<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Gex Melk</title>
    <description>The latest articles on DEV Community by Gex Melk (@gex_melk_8e5bfbf8b7477871).</description>
    <link>https://dev.to/gex_melk_8e5bfbf8b7477871</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4249192%2Ffaaaf8d1-32fb-46be-8f8d-ecacb28ae25a.png</url>
      <title>DEV Community: Gex Melk</title>
      <link>https://dev.to/gex_melk_8e5bfbf8b7477871</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/gex_melk_8e5bfbf8b7477871"/>
    <language>en</language>
    <item>
      <title>AI agents for business: what they are and how they work</title>
      <dc:creator>Gex Melk</dc:creator>
      <pubDate>Tue, 17 Nov 2026 07:41:47 +0000</pubDate>
      <link>https://dev.to/gex_melk_8e5bfbf8b7477871/ai-agents-for-business-what-they-are-and-how-they-work-4jc5</link>
      <guid>https://dev.to/gex_melk_8e5bfbf8b7477871/ai-agents-for-business-what-they-are-and-how-they-work-4jc5</guid>
      <description>&lt;p&gt;"AI agent" has become one of those terms that means everything and nothing. Vendors use it for chat widgets, for Zapier flows with a GPT step in the middle, and for systems that genuinely plan and act on their own. If you are a developer asked to "add an agent" to a business process, it helps to have a precise definition and a clear picture of what you are actually going to build.&lt;/p&gt;

&lt;p&gt;This post is a practical overview: what separates an agent from a chatbot or a classic automation, where agents tend to pay off first, what the code around the model looks like, and how to roll one out without waking up to a mess.&lt;/p&gt;

&lt;h2&gt;
  
  
  Chatbot, automation, agent: the short version
&lt;/h2&gt;

&lt;p&gt;A &lt;strong&gt;chatbot&lt;/strong&gt; takes a question and returns an answer. It understands free text, but it rarely does anything outside the conversation.&lt;/p&gt;

&lt;p&gt;A &lt;strong&gt;classic automation&lt;/strong&gt; (n8n, Zapier, Make, a cron job with some scripts) does things in your systems, but only along a path you defined in advance. It does not understand free text and breaks on cases you did not anticipate.&lt;/p&gt;

&lt;p&gt;An &lt;strong&gt;AI agent&lt;/strong&gt; combines the two. It understands an open-ended task, and it can take actions: query a database, read a CRM record, send a message, create a document. Crucially, it decides the next step itself, based on what the previous step returned.&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;&lt;/th&gt;
&lt;th&gt;Chatbot&lt;/th&gt;
&lt;th&gt;Classic automation&lt;/th&gt;
&lt;th&gt;AI agent&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Understands free text&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Acts in your systems&lt;/td&gt;
&lt;td&gt;Rarely&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;td&gt;Yes&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Handles unexpected cases&lt;/td&gt;
&lt;td&gt;Partly&lt;/td&gt;
&lt;td&gt;No&lt;/td&gt;
&lt;td&gt;Yes, within limits you set&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Good fit for&lt;/td&gt;
&lt;td&gt;FAQ, simple answers&lt;/td&gt;
&lt;td&gt;Fixed, predictable steps&lt;/td&gt;
&lt;td&gt;Multi-step tasks that need judgement&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;The rule of thumb: if you can draw the whole process as a flowchart with no "it depends" boxes, use a normal automation. It is cheaper and more predictable. Reach for an agent when the steps depend on what the data says.&lt;/p&gt;

&lt;h2&gt;
  
  
  The core idea: a loop
&lt;/h2&gt;

&lt;p&gt;Strip away the marketing and an agent is a loop:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Send the task (plus context) to the model, along with a list of tools it is allowed to use.&lt;/li&gt;
&lt;li&gt;The model either answers, or asks to call a tool with specific arguments.&lt;/li&gt;
&lt;li&gt;Your code runs that tool and sends the result back to the model.&lt;/li&gt;
&lt;li&gt;Repeat until the model says it is done, or until a limit is hit.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Here is the skeleton in Python-flavoured pseudocode:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;run_agent&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;task&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;max_steps&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;10&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="n"&gt;messages&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;{&lt;/span&gt;&lt;span class="s2"&gt;"role"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"user"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"content"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;task&lt;/span&gt;&lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;

