<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Antonio D&#39;Souza on Medium]]></title>
        <description><![CDATA[Stories by Antonio D&#39;Souza on Medium]]></description>
        <link>https://medium.com/@ab6714462?source=rss-891b48146094------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*p0GC63uhyuwDor1K0x-uTA.png</url>
            <title>Stories by Antonio D&amp;#39;Souza on Medium</title>
            <link>https://medium.com/@ab6714462?source=rss-891b48146094------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 16:30:28 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ab6714462/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I Started Taking Agentic AI Seriously When I Realised AI Could Do More Than Answer]]></title>
            <link>https://medium.com/@ab6714462/i-started-taking-agentic-ai-seriously-when-i-realised-ai-could-do-more-than-answer-47870fb16ed0?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/47870fb16ed0</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[upgrad]]></category>
            <category><![CDATA[genai]]></category>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Thu, 03 Sep 2026 09:40:40 GMT</pubDate>
            <atom:updated>2026-09-03T09:40:40.476Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Agentic AI planning, reasoning, using tools, and acting through upGrad’s Agentic AI learning programmes." src="https://cdn-images-1.medium.com/max/1024/1*BX-WDqzs7CtL9BdMSpSRQA.png" /><figcaption>Moving beyond prompts to intelligent action with upGrad’s Agentic AI programmes.</figcaption></figure><p>For a while, I thought the biggest change AI had brought to work was that we could ask a machine almost anything and get an answer in seconds.</p><p>Write an email.</p><p>Summarise a report.</p><p>Explain a concept.</p><p>Generate code.</p><p>Analyse some information.</p><p>Then I started hearing the phrase “Agentic AI” everywhere.</p><p>At first, I assumed it was just another way of describing more advanced generative AI.</p><p>It isn’t.</p><p>The more I explored it, the more I realised that the interesting part of Agentic AI isn’t simply that AI can produce better answers. It’s that an AI system can potentially <strong>plan, use tools, make decisions and execute multiple steps to reach an objective</strong>.</p><p>That changed the way I started looking at AI.</p><p>I began exploring Agentic AI more seriously, including structured programmes through upGrad, and a few concepts stood out to me.</p><h3>1. An agent isn’t just a chatbot with a longer prompt</h3><p>This was probably the first distinction I had to make.</p><p>A chatbot generally waits for an instruction and responds.</p><p>An agent can be designed around an objective.</p><p>Suppose I ask an AI:</p><p>“Summarise these customer reviews.”</p><p>That’s a straightforward generative AI task.</p><p>Now imagine asking:</p><p>“Analyse this month’s customer reviews, identify the biggest recurring complaints, compare them with last month’s data, find possible causes and prepare recommendations for the customer experience team.”</p><p>That involves several steps.</p><p>The system has to retrieve information, analyse it, decide what is relevant and organise the result.</p><p>The difference isn’t necessarily the language model.</p><p>It’s the <strong>workflow surrounding the model</strong>.</p><p>That’s what made Agentic AI much more interesting to me.</p><h3>2. Tool use changes everything</h3><p>An AI model can be incredibly capable and still be limited by what information and capabilities it has access to.</p><p>Tools change that.</p><p>An agent can potentially interact with APIs, databases, search systems, documents and other software.</p><p>This means AI doesn’t have to exist as a standalone conversational interface.</p><p>It can become part of a larger system.</p><p>For example, an agent working in customer support could retrieve information from a knowledge base, check an order database, identify the customer’s issue and then trigger an appropriate workflow.</p><p>Suddenly, we’re no longer talking about AI simply generating text.</p><p>We’re talking about AI <strong>interacting with software</strong>.</p><p>That is a much bigger shift.</p><p>current <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/">Agentic AI</a> learning offerings via <a href="https://www.upgrad.com/">upGrad</a> cover areas including prompting, RAG pipelines, multi-agent collaboration and connecting agents with external tools, APIs and databases.</p><figure><img alt="Agentic AI workflow showing planning, tool use, reasoning, and controlled autonomy through upGrad’s Agentic AI programmes." src="https://cdn-images-1.medium.com/max/1024/1*JhlphmmsT14dY02GAyApaw.png" /><figcaption>From answering questions to planning and acting with upGrad’s Agentic AI programmes.</figcaption></figure><h3>3. RAG made me think differently about AI’s memory</h3><p>Another concept that kept appearing was Retrieval-Augmented Generation, or RAG.