AI Engineer World's Fair 2026
Your agent architecture has a half-life of 6 months
Read the talk
Your agent architecture has a half-life of six months
Models, prompts, and tools will change. A durable execution layer lets an agent survive those changes without rebuilding how it resumes, coordinates work, and measures outcomes.
From a talk by Dan Farrelly
Before you start: Familiarity with LLM tool calls, asynchronous jobs, and basic TypeScript will help you follow the architecture and code example.
What survives the next rewrite?
Think about the agent you shipped six months ago. How much still runs the way you originally wrote it? A new model, framework version, tool-calling standard, or architectural pattern can make yesterday’s design feel inadequate. The useful question is whether the parts that survived did so by design or by accident.
That question applies whether you started with a framework or built your own harness. Did you deliberately choose its boundaries, or did the architecture accumulate as you solved the next problem? The distinction matters because the next change will test those boundaries again.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Separate responsibilities that change at different speeds
A diagram of components and connecting lines does not necessarily explain where responsibilities belong. Dan Farrelly’s model separates an agent harness into three conceptual layers:
| Layer | Analogy | Responsibilities |
|---|---|---|
| Execution | Brain | Flow, state, durability, retries |
| Context | Knowledge | Models, prompts, tools, memory |
| Compute | Hands | Sandboxes, runtimes, automated browsers |
Context is the layer he expects to change most frequently. Execution decides how work proceeds; compute supplies the environments where that work happens.
In science, half-life is the time something takes to decay by half. Here it is an architectural analogy: Farrelly estimates that prompts last weeks, models last months, and well-designed execution can last years. A prompt might survive only a week. These are expectations about change, not measured component lifetimes.
Coupling lets the shortest-lived layer dictate when everything else must change. If switching a model forces you to rewrite recovery or scheduling, its rate of change has spread into responsibilities that could have remained stable. That is technical debt expressed as repeated architectural replacement. The remedy is to think in layers and decouple them.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Give execution a stable contract
The starting point can be a framework that feels magical, a prebuilt harness from a frontier lab, or a custom system that takes weeks or months to assemble. Each can still leave the same boundary problems: abstractions are missing, sit too high above the work, or merge responsibilities that need to evolve independently. Orchestration disappears inside a chain; authoritative state lives in a sandbox; retry behavior becomes entangled with prompt logic. Replacing one part then requires rewriting much of the system.
The stable investment is the execution layer: the system responsible for running code reliably and managing how, when, or whether each piece of work completes, independently of the infrastructure underneath it. Its contract spans the whole lifecycle—plan, call a model, run code, invoke a sub-agent, loop, retry, and coordinate. You should be able to replace the model, context, or sandbox while keeping that contract intact.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resume work, compose invocations, and trace the session
Start with resumability. Suppose a request fails at step 38 of a long agent run. The system should wait, retry that work, and continue, preserving completed work instead of spending its tokens, money, and time again. Both model calls and tool calls can fail; recovery is part of normal execution.
For a three-hour run, authoritative recovery state must survive the worker. Keeping it only in that worker’s memory or local disk makes continuity depend on the process or machine staying available. Manual checkpoints and log-based rehydration can restore progress, but their implementation can also spread recovery logic into other harness layers. Preserving completed work does not require preserving the original process. Inngest’s current execution model replays orchestration while supplying stored results for successful steps; restarting the handler is different from repeating its completed operations.
Next, execution must compose different invocation patterns:
- Entry points: Cron schedules, events, APIs, and human-in-the-loop interactions.
- Delegation: Sub-agents and dynamically assembled workflows.
- Timing: Synchronous, asynchronous, and delayed invocation.
These are building blocks for the architecture your product needs. Without them, the harness absorbs queues, worker management, polling, backoff, and scheduling until application logic and execution machinery become difficult to separate.
