Want smarter insights in your inbox? Sign up for our weekly newsletters to get only what matters to enterprise AI, data, and security leaders.<\/em> Subscribe Now<\/em><\/p>\n\n\n\n Everywhere you look, people are talking about AI agents like they\u2019re just a prompt away from replacing entire departments. The dream is seductive: Autonomous systems that can handle anything you throw at them, no guardrails, no constraints, just give them your AWS credentials and they\u2019ll solve all your problems. But the reality is that\u2019s just not how the world works, especially not in the enterprise, where reliability isn\u2019t optional.<\/p>\n\n\n\n Even if an agent is 99% accurate, that\u2019s not always good enough. If it\u2019s optimizing food delivery routes, that means one out of every hundred orders ends up at the wrong address. In a business context, that kind of failure rate isn\u2019t acceptable. It\u2019s expensive, risky and hard to explain to a customer or regulator.<\/p>\n\n\n\n In real-world environments like finance, healthcare and operations, the AI systems that actually deliver value don\u2019t look anything like these frontier fantasies. They aren\u2019t improvising in the open world; they\u2019re solving well-defined problems with clear inputs and predictable outcomes.<\/p>\n\n\n\n If we keep chasing open-world problems with half-ready technology, we\u2019ll burn time, money and trust. But if we focus on the problems right in front of us, the ones with clear ROI and clear boundaries, we can make AI work today.<\/p>\n\n\n\n This article is about cutting through the hype and building AI agents that actually ship, run and help.<\/p>\n\n\n\n The tech industry loves a moonshot (and for the record, I do too). Right now, the moonshot is open-world AI \u2014 agents that can handle anything, adapt to new situations, learn on the fly and operate with incomplete or ambiguous information. It\u2019s the dream of general intelligence: Systems that can not only reason, but improvise.<\/p>\n\n\n\n Open-world problems are defined by what we don\u2019t<\/em> know.<\/p>\n\n\n\n More formally, drawing from research defining these complex environments, a fully open world is characterized by two core properties:\u00a0<\/p>\n\n\n\n In such environments, the AI operates with incomplete information; it cannot assume that what isn\u2019t known to be true is false, it\u2019s simply unknown. The AI is expected to adapt to these unforeseen changes and novel tasks as it navigates the world. This presents an incredibly difficult set of problems for current AI capabilities.<\/p>\n\n\n\n In contrast, closed-world problems are ones where the scope is known, the rules are clear and the system can assume it has all the relevant data. If something isn\u2019t explicitly true, it can be treated as false. These are the kinds of problems most businesses actually face every day: invoice matching, contract validation, fraud detection, claims processing, inventory forecasting.<\/p>\n\n\n\n These aren\u2019t the use cases that typically make headlines, but they\u2019re the ones businesses actually care about solving.<\/p>\n\n\n\n However, the hype is harmful: By setting the bar at open-world general intelligence, we make enterprise AI feel inaccessible. Leaders hear about agents that can do everything, and they freeze, because they don\u2019t know where to start. The problem feels too big, too vague, too risky.<\/p>\n\n\n\n It\u2019s like trying to design autonomous vehicles before we\u2019ve even built a working combustion engine. The dream is exciting, but skipping the fundamentals guarantees failure.<\/p>\n\n\n\n Open-world problems make for great demos and even better funding rounds. But closed-world problems are where the real value is today. They\u2019re solvable, testable and automatable. And they\u2019re sitting inside every enterprise, just waiting for the right system to tackle them.<\/p>\n\n\n\n The question isn\u2019t whether AI will solve open-world problems eventually. The question is: What can you actually deploy right now that makes your business faster, smarter and more reliable?<\/p>\n\n\n\n When people imagine AI agents today, they tend to picture a chat window. A user types a prompt, and the agent responds with a helpful answer (maybe even triggers a tool or two). That\u2019s fine for demos and consumer apps, but it\u2019s not how enterprise AI will actually work in practice.<\/p>\n\n\n\n In the enterprise, most useful agents aren\u2019t user-initiated, they\u2019re autonomous.