</p><p>I found this particularly interesting because it addresses a very practical problem.</p><p>A general AI model may know a huge amount, but businesses usually need answers based on their own information.</p><p>Internal documents.</p><p>Product information.</p><p>Policies.</p><p>Research.</p><p>Customer data.</p><p>Company knowledge.</p><p>RAG allows an AI system to retrieve relevant information and provide that context to the model before generating an answer.</p><p>So instead of asking only:</p><p>“How smart is the model?”</p><p>we start asking:</p><p>“What information can the system access?”</p><p>and:</p><p>“Can it retrieve the right information at the right time?”</p><p>For enterprise AI, those questions can matter just as much as the model itself.</p><h3>4. Multi-agent systems made the idea even more interesting</h3><p>Then there is the idea of multiple specialised agents working together.</p><p>Instead of creating one system that does everything, different agents can have different responsibilities.</p><p>One could research.</p><p>Another could analyse.</p><p>Another could verify.</p><p>Another could coordinate the workflow.</p><p>Conceptually, it starts looking less like a chatbot and more like a digital team.</p><p>Of course, making this reliable is not simple.</p><p>Agents can make incorrect decisions. They can use the wrong tool. They can misunderstand instructions. They can produce confident but incorrect outputs.</p><p>So building an agentic system also means thinking about evaluation, guardrails, permissions, failure handling and human oversight.</p><p>That was an important realisation for me.</p><p>Agentic AI isn’t just about autonomy.</p><p>It’s about <strong>controlled autonomy</strong>.</p><h3>5. The hardest part isn’t making an AI demo</h3><p>This is probably what changed my expectations the most.</p><p>It’s relatively easy to build an impressive AI demo.</p><p>Give an LLM a prompt.</p><p>Connect a tool.</p><p>Add a nice interface.</p><p>Show what it can do.</p><p>But production systems have to deal with much more.</p><p>What happens when the tool fails?</p><p>What happens when the retrieved information is wrong?</p><p>What happens when the agent takes an unnecessary action?</p><p>How much does every interaction cost?</p><p>How quickly does the system respond?</p><p>How do you monitor it?</p><p>Who is responsible when it makes a mistake?</p><p>Those are engineering questions as much as AI questions.</p><p>And that’s why I don’t think learning Agentic AI should mean learning a collection of frameworks and calling it a day.</p><p>The frameworks will change.</p><p>The underlying principles matter more.</p><h3>What I’m paying attention to now</h3><p>My biggest takeaway from exploring Agentic AI is that AI seems to be moving through an interesting progression.</p><p>First, we asked AI to <strong>generate</strong>.</p><p>Then we started asking it to <strong>understand and retrieve</strong>.</p><p>Now we’re increasingly asking it to <strong>reason, coordinate and act</strong>.</p><p>That doesn’t mean humans disappear from the process.</p><p>In many situations, the opposite may be true.</p><p>The better question is where humans should remain in the loop and where AI can safely handle repetitive or multi-step work.</p><p>That’s the part of Agentic AI I find most interesting.</p><p>Not the idea of replacing every task with an autonomous agent.</p><p>But the possibility of designing systems where AI can take responsibility for parts of a workflow while humans focus on judgement, creativity and decisions that actually require them.</p><p>I started looking at Agentic AI because it sounded like another AI buzzword.</p><p>I ended up finding a much bigger idea underneath it:</p><p><strong>What if AI isn’t just something we talk to, but something we design to work alongside us?</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=47870fb16ed0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[5 Things I Got Wrong Building an AI Agent for My Team’s Support Queue]]></title>
            <link>https://medium.com/@ab6714462/5-things-i-got-wrong-building-an-ai-agent-for-my-teams-support-queue-1a9cee27393f?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a9cee27393f</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[upgrad]]></category>
            <category><![CDATA[learn-online]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Mon, 24 Aug 2026 05:54:47 GMT</pubDate>
