Why Agentic AI Requires a New Approach to Orchestration | Niels Bantilan, Union.ai

Why Agentic AI Requires a New Approach to Orchestration | Niels Bantilan, Union.ai

by Niels Bantilan

Applied Sciences & Engineering AI — News & Applications UYComputer scienceUYQArtificial intelligence (AI)
🎙 Niels Bantilan 👥 5K 📅 September 29, 2025 ⏱ 34 min 👁 78 📄 expert opinion 🧭 2026-08-15
Available in: English (current) Français

Keywords

agentic AIorchestrationFlyte 2.0dynamic workflowsfault tolerance

Summary

Niels Bantilan, Chief ML Engineer at Union.ai, presents a talk on why agentic AI requires a new approach to orchestration. He begins by tracing the evolution of ML tooling from ETL to modern agent orchestration, highlighting the shift from static pipelines to dynamic, adaptive systems. He identifies six common pain points for builders: steep learning curves for DSLs, difficulty in debugging, limitations of static execution, challenges with fan-out, brittle pipelines, and difficulty implementing best practices. To address these, he announces Flyte 2.0, a durable, fully dynamic, resource-aware AI orchestration platform. He demonstrates its Pythonic authoring, support for async and parallel execution, dynamic workflow construction, and fault tolerance through a proprietary logging system. He also covers automatic versioning, multi-tenancy, and declarative infrastructure. The roadmap includes a private beta, open-sourcing the SDK, a Rust core, plugins for distributed training, a local sandbox, and eventual full open-source availability. The talk concludes with a Q&A addressing guardrails, state recovery, resource allocation, and debugging.

159 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the limitations of existing orchestration tools for agentic AI, backed by practical examples from real-world deployments. The argumentation is coherent and grounded in the speaker’s experience, but it is primarily a product pitch for Flyte 2.0, which limits its objectivity. The speaker effectively demonstrates the need for dynamic, fault-tolerant orchestration and presents a plausible solution, but the claims are not independently verified.

Scientific Rigor, Source Quality, Title Accuracy

The talk is an expert opinion rather than a peer-reviewed study, so scientific rigor is moderate. The speaker references his own open-source projects (Pandera, Flyte) and industry trends, but does not cite external research. The title accurately reflects the content, and the presentation is well-structured. No comments were provided for analysis.

134 words

Title / Content Match

The title accurately reflects the content: the talk argues that agentic AI requires new orchestration approaches and presents Flyte 2.0 as a solution.

Quality & Reliability

7/10

The talk presents expert opinion from a practitioner with deep experience in ML orchestration, backed by concrete demos of the Flyte 2.0 platform. Claims are not formally verified but are plausible and grounded in practical experience. The presentation is transparent about limitations and future roadmap.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • MLOps World — Event website for the GenAI World session where this talk was recorded.

Concurring Sources

  • Flyte — The open-source project that Flyte 2.0 is based on, supporting the claims about its capabilities.
  • LangGraph — A framework for building agentic AI applications, illustrating the need for dynamic orchestration.
  • AutoGen — A framework for multi-agent AI systems, illustrating the need for dynamic orchestration.

Contribution & Novelties

The talk presents Flyte 2.0 as a novel orchestration platform designed for agentic AI, emphasizing dynamic workflows, fault tolerance, and resource awareness. It contributes to the discourse on AI infrastructure by arguing that traditional static orchestration is insufficient for modern AI agents. The speaker provides concrete examples and demos, making the concepts tangible.

Pour aller plus loin :

  • Flyte — The open-source workflow automation platform that Flyte 2.0 builds upon.
  • LangGraph — A framework for building stateful, agentic AI applications, relevant to the orchestration discussion.
  • AutoGen — A framework for multi-agent AI systems, relevant to the orchestration discussion.

98 words

Radar Profile

The radar profile shows a balanced performance across all dimensions, with slightly lower scores in technical depth and information quality, reflecting the talk's focus on product demonstration rather than deep technical analysis. The overall reliability is moderate, consistent with an expert opinion presentation.

Reliability 7/10