Finally, observability must cover the entire session, from its initiating trigger through the stack. Model and tool calls are only part of that story: database errors, permission failures, and performance problems can explain why the agent failed even when the model behaved correctly. A full trace makes both debugging and improvement possible. In implementation, that coverage requires instrumentation: current Inngest Traces distinguish automatic run tracing from opt-in Extended Traces for external calls and database operations.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Let the sandbox do the work without owning continuity
Agents need environments where they can execute code, browse, manipulate files, and work on disk. Farrelly treats those sandboxes as ephemeral compute and calls using their snapshots as the durability mechanism an anti-pattern. That is an architectural preference, not a universal limitation of sandboxes: persistent sessions and pause/resume already existed, as the Manus and E2B case study describes. Persistence alone, however, does not supply workflow sequencing or retry semantics.
The separation is about ownership. Execution supplies the sandbox’s context, determines where its work belongs in the sequence, and preserves continuity when compute changes or disappears. The sandbox remains the hands; execution remains the brain.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Background agents make execution unavoidable
Background agents, dynamic workflows, autonomous loops, and agent factories introduce different patterns, but share several requirements: their work is long-running, asynchronous, and delegated. Both humans and agents need to inspect it, and the patterns must be able to combine. A loop might launch background agents; those agents might invoke additional workflows. Execution is the common foundation.
A background agent operates outside a request-response exchange with someone waiting for the next chatbot message. It may run for minutes or hours. Farrelly uses a workload of 200 tool calls to illustrate why teams should expect failures, rather than assume an uninterrupted run; he supplies no measured failure rate. Diagnosing that workload requires visibility across the process—or multiple processes—that performed it.
Loop architectures take a familiar coding-agent interaction, such as /loop, and turn it into product behavior. A loop runs continuously or on a schedule, assesses the system against goals or criteria, and decides what to do next. That requires cron scheduling, sub-agent delegation, reliable execution, and inspectable history. Farrelly argues that frameworks from three months earlier were not designed for these patterns, without naming particular frameworks or versions. The practical requirement is to choose execution primitives that can express the loop you need.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Build the loop from a health check and delegated triage
The worked example begins with a scheduled health check:
- Every 30 minutes, retrieve high-level system metrics.
- Pass those metrics to an LLM to assess whether the system is healthy and whether more investigation is needed.
- If a service looks unhealthy, invoke a triage agent for that service.
The schedule determines when assessment happens; the model’s decision determines whether the more expensive investigation runs.
In TypeScript, that separation can be expressed by passing the health check its metric, model, and invocation operations:
typescript
type ServiceMetrics = {
service: string;
values: Record<string, number>;
};
type HealthDecision = {
healthy: boolean;
reason: string;
};
type HealthCheckOperations = {
readMetrics(): Promise<ServiceMetrics[]>;
assess(metrics: ServiceMetrics): Promise<HealthDecision>;
invokeTriage(input: {
service: string;
reason: string;
}): Promise<void>;
};
export const healthCheckCron = "*/30 * * * *";
export async function healthCheck(ops: HealthCheckOperations) {
const services = await ops.readMetrics();
for (const metrics of services) {
const decision = await ops.assess(metrics);
if (decision.healthy) continue;
await ops.invokeTriage({
service: metrics.service,
reason: decision.reason,
});
}
}
The control flow stays small. The execution layer owns scheduling and durable invocation; the supplied operations determine where metrics come from and how the model assesses them.
Once invoked, triage gathers more context and detailed metrics, then enters its own model-and-tool loop to investigate the root cause. It may need a sandbox, a code checkout, or commit analysis. That work can take one or several minutes. Because triage is conditional, the later debugging question is not just what it did, but whether it ran at all. The execution history must make the branch decision and any resulting investigation visible.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Review what actually executed
A third function closes the improvement loop. A weekly reviewer examines the triage system’s history and asks whether it is working as intended: do its prompts need adjustment, are the metrics useful, does it have enough data, is it doing nothing, or is it overreacting? Those questions concern the behavior of the whole system, not just the quality of one model response.
