> Markdown version of [/videos/100207-in-depth-net-azure-functions-isolated-mode-performance-and-durable-ai-agents?t=448](https://www.wearedevelopers.com/videos/100207-in-depth-net-azure-functions-isolated-mode-performance-and-durable-ai-agents?t=448). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# In-depth .NET Azure Functions: Isolated mode, performance and durable AI agents

The smallest compute tier in Azure Flex Consumption actually worsens cold starts. Discover the optimal .NET configurations for Isolated workers to slash latency and orchestrate resilient AI agents.

- **Speakers:** [Stas Lebedenko](https://www.wearedevelopers.com/@stanislav-lebedenko)
- **Event:** World Congress 2026 Europe
- **Published:** July 10, 2026
- **Duration:** 26:58
- **URL:** https://www.wearedevelopers.com/videos/100207-in-depth-net-azure-functions-isolated-mode-performance-and-durable-ai-agents

## Summary

The .NET Azure Functions ecosystem is undergoing a significant architectural shift as in-process support ends, making the Isolated worker model and Flex Consumption hosting plans the new standard. Migrating requires adapting to Linux-only environments, acknowledging region-specific compute capacity limits that often require manual quota requests, and replacing deprecated deployment slots with rolling upgrades. Because hosting costs and architectural behaviors differ drastically under Flex Consumption, careful configuration of concurrency and dynamic scaling parameters inside `host.json` becomes non-negotiable for stable operations.

Achieving low latency in Isolated mode—particularly bridging the cold-start gap—relies heavily on application configuration and dependency trimming. Utilizing the `ReadyToRun` compiler option noticeably shrinks application bundle sizes, which facilitates tighter instance placement and resource reuse on 2GB and 4GB memory tiers. Interestingly, load testing reveals a platform quirk where the smallest compute tier (0.5GB) occasionally suffers worse cold-start delays reaching up to 800 milliseconds, establishing the 2GB compute tier paired with `ReadyToRun` optimization as the practical sweet spot for production APIs. Continuously validating capacity with high-volume tools like loader.io ensures predictable behavior across varied execution loads.

Beyond basic serverless triggers, Durable Functions excel as a robust orchestration backbone for the Microsoft Agent Framework. Resolving complex architectures like distributed transactions or sudden "fan-in/fan-out" traffic spikes requires specific design constraints; developers should implement sub-orchestrators to prevent heavy request loads from bottlenecking a single execution node. Crucially, the main orchestrator must remain exceptionally lean, delegating all data persistence and external system interactions strictly to activity functions. By natively integrating AI agents into Durable Functions, developers gain reliable state control and disaster recovery while maximizing cost efficiency, automatically eliminating billing charges while execution is paused for human-in-the-loop reviews or long-running Model Context Protocol (MCP) interactions.

**Keywords:** .NET azure functions isolated mode, azure flex consumption plan, durable functions orchestration, microsoft agent framework, aot readytorun compilation, cold start latency profiling, host.json concurrency optimizations, loader.io load testing, fan-in fan-out pattern, sub-orchestrator architecture, http response data bindings, mcp tools integration, human-in-the-loop serverless computing, azure region capacity limits

## Chapters

1. **Azure functions hosting plans and migration challenges** (00:02) — Because older consumption plans use Windows containers, teams must update applications to adopt the superior Flex consumption model.
1. **Flex consumption plan features and capacity limitations** (02:08) — Managing the Linux-only Flex consumption plan requires acknowledging missing deployment slots and requesting adequate regional compute quotas.
1. **Adopting the .NET isolated worker model for functions** (04:29) — Transitioning away from legacy in-process logic enables teams to deploy standard console applications featuring multiple output bindings.
1. **Load testing and native AOT pre-compilation performance** (07:28) — Testing unoptimized cold starts reveals how native AOT compilation significantly reduces startup latency across various compute sizes.
1. **Optimizing durable functions scaling with sub-orchestrators** (12:35) — Processing heavy event streams necessitates deploying sub-orchestrators to eliminate single-node bottlenecks found in traditional fan-in patterns.
1. **Tuning dynamic concurrency and durable tasks in host.json** (14:11) — Tuning the default configuration parameters allows Azure to automatically manage concurrent activity execution and reduce host logging overhead.
1. **Best practices for function builds and durable orchestration** (17:01) — Reducing cold start footprints requires trimming framework dependencies and carefully managing connection pools within orchestrated activity functions.
1. **Building durable AI agents with Microsoft agent framework** (19:06) — Executing long-running workflows with human feedback benefits from adopting integrated durability structures within the Microsoft agent framework.
1. **Key takeaways for isolated mode and future migrations** (22:50) — Ensuring predictable performance upon scaling highly depends on establishing thorough load tests before the isolated mode deprecation deadline.