<\/p>\n\n\n\n They don\u2019t sit idly waiting for a human to prompt them. They\u2019re long-running processes that react to data as it flows through the business. They make decisions, call services and produce outputs, continuously and asynchronously, without needing to be told when to start.<\/p>\n\n\n\n Imagine an agent that monitors new invoices. Every time an invoice lands, it extracts the relevant fields, checks them against open purchase orders, flags mismatches and either routes the invoice for approval or rejection, without anyone asking it to do so. It just listens for the event (\u201cnew invoice received\u201d) and goes to work.<\/p>\n\n\n\n Or think about customer onboarding. An agent might watch for the moment a new account is created, then kick off a cascade: verify documents, run know-your-customer (KYC) checks, personalize the welcome experience and schedule a follow-up message. The user never knows the agent exists. It just runs. Reliably. In real time.<\/p>\n\n\n\n This is what enterprise agents look like:<\/p>\n\n\n\n You don\u2019t build these agents by fine-tuning a giant model. You build them by wiring together existing models, tools and logic. It\u2019s a software engineering problem, not a modeling one.<\/p>\n\n\n\n At their core, enterprise agents are just modern microservices with intelligence. You give them access to events, give them the right context and let a language model drive the reasoning.<\/p>\n\n\n\n Agent = Event-driven microservice + context data + LLM<\/strong><\/p>\n\n\n\n Done well, that\u2019s a powerful architectural pattern. It\u2019s also a shift in mindset. Building agents isn\u2019t about chasing artificial general intelligence (AGI). It\u2019s about decomposing real problems into smaller steps, then assembling specialized, reliable components that can handle them, just like we\u2019ve always done in good software systems.<\/p>\n\n\n\n If this sounds familiar, it should. We\u2019ve been here before.<\/p>\n\n\n\n When monoliths couldn\u2019t scale, we broke them into microservices. When synchronous APIs led to bottlenecks and brittle systems, we turned to event-driven architecture. These were hard-won lessons from decades of building real-world systems. They worked because they brought structure and determinism to complex systems.<\/p>\n\n\n\n I worry that we\u2019re starting to forget that history and repeat the same mistakes in how we build AI.<\/p>\n\n\n\n Because this isn\u2019t a new problem. It\u2019s the same engineering challenge, just with new components. And right now, enterprise AI needs the same principles that got us here: clear boundaries, loose coupling and systems designed to be reliable from the start.<\/p>\n\n\n\n The problems worth solving in most businesses are closed-world: Problems with known inputs, clear rules and measurable outcomes. But the models we\u2019re using, especially LLMs, are inherently non-deterministic. They\u2019re probabilistic by design. The same input can yield different outputs depending on context, sampling or temperature.<\/p>\n\n\n\n That\u2019s fine when you\u2019re answering a prompt. But when you\u2019re running a business process? That unpredictability is a liability.<\/p>\n\n\n\n So if you want to build production-grade AI systems, your job is simple: Wrap non-deterministic models in deterministic infrastructure.<\/p>\n\n\n\n Too many teams are reinventing runtime orchestration with every agent, letting the LLM decide what to do next, even when the steps are known ahead of time. You\u2019re just making your life harder.<\/p>\n\n\n\n Event-driven multi-agent systems break the problem into smaller steps. When you assign each one to a purpose-built agent and trigger them with structured events, you end up with a loosely coupled, fully traceable system that works the way enterprise systems are supposed to work: With reliability, accountability and clear control.<\/p>\n\n\n\n And because it\u2019s event-driven:<\/p>\n\n\n\n Closed-world problems don\u2019t require magic. They need solid engineering. And that means combining the flexibility of LLMs with the structure of good software engineering. If something can be made deterministic, make it deterministic. Save the model for the parts that actually require judgment.<\/p>\n\n\n\n That\u2019s how you build agents that don\u2019t just look good in demos but actually run, scale and deliver in production.<\/p>\n\n\n\n One of the most overlooked challenges in building agents is testing, but it is absolutely essential for the enterprise.