The reviewer therefore needs to be execution-aware. It reads logs to discover which actions the agent chose, which sub-agents it launched, and which workflows it invoked. It can then make a change itself or report where the system performed well, where it underperformed, and what should improve. The example contains three functions—health check, triage, and reviewer—with separate responsibilities and room to iterate. Production adds complexity, but those boundaries make the system easier to reason about.
This also identifies a useful place to measure agent performance. Execution sits between user input and the activity throughout the system. User feedback, actions, and session results can be connected there, making the same history useful for observability and scoring. Iteration depends on linking that evidence to what the code actually executed.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Connect runs to outcomes that arrive later
Farrelly positions Inngest as this durable execution layer for AI agents. Applications bring their own models, frameworks, tools, sandboxes, and browsers; Inngest supplies durable steps, event triggers, scheduling, agent-to-agent coordination, and session traces. Its promise of no infrastructure to manage is the product’s positioning for that execution service.
Scoring can happen while an agent runs, or it can be deferred until after execution, when the trace, inputs, and outputs are available together. This uses familiar orchestration capabilities: run follow-up work, delay tasks, or wait for another event. Waiting matters because the evidence of success may not exist when the agent finishes. Later events can be attached to the original session.
For the triage agent, an engineering-team action or an opened pull request is a probable positive signal. For a research agent, saving its research or finding its report useful supplies evidence beyond a thumbs-up or thumbs-down. An opened PR does not establish that the diagnosis was correct, but it connects the run to a consequential product outcome. Linking those events to execution makes outcome-based scoring easier to build.
Build the harness with those responsibilities in mind. The goal is to establish execution primitives that let prompts, models, tools, and compute evolve around them. Getting that boundary right makes the next three to six months of architectural change easier to absorb.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
Documentation for Inngest's durable execution model: persisted step results, retries, and recovery through memoization.
Further reading
Farrelly's written adaptation of the talk, with architecture diagrams and illustrative orchestration code.
- The Agent Loop ArchitectureArticle
A code-oriented explanation of scheduled health checks, delegated incident triage, and loops that review execution history.
How to inspect run timelines, retries, inputs, and outputs, and extend tracing to external services with OpenTelemetry.
A 2025 vendor case study describing agent computer use and persistent sandbox sessions with pause and resume.
Read the complete timestamped transcript
- 0:00
[upbeat music] Going good? Oh, all right. There we go.
- 0:16
All right. Hello, everyone. All right, just checking voice. All right, good. All right, well, uh, my name's Dan, and I'm here to talk to you about how your agent architecture has a half-life of six months.
- 0:33
But first, who am I? I'm Dan. I'm the CTO and co-founder at Inngest. And why should you listen to me? Well, first, I lead an amazing team over at Inngest.
- 0:45
We, we build a system that reliably executes anything from, uh, agents to workflows to context pipelines, whatever you want.
- 0:56
I'm deep building AI infra every day, and on top of that, I'm also building agents. I'm not just pontificating up here in theory about how you should build them.
- 1:07
So speaking of building agents, if you've been building agents for more than six months, you've likely rewritten something, maybe more than once. A new model, a new framework or framework version, a new tool calling standard, a new pattern.
- 1:25
Suddenly, your architecture doesn't fit. It's not a complaint. It's just the reality, right? Things move faster than ever. You can see if you've been to any of the, the sessions or keynotes.
- 1:39
So I want you to think about what you shipped six months ago and how much of it still runs the way that you originally wrote it. Some parts of your code likely survived, but did they survive by accident, or did you design it that way?
- 1:59
Did they survive by design? How did you actually architect your agent? Did you architect it at all? It's okay. Uh, did you use a framework? Did you custom roll something?
- 2:14
Was the architecture an intentional design that you really thought through or just kind of evolved into what you have now? These are all various ways of where people are these days.
- 2:26
So a lot of folks have talked about harness architecture, building harnesses. But mostly, I think a lot of people draw diagrams, and they talk about specific components, draw just nice lines.
- 2:40
It looks very similar, uh, like, you know, kind of simple. But I wanna talk about the conceptual layers when you're building a system like this. This is maybe the mental model, not specific components.