    &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;step&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="nb"&gt;range&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;max_steps&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="n"&gt;reply&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;llm&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;chat&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;messages&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;messages&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;schemas&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
        &lt;span class="n"&gt;messages&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

        &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;tool_calls&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
            &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;text&lt;/span&gt;                &lt;span class="c1"&gt;# model decided it is done&lt;/span&gt;

        &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;call&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;reply&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;tool_calls&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
            &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;needs_approval&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
                &lt;span class="n"&gt;result&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;ask_human&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;     &lt;span class="c1"&gt;# pause for a person&lt;/span&gt;
            &lt;span class="k"&gt;else&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
                &lt;span class="n"&gt;result&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;tools&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;run&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;args&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
            &lt;span class="nf"&gt;log&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;step&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;result&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
            &lt;span class="n"&gt;messages&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nf"&gt;tool_result&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;call&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nb"&gt;id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;result&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="s2"&gt;"Stopped: step limit reached"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That is genuinely most of it. The model never touches your CRM directly. It only produces a structured request ("call &lt;code&gt;find_customer&lt;/code&gt; with &lt;code&gt;email=...&lt;/code&gt;"), and your code decides whether and how to execute it. This is the property that makes agents controllable.&lt;/p&gt;

&lt;h2&gt;
  
  
  What lives around the model (the harness)
&lt;/h2&gt;

&lt;p&gt;The model on its own turns text into text. Everything that turns it into an agent is the code around it, often called the harness:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;The loop&lt;/strong&gt; above, with step limits and timeouts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Tools:&lt;/strong&gt; thin, well-described wrappers over your CRM, e-mail, messengers, databases and internal APIs. Good tool names and descriptions matter more than clever prompts.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Context and memory:&lt;/strong&gt; which documents, conversation history and business rules the model sees on each step. Often this is a RAG layer over your docs.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Permissions:&lt;/strong&gt; what the agent may do on its own and what requires a human to approve. "Read order status" can be automatic; "issue a refund" probably should not be.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Logging and monitoring:&lt;/strong&gt; every step, every tool call, its cost and its outcome, stored somewhere you can query.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Once this layer exists and is reused, each new agent becomes mostly configuration: a role, a set of data sources and a set of tools. That is where teams get speed and consistency.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where agents pay off first
&lt;/h2&gt;

&lt;p&gt;In practice, the first wins tend to come from processes that are repetitive, text-heavy and slightly messy:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Lead handling.&lt;/strong&gt; Reply to an inbound request within a minute, ask a couple of qualifying questions, write the lead into the CRM and notify sales.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Customer support.&lt;/strong&gt; Answer from your documentation, check order or account status through an API, and escalate complex cases to a person with a summary attached.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Inbox and documents.&lt;/strong&gt; Sort incoming e-mail, extract fields from invoices or contracts, prepare draft replies.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Internal assistant.&lt;/strong&gt; Answer staff questions from policies, wiki pages and CRM data.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Reporting.&lt;/strong&gt; Pull numbers from several systems and write a short weekly summary.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Content operations.&lt;/strong&gt; Draft product descriptions or replies for a human to review and publish.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Notice that none of these require the agent to be fully autonomous on day one.&lt;/p&gt;

&lt;h2&gt;
  
  
  Rolling one out safely
&lt;/h2&gt;

&lt;p&gt;A few habits prevent most of the horror stories:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Start with one process and measure the baseline.&lt;/strong&gt; Time per task, response time, error rate. Without a baseline you cannot tell if the agent helps.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Least privilege.&lt;/strong&gt; Give the agent only the tools and scopes it needs for this process. Read-only where possible.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Human approval for risky actions.&lt;/strong&gt; Payments, deletions, and messages to important clients go through a person at first.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Log everything and review samples weekly.&lt;/strong&gt; Read real transcripts, not just dashboards.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Expand autonomy gradually.&lt;/strong&gt; Remove approval steps only after the agent has proven itself on real cases.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Wrapping up
&lt;/h2&gt;

&lt;p&gt;An AI agent is not magic and not just a chatbot. It is a model in a loop, calling tools you define, inside guardrails you control. The model provides the judgement; your harness provides the safety, the integrations and the observability. Get the harness right and adding the next agent becomes a configuration task rather than a new project.&lt;/p&gt;

&lt;p&gt;Written by the Amsoft team (amsoft.am), we build AI agents and automation for businesses.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>llm</category>
      <category>agents</category>
      <category>automation</category>
    </item>
  </channel>
</rss>
