{"slug": "agents-don-t-start-with-prompts-they-start-with-events", "title": "Agents Don't Start with Prompts. They Start with Events.", "summary": "A developer argues that AI agents should be designed around events rather than prompts, citing hiring processes as an example. The developer contends that an agent is the complete system that observes facts and generates new facts, not just an LLM connected to tools.", "body_md": "Lately, it seems like everything is an agent. Open LinkedIn and you'll\n\nfind posts about AI Agents. Go to YouTube and you'll see tutorials on\n\nLangChain, OpenAI Agents, CrewAI, AutoGen, and a long list of frameworks\n\npromising to build autonomous agents capable of performing complex\n\ntasks.\n\nAfter consuming enough of this content, it's easy to conclude that an\n\nagent is simply an LLM connected to a set of tools. From my perspective,\n\nthat's the wrong way to think about it.\n\nMost diagrams look something like this:\n\n```\nPrompt\n  ↓\nLLM\n  ↓\nTool\n  ↓\nLLM\n  ↓\nResponse\n```\n\nThere's nothing inherently wrong with this model. In fact, it's\n\nextremely useful for solving a wide range of problems. The problem\n\nbegins when we confuse one implementation pattern with the complete\n\ndefinition of an agent.\n\nTo understand why, let's think about something millions of people do\n\nevery day: hiring.\n\nWhen someone applies for a job, an important fact has occurred. Someone\n\nsubmitted an application, and the recruiting process has begun. From\n\nthat moment on, many things can happen: the resume is analyzed, an\n\ninterview is scheduled, feedback is collected, experience is validated,\n\nan offer is approved, and eventually a decision is made.\n\nIf we represented that process as events, it might look something like\n\nthis:\n\n```\nCandidateApplied\n        ↓\nResumeAnalyzed\n        ↓\nInterviewScheduled\n        ↓\nInterviewCompleted\n        ↓\nFeedbackReceived\n        ↓\nOfferSent\n```\n\nNow let's ask an interesting question: **where is the agent?**\n\nUnder the current popular paradigm, we might say the agent is the model\n\nthat analyzed the resume. But that explanation is incomplete. Resume\n\nanalysis represents only a small part of the overall process. The\n\ncomplete system observed new information, evaluated the current\n\nsituation, and executed actions to move the process forward.\n\nIn other words, the agent didn't appear when someone wrote a prompt.\n\n**The agent started acting when an event occurred.**\n\nThis distinction may seem subtle, but it completely changes how we\n\ndesign systems.\n\nA recruiter doesn't wake up in the morning waiting for prompts. They\n\nreceive events. A candidate applies for a job. An interviewer submits\n\nfeedback. A manager approves a hire. A candidate rejects an offer. Each\n\nnew fact changes the state of the process and creates the need for new\n\ndecisions.\n\nThe same thing happens in almost every organization. A customer sends an\n\nemail. A payment is approved. A shipment changes status. A document is\n\nuploaded. A support ticket is created.\n\n**The real world is made of events, not prompts.**\n\nOf course, an AI model may become involved at some point. It might\n\nanalyze a resume, classify an email, summarize a document, or determine\n\na customer's intent. It might even be an LLM. But regardless of the\n\ntechnology being used, the agent is still the complete system. The model\n\nparticipates in a decision; it doesn't define the agent by itself.\n\nImagine, for example, that a model analyzes a resume and determines that\n\nthe candidate meets the minimum requirements for the position. The\n\nimportant output isn't the textual response generated by the model. The\n\nimportant output is the new fact that was discovered:\n\n```\nResumeAnalyzed\nCandidateQualified\n```\n\nFrom that point, the system can continue operating. It can schedule an\n\ninterview, request additional information, or notify the recruiter. The\n\nmodel participated in a decision, but the overall flow remains\n\nevent-driven.\n\nAnd this is where I think a more useful definition of an agent begins to\n\nemerge.\n\nIf we look at the complete process, the agent isn't constantly producing\n\nresponses. It's observing facts, evaluating information, and generating\n\nnew facts that allow the process to continue moving forward.\n\nEach decision generates new facts that feed the next step in the\n\nprocess.\n\nFrom this perspective, an agent stops looking like a conversation with\n\ntools and starts looking like an **event-driven decision-making\nsystem**.\n\nThose decisions can be made using business rules, machine learning\n\nmodels, LLMs, or a combination of them. The important thing isn't the\n\ntechnology used to make the decision. The important thing is that there\n\nis a system capable of observing what happens, evaluating the available\n\ninformation, and reacting accordingly.\n\nI think this is where the current conversation about agents becomes\n\nconfusing. We've spent enormous amounts of time talking about prompts,\n\ntool calling, context windows, and reasoning, but relatively little time\n\ntalking about what causes an agent to act in the first place.\n\nBecause before a prompt exists, something has usually already happened:\n\n**The event is the trigger.** The prompt, when one exists, is simply a\n\ntool used during the evaluation process.\n\nThat's why I'm increasingly convinced that we've started this\n\nconversation from the wrong end. Agents are much more than prompt\n\nworkflows connected to tools. They are systems that observe events, make\n\ndecisions, and execute actions.\n\nSome of those decisions may be made by business rules. Others by machine\n\nlearning models. Others by LLMs. What matters is that the agent isn't\n\ndefined by the technology used to make a decision, but by its ability to\n\nreact to what happens.\n\nMaybe we're not just beginning to build agents. Maybe we've been\n\nbuilding them for years, and we've simply added new tools that help them\n\nmake better decisions.\n\nPrompts matter. LLMs matter too. But neither answers the fundamental\n\nquestion:\n\n**What caused the agent to act?**\n\nMost of the time, the answer isn't a prompt. It's an event.\n\n**Agents don't start with prompts. They start with events.**", "url": "https://wpnews.pro/news/agents-don-t-start-with-prompts-they-start-with-events", "canonical_source": "https://dev.to/clandro89/agents-dont-start-with-prompts-they-start-with-events-3524", "published_at": "2026-08-13 04:02:09+00:00", "updated_at": "2026-08-13 04:21:43.601423+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research"], "entities": ["OpenAI", "LangChain", "CrewAI", "AutoGen"], "alternates": {"html": "https://wpnews.pro/news/agents-don-t-start-with-prompts-they-start-with-events", "markdown": "https://wpnews.pro/news/agents-don-t-start-with-prompts-they-start-with-events.md", "text": "https://wpnews.pro/news/agents-don-t-start-with-prompts-they-start-with-events.txt", "jsonld": "https://wpnews.pro/news/agents-don-t-start-with-prompts-they-start-with-events.jsonld"}}