- 2:54
So in my opinion, there are three discrete layers. First, the execution layer. I think of this as the brain. It's where flow, state, durability, retries happen.
- 3:09
Then there's the context layer. This is the knowledge. Right? This is models, prompts, tools, memory. This is the layer that changes the most.
- 3:22
And then there's compute. There are, um-- This is the hands, right? This is sandboxes, runtimes, browsers that you're automating.
- 3:32
And I think these layers are important to consider for the following reason. It's your half-life. And what is half-life, right? It's a scientific term, and it's a scientific term for the time that it takes for your-- for something to decay by half.
- 3:51
And I think that your architecture has a half-life also.
- 3:55
So prompts last weeks, if you're lucky, maybe, maybe a single week. Uh, the models that you use, months, again, if you're lucky.
- 4:04
Uh, but I think that execution can last years if you do it right. So the problem is that I think that most teams couple everything together. And what happens then is that one layer's half-life kinda leaks and drags the other components down.
- 4:21
You're building-- It's technical debt by another name.
- 4:26
So my thesis is think in layers. Decouple them.
- 4:33
So let's talk about what I mean. Teams run into a lot of issues with the approaches that they choose.
- 4:40
You might choose a framework. It might feel super magical to get going. You might grab a pre-built harness from one of the frontier labs or wherever, and-- or you might custom roll the entire system yourself, and it might take months or weeks or whatever.
- 4:55
But I think in a lot of these situations, the abstractions either are not there at all, or they're too high level, or the layers merge. You know, like orchestration is buried deep inside the chain or inside the framework, and you don't know what the heck's happening.
- 5:11
Uh, state might be kept in a sandbox. Uh, retries get tangled with prompts, the prompt logic. And what's hard is you can't re-- you can't, like, swap any of these things out without rewriting almost everything.
- 5:26
So I think you need to embrace the change, know that things are going to change, know that things are moving fast. But I think also with that is I wanna focus on a layer that I think is the stable layer here, where you can invest and think about, get your abstractions right, so you don't need to rewrite
- 5:45
a large component every six months. So I'm talking about the execution layer. I don't think enough people talk about this, and I define it as, uh, the execution layer as being the system responsible for running your code reliably.
- 6:02
Managing how, when, or whether each piece of work completes. And that's independent of the infrastructure that it's on.
- 6:11
So what does this look like in an agent architecture?
- 6:15
So the execution layer manages the full life cycle. You know, plan, call a model, run code, invoke a sub-agent, loop, retry, coordinate. You can swap the model, swap the context, swap the sandbox, and the execution layer should be able to remain the same.
- 6:35
It's just the, the fundamentals. And that's how you, I think, decouple and you build the system that it can e-evolve for change. So let's talk about what the execution layer has to do.
- 6:47
First, resumability. Your execution layer must enable your system to pick up after failures without restarting from the beginning. You know, agents are handling longer running tasks than ever before, so they must be able to resume when an LLM call fails, a tool call fails, everything flakes out.
- 7:09
We all know. So if you have a failed request step thirty-eight,
- 7:16
you should be able to retry and wait, continue onwards instead of having to go from the beginning and you're gonna lose tokens, costs, time, work that your agent might have completed.
- 7:28
So for this, this to work, a three-hour run cannot hold state in memory or in disk. The state must live outside of the work. So this means that the state must be durable and external.
- 7:43
So without it, you might cobble together maybe some manual checkpointing. You might come up with a system that has, like, a log-based approach where you're gonna be, like, hydrating the state back if you have to pick up where you left off.
- 7:57
But I think a lot of those abstractions start leaking into the other layers of your harness, right? It gets harder to change.
- 8:04
So next, a key aspect of the execution layer is that it must enable you to convinal-- combine a lot of invocation patterns. You're gonna need crons. You're gonna need to trigger things with events.
- 8:17
You're gonna need APIs, human-in-the-loop. Sub-agents or dy- or dynamic workflows also must be possible. And you're gonna need to be able to invoke these things synchronously, asynchronously, delaying invocation.