## Related Moments

- [Introduction to serverless websites with Blazor and Azure Functions](https://www.wearedevelopers.com/videos/115-blazor-and-azure-functions-for-serverless-website) (from "Blazor and Azure functions for Serverless Website")
- [Audience Q&A on language support, performance, and tradeoffs](https://www.wearedevelopers.com/videos/588-using-webassembly-for-in-database-machine-learning) (from "Using WebAssembly for in-database Machine Learning")
- [Shifting developer workloads and realistic AI productivity gains](https://www.wearedevelopers.com/videos/1830-wearedevelopers-live-speculaitions) (from "WeAreDevelopers LIVE - SpeculAItions")
- [Scaling operations using Azure AI Foundry tools](https://www.wearedevelopers.com/videos/1535-from-traction-to-production-maturing-your-genaiops-step-by-step) (from "From Traction to Production: Maturing your GenAIOps step by step")
- [Overview of serverless architectures and function as a service](https://www.wearedevelopers.com/videos/814-serverless-architectures-with-spring-cloud-functions-and-knative) (from "Serverless Architectures with Spring Cloud Functions and Knative")
- [Comparing serverless functions against traditional monolithic architectures](https://www.wearedevelopers.com/videos/590-building-your-way-to-a-serverless-powered-iot-buzzwire-game) (from "Building your way to a serverless powered IOT Buzzwire game")

## Related Articles

- [Dev Digest 132 - Binging WADFlix?](https://www.wearedevelopers.com/magazine/473-dev-digest-132-binging-wadflix)
- [MLOps And AI Driven Development](https://www.wearedevelopers.com/magazine/82-mlops-and-ai-driven-development)
- [Building AI Solutions with Rust and Docker](https://www.wearedevelopers.com/magazine/494-building-ai-solutions-with-rust-and-docker)
- [Stephan Gillich - Bringing AI Everywhere](https://www.wearedevelopers.com/magazine/489-stephan-gillich-bringing-ai-everywhere)

## Related Jobs

- [Cloud Foundations Team](https://www.wearedevelopers.com/jobs/ext/1483289-cloud-foundations-team) at **GitHub**
- [Principal Software Engineer, Enterprise AI Platform](https://www.wearedevelopers.com/jobs/ext/1467292-principal-software-engineer-enterprise-ai-platform) at **GitHub**
- [Senior Cloud Software Architect (all genders welcome) for our Intelligent Service Operations Hub](https://www.wearedevelopers.com/jobs/ext/1693682-senior-cloud-software-architect-all-genders-welcome-for-our-intelligent-service-operations-hub) at **Rosenxt Group**
- [Staff Software Engineer, Copilot Experiences](https://www.wearedevelopers.com/jobs/ext/164361-staff-software-engineer-copilot-experiences) at **GitHub**
- [Senior Cloud Software Architect (all genders welcome) for our Intelligent Service Operations Hub](https://www.wearedevelopers.com/jobs/ext/1284556-senior-cloud-software-architect-all-genders-welcome-for-our-intelligent-service-operations-hub) at **Rosenxt Group**
- [Senior Software Engineer, Enterprise Products](https://www.wearedevelopers.com/jobs/ext/1841248-senior-software-engineer-enterprise-products) at **GitHub**