RT @StephanEwen: I wrote a piece on building 𝗗𝘂𝗿𝗮𝗯𝗹𝗲 𝗔𝗴𝗲𝗻𝘁𝘀 using durable execution, communication, and state in @restatedev. It matches s….

RT @tnachen: Durable agents is inportant for both stateful and stateless agents, but I would argue it would be crucial for stateful agents.….

Some of this came out of a discussion with @cramforce, so credit where credit is due.

The full article goes into a few more details and links more examples.

Multi-agent Orchestration. Agents can call other agents and suspend while awaiting their response. It looks just like ordinary RPC, but is a persistent channel that can suspend callers, can be detached / re-attached, canceled, and lets you fan out/in across parallel remote

Human-in-the-Loop. HIL interactions look like they would in common code: a durable future/promise that is completed with external input. Simple, reliable, doesn't break the flow, and suspends the agent while it awaits input.

Observability. Because the log tracks all inference calls, tool calls, and results, it can visualize the agents work in detail without any extra tools or telemetry.

Recovery in action:

Durable Execution. Put LLM inference calls and tools in to durable blocks. Most SDKs have hooks/wrappers for that. Failures lead to retries that bring the agent back to last completed step.

I wrote a piece on building 𝗗𝘂𝗿𝗮𝗯𝗹𝗲 𝗔𝗴𝗲𝗻𝘁𝘀 using durable execution, communication, and state in @restatedev. It matches so well, it still sometimes feels too good. The approach works with popular frameworks like @aisdk and @OpenAIDevs agent SDK, or even with

RT @restatedev: 🚀 Restate 1.4 just dropped! 10x faster failure detection, automatic workload rebalancing, and 2.5x quicker startup. Your di….

Here is the full article:.

A nice write-up about runtime for AI Agents, and why durable execution is a nicer way than workflow graphs:. My two favorite quotes:. Every single one of these [AI workflow] patterns is 𝗷𝘂𝘀𝘁 𝗽𝗿𝗼𝗰𝗲𝗱𝘂𝗿𝗮𝗹 𝗰𝗼𝗱𝗲 𝘄𝗶𝘁𝗵 𝗰𝗼𝗻𝘁𝗿𝗼𝗹 𝗳𝗹𝗼𝘄 𝘀𝘁𝗮𝘁𝗲𝗺𝗲𝗻𝘁𝘀.

RT @thdxr: 5% of my energy goes to coding. 95% of my energy goes to figuring out wtf we should be doing. thanks to AI it's now 96%.

RT @vanlightly: How to reliably distribute work across microservices, stream processors, durable execution, event-driven, orchestration and….

RT @StephanEwen: I had the pleasure of speaking with @seanfalconer about distributed applications, resilience, scalability, from @ApacheKaf….

RT @restatedev: Restate auto-retries temporary issues like network glitches. But for business logic failures ("Hotel fully booked"), retryi….

I had the pleasure of speaking with @seanfalconer about distributed applications, resilience, scalability, from @ApacheKafka and @ApacheFlink to @RestateDev. Restate started with the question: How close can we get to replicating the magic correctness of exactly-once stream.

RT @software_daily: Stephan Ewen is the Founder at @restatedev which aims to simplify modern distributed applications. Stephan joins the sh….

RT @restatedev: Serverless is great, until you need to manage state. This is where Restate comes in!.Learn how to combine @KnativeProject w….