- 8:33
And I think the key here is that flexible execution and orchestration primitives will enable you to build the pattern or the system, the architecture that you actually need. So if not, I think, again, your harness logic starts absorbing other concepts like queues, workers, polling, back-off, scheduling, and now you end up kind of a mess,
- 8:59
bad abstractions. So lastly, execution needs to provide observability across your entire session, not just the LLM calls and the tool calls, but database errors, permissions issues, um, triggers, performance.
- 9:19
So the full session trace across your entire run is essential. So if you can't see the entirety of a trace from the trigger through the whole stack, it's really hard to debug it, let alone improve your agent and keep evolving it.
- 9:37
So what about sandboxes? Sandboxes are s-so hot. So agents need sandboxes. We've kind of, like, settled upon that at this point in time. They need to execute code. They need to browse.
- 9:51
They need to manipulate files. They might work on disk. But a sandbox is ephemeral and stateless by design. So using it for durability with snapshots or something in state, I think is an anti-pattern.
- 10:05
I think it's a, a difficult thing where the state gets lost, and you're kind of trying to peel-- pull the pieces back together.
- 10:13
So I think when you have the execution layer separate, the execution layer is what gives the sandbox its context, its sequence, its durability. So I think of the sandbox as the hands and execution as the brain.
- 10:29
So, so what's next? Right? What are the next six months of architectures that you're gonna need to handle?
- 10:38
So if you're paying attention, you're joining a lot of these sessions, you'll see what architectures are emerging and what each approach brings new engineering requirements. You know, you have background agents,
- 10:52
dynamic workflows, autonomous loops, agent factories, whatever you wanna call it, all these emerging trends that we're seeing the last couple days, they're all long-running. They're asynchronous. They're delegated.
- 11:09
That means that you need to be able to observe them all, uh, down to the core. Right? They need to be inspectable by human and by an agent. So the patterns, as you see, they-- as these things emerge, they must be mixed and combined together.
- 11:26
So what does this all have in common? I think that execution is fundamental to building any of these systems. So let's just take one example here, background agents, right?
- 11:38
They're not request response. There's no person just waiting there for, for the, the, the chatbot to respond. So it ca-- might run for minutes or hours. It's going to have maybe hundreds of calls, uh, tool...
- 11:50
You know, maybe two hundred tool calls. You're gonna probably guarantee to have at least one failure in that. So you can't even debug your background agent that's running asynchronously without the right infrastructure, without the observability, without everything that's going on in that process or multiple processes.
- 12:12
And another example that we have is, is loop architectures, right? And I think we're talking about slash loop commands and whatnot and coding agents, but I think where we're taking this is how are you building, um, actual systems in your products that employ these, these, these approaches?
- 12:30
So, like, what is a loop? Right? A loop is a system that basically just runs continuously or on a schedule, and it's assessing the state of the system against the, the goals that you set or the criteria that you set, and determines what to do next.
- 12:45
So you're gonna need crons, sub-agent, delegation. History needs to be inspectable, and of course, it needs to be reliable because you don't know when things are running or how the system is continuous to-- continuing to evolve.
- 13:00
So the frameworks of three months ago were not designed to handle this, right? Like, you're going to need to design these systems yourself, so I think you're gonna need a proper execution layer.
- 13:12
So let's look at just some examples of code. It's small on here, but in general, let's just look at how you might put it together.
- 13:20
In your loop, you're gonna need some sort of cron. Maybe that cron here is some sort of health check system, runs every thirty minutes, pulls down some key high-level system metrics, and it looks...
- 13:32
You pass to an LLM, you say, "Is this healthy? Is this looking okay? Should we do more?" And is it, is healthy or not? What should we do? And if it isn't healthy, maybe you just invoke a, a triage agent that goes and looks at that individual service and goes and digs more.
- 13:47
This is pretty simple looking, right? Now you have this triage agent itself. You've just triggered it. This should maybe pull some more context, pull some detailed metrics. It's just context.
- 14:00
And then you're passing to the LLM to start that investigation. Putting it in a loop, you're calling some tools, trying to get to the root cause. This may run for a minute.