<\/p>\n\n\n\n In an open-world context, it\u2019s nearly impossible to do well. The problem space is unbounded so the inputs can be anything, the desired outputs are often ambiguous and even the criteria for success might shift depending on context.<\/p>\n\n\n\n How do you write a test suite for a system that can be asked to do almost anything? You can\u2019t.<\/p>\n\n\n\n That\u2019s why open-world agents are so hard to validate in practice. You can measure isolated behaviors or benchmark narrow tasks, but you can\u2019t trust the system end-to-end unless you\u2019ve somehow seen it perform across a combinatorially large space of situations, which no one has.<\/p>\n\n\n\n In contrast, closed-world problems make testing tractable. The inputs are constrained. The expected outputs are definable. You can write assertions. You can simulate edge cases. You can know what \u201ccorrect\u201d looks like.<\/p>\n\n\n\n And if you go one step further, decomposing your agent\u2019s logic into smaller, well-scoped components using an event-driven architecture, it gets even more tractable. Each agent in the system has a narrow responsibility. Its behavior can be tested independently, its inputs and outputs mocked or replayed, and its performance evaluated in isolation.<\/p>\n\n\n\n When the system is modular, and the scope of each module is closed-world, you can build test sets that actually give you confidence.<\/p>\n\n\n\n This is the foundation for trust in production AI.<\/p>\n\n\n\n The future of AI in the enterprise doesn\u2019t start with AGI. It starts with automation that works. That means focusing on closed-world problems that are structured, bounded and rich with opportunity for real impact.<\/p>\n\n\n\n You don\u2019t need an agent that can do everything. You need a system that can reliably do something:<\/p>\n\n\n\n Those wins add up. They reduce costs, free up time and build trust in AI as a dependable part of the stack.<\/p>\n\n\n\n And getting there doesn\u2019t require breakthroughs in prompt engineering or betting on the next model to magically generalize. It requires doing what good engineers have always done: Breaking problems down, building composable systems and wiring components together in ways that are testable and observable.<\/p>\n\n\n\n Event-driven multi-agent systems aren\u2019t a silver bullet, they\u2019re just a practical architecture for working with imperfect tools in a structured way. They let you isolate where intelligence is needed, contain where it\u2019s not and build systems that behave predictably even when individual parts don\u2019t.<\/p>\n\n\n\n This isn\u2019t about chasing the frontier. It\u2019s about applying basic software engineering to a new class of problems.<\/p>\n\n\n\n Sean Falconer is Confluent\u2019s AI entrepreneur in residence.<\/em><\/p>\n Daily insights on business use cases with VB Daily<\/strong><\/p>\n If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI.<\/p>\n \n\t\t\t\t\tThanks for subscribing. Check out more VB newsletters here.\n\t\t\t\t<\/p>\n An error occured.<\/p>\n<\/p><\/div>\n
\n<\/div>The problem with the open world hype<\/h2>\n\n\n\n
What makes a problem \u201copen world\u201d?<\/h3>\n\n\n\n
\n
Most enterprise problems aren\u2019t like this<\/h3>\n\n\n\n
Feature<\/strong><\/td> Open world<\/strong><\/td> Closed world<\/strong><\/td><\/tr> Scope<\/td> Unbounded<\/td> Well-defined<\/td><\/tr> Knowledge<\/td> Incomplete<\/td> Complete (within domain)<\/td><\/tr> Assumptions<\/td> Unknown \u2260 false<\/td> Unknown = false<\/td><\/tr> Tasks<\/td> Emergent, not predefined<\/td> Fixed, repetitive<\/td><\/tr> Testability<\/td> Extremely hard<\/td> Well-bounded<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n The risk of hype and inaction<\/h3>\n\n\n\n
Solve what\u2019s right in front of you<\/h3>\n\n\n\n
What enterprise agents actually look like<\/h2>\n\n\n\n
\n
![\"\"](\"https:\/\/venturebeat.com\/wp-content\/uploads\/2025\/07\/image1.png?w=800\")
We\u2019ve solved this kind of problem before<\/h2>\n\n\n\n
AI models are not deterministic, but your systems can be<\/h3>\n\n\n\n
Build determinism around the model<\/h3>\n\n\n\n
\n
Where event-driven multi-agent systems shine<\/h3>\n\n\n\n
\n
Don\u2019t chase magic<\/h3>\n\n\n\n
Why testing is so much harder in an open world<\/h2>\n\n\n\n
Building the right foundation<\/h2>\n\n\n\n
\n
![\"\"\/](\"https:\/\/venturebeat.com\/wp-content\/themes\/vb-news\/brand\/img\/vb-daily-phone.png\")
\n\t\t\t\t<\/div>\n<\/p><\/div>\n<\/div>\t\t\t<\/div>\r\n