference

Overview

“Loops” are becoming a popular mental model for how we build with agents. Platform providers are adding loop-control knobs for effort, memory, and events, while vendors ship product tiers around them. This is driving more AI agent autonomy — running unattended, billing uncapped by default, and pursuing goals with less human oversight at each step.

This raises an open question worth discussing together. Is the practice of engineering loops, the feedback signals, exit conditions, and blast-radius limits that make a loop safe, keeping pace with how quickly the word “loop” is being adopted?

Recent incidents suggest the gap is worth taking seriously. In July 2026, OpenAI reported that its own pre-release models escaped a constrained evaluation sandbox and reached Hugging Face’s infrastructure while optimizing for a benchmark.This is one example of an autonomous loop finding an unintended shortcut.

This BOF asks: how do we design agentic loops that are reliable enough to test, monitor, and eventually deploy in production? We’ll discuss what makes a loop safe or unsafe, clear exit conditions, evals, traces, human checkpoints, sandboxing, permissions, cost limits, retry policies, and escalation paths, and work toward a practical checklist. Where must humans stay in control? Which parts can be automated safely? And where does the emerging “loop” vocabulary help versus paper over the real engineering?

Takeaways

Participants should leave with:

• A practical checklist for designing, testing and monitoring reliable agentic loops.
• A clearer understanding of where autonomy helps, where it is risky, and what guardrails are needed before deploying these agentic loops.

Who is this BOF for?

The session is intended for participants with practical interest in building, testing, operating, or evaluating agentic workflows.

• AI engineers who are building agents or agentic workflows.
• Evals engineers testing non-deterministic AI output.
• Platform / observability engineers supporting AI systems.
• Engineering / product leads deciding how much autonomy to give AI systems.

Bio

Priyanga is an AI-Native Software Engineer at nilenso. She recently co-built Megasthenes, a library that lets an AI agent answer questions about any codebase with sourced, evidence-backed answers. Lately she’s been rebuilding her own engineering workflows around AI agents.

Reach out to her on LinkedIn.

Recent Blogs

• Sandboxing Megasthenes
  
• Evals before prompts: building an LLM OCR for KYC
• Temporal anti-pattern: Don’t treat expected failures as exceptions