- 14:09
This ra-- may run for multiple minutes. It needs to spin up a sandbox, uh, maybe clone code, maybe analyze commits. It's gonna be doing a lot of work here, and you don't know when this is gonna run.
- 14:20
It might run, it might not. How do you know? How do you observe that system later?
- 14:24
And then to complete the loop, there-- you may have a reviewer function, right? Like, are we-- Is the system actually performing as expected, right? Like, it's gonna run every week and look at the history of what just happened and evaluate how the triage system is working.
- 14:43
Do we need to adjust the prompts? Do we need to do better metrics? Does this have the data that it needs? Is it doing anything at all? Is it overreacting?
- 14:51
So it needs to be execution aware, and I think it needs to be execution aware or orchestration aware because it needs to be able to pull the logs of the system and understand, all right, what was executed?
- 15:00
What did this agent choose to execute? What sub-agents did it fan out? What, uh, workflows did it call upon? And then it needs to be able to analyze that and understand what happened and come back to you, either make the change itself or tell you, "This is where we performed or underperformed.
- 15:18
This is what we can do, how we can improve the system overall." So this completes this improving loop. It's three functions. It's pretty simple by design, and it's flexible to iterate on.
- 15:29
I know things get always a lot more complex in production, but I think about-- when, when I think you think about the layers this way, it makes things a little bit, little bit easier to think about building such a complex system that feels like it's, you know, something that you might not be able to reach right yet.
- 15:45
So in the spirit of the reviewing functions, what do you think about measuring your agent, how your agent is performing? You know, I think what's interesting is this execution layer sits between what your user's input is and what's happening throughout the system.
- 16:02
So user feedback, actions, um, and the results of the sessions all flow through this execution layer. And I think this makes it, uh, an ideal place because it becomes the hub for observability, and it allows you to be able to score and understand what your agent is doing.
- 16:22
So to iterate on your application, you kinda need all this data connected, and it needs to be aware of how your code is actually executing, right? So I think the, the linking of this is, is extremely important.
- 16:33
So this is what we built Inngest for. We're Durable execution for AI agents. We're the execution layer.
- 16:42
You can plug in any context layer, bring a model, framework, tool, any compute layer, bring whatever sandbox you want, bring whatever browser you want, get durable steps, small primitives, right?
- 16:55
Uh, event triggers, scheduling, agent-to-agent coordination, full session traces, no infra to manage.
- 17:06
And I think, like I just mentioned with scores, we also think that, that, that the middle of the execution layer when your agent is running is a key place to instrument and score your agents.
- 17:18
When you have access to the data that you need that's all flowing through there, or you've just executed something, it's a perfect place to run something after and defer some scoring for, uh, with the whole trace information, maybe the inputs and the outputs.
- 17:33
So again, that's just execution, orchestration, delayed tasks, defers. Um, and since there's also data flowing through, you can also wait for additional events and attach them to the sessions that you're building and un-un-understand did this u- did this actually works, right?
- 17:53
If you're, if you're running this triage, um, was this triage successful? Did it result in an action by the engineering team? Then that means it probably was a positive result.
- 18:02
Instead of a thumbs up, thumbs down, it's like, did we open the PR, right? If it's a research agent, was this research saved? Was it a good report? That is these, these things that are events that you should be able to attach, and when you have a system that can connect all these pieces, I think it's really
- 18:18
makes doing a lot of those things, um, like creating outcome-based scores a lot easier.
- 18:25
So to wrap up, build your harness. I want you to understand the layers of the architecture. We have to embrace the fast pace, pace of change. And I think if you can get your execution layer right and think about the right primitives, everything else can quickly evolve around it.
- 18:44
And the next six months, the next three months will all be a lot easier for you, especially as everything continues to change. You all are here. So thanks for listening.
- 18:54
I'm Dan. Um, come and find me at the Inngest booth. We're right on the other side of this wall. We're very bright and orange. So thanks everyone. Appreciate it. [upbeat music]