            <atom:updated>2026-08-24T05:54:47.072Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="AI agent support triage for the Agentic AI Programme through upGrad." src="https://cdn-images-1.medium.com/max/1024/1*dAa2YO-N_X2RxYWiI9Wfng.png" /><figcaption>Building reliable AI agents with the Agentic AI Programme through upGrad.</figcaption></figure><p>Our support team was drowning. Not in ticket volume exactly, in the boring middle part, reading each ticket, checking it against three different internal docs, deciding which team should actually own it, then manually reassigning it. My manager suggested, almost offhand, that I try building something to automate the triage step. I said sure, expecting a weekend project. It took three weeks, and taught me more about agentic AI than any explainer video had.</p><p>Here’s what I actually got wrong, in order.</p><p><strong>1. I assumed the model would infer context I never gave it</strong></p><p>My first version worked fine on obvious tickets, “password reset” clearly goes to IT. It fell apart on anything ambiguous, a billing complaint that was actually a bug report in disguise, for instance. I’d assumed the model would somehow infer our internal routing logic. It didn’t, because I’d never actually written that logic down anywhere it could access. Turns out an agent is only as good as the context you explicitly hand it, intuition doesn’t transfer just because it feels obvious to a human.</p><p><strong>2. I gave it too much autonomy before it earned any trust</strong></p><p>Version two let the agent auto-assign tickets without any human review, mostly because reviewing every single one felt like defeating the point of automating it. Within two days it had misrouted a genuinely urgent security ticket to general support, where it sat for six hours before anyone noticed. Nothing catastrophic happened, but it easily could have. I’d skipped the step where you let an agent prove itself on low-stakes decisions before handing it anything that actually matters.</p><figure><img alt="AI support agent workflow with guardrails and human review from the Agentic AI Programme through upGrad." src="https://cdn-images-1.medium.com/max/1024/1*XbUJVaX5YyOUJNUcxbM0Ng.png" /><figcaption>Applying agentic AI principles through Agentic AI Programme via upGrad</figcaption></figure><p><strong>3. I didn’t build in a way to catch its own mistakes</strong></p><p>The agent had no mechanism to flag its own uncertainty. It would confidently route a genuinely confusing ticket the same way it routed an obvious one, no difference in tone, no “I’m not fully sure about this one.” Adding a confidence threshold, where anything below a certain certainty got flagged for a human instead of auto-routed, fixed more problems than any other single change I made.</p><p><strong>4. I underestimated how much the “boring” parts mattered</strong></p><p>I spent most of my initial effort on the routing logic itself, the interesting part. I spent almost no time on logging, on tracking every decision the agent made and why. That came back to bite me the first time something went wrong and I had no way to actually diagnose what happened, I just had a wrong outcome and no trail leading to it.</p><p><strong>5. I treated “it works in testing” as the finish line</strong></p><p>My test set was clean, well-formatted example tickets. Real tickets are messy, typos, half-finished sentences, people describing three unrelated problems in one message. The agent that performed beautifully in testing struggled noticeably against actual production chaos, which taught me that testing on realistic mess matters far more than testing on a large volume of clean examples.</p><p><strong>Where things actually stand now</strong></p><p>The agent handles about 70 percent of ticket routing on its own now, correctly, with the remaining 30 percent flagged for a human based on that confidence threshold. Support response time dropped noticeably, not because the agent is smarter than a person, but because it removed the repetitive part of the job that used to eat hours nobody particularly enjoyed spending.</p><p>I picked up most of what actually fixed these mistakes through an <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/">Agentic AI programme through upGrad</a> that I went through afterward, mainly to understand properly what I’d been improvising badly on my own. The modules on guardrails and evaluation specifically addressed almost every mistake on this list, which was slightly humbling to sit through in hindsight, realising how avoidable most of it had actually been.</p><p><strong>What I’d tell someone starting from scratch</strong></p><p>If you’re about to build your first real agent instead of just a prompt-and-response chatbot, the honest advice is this: assume it knows nothing you haven’t explicitly told it, build in a way for it to flag its own uncertainty before you trust it with real decisions, and test it against the messiest, most realistic version of your actual use case, not the clean examples that make a demo look impressive.</p><p>The gap between “technically works” and “actually reliable” is where almost all the real engineering effort lives, and honestly, where most of the genuinely useful learning happens too.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a9cee27393f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[6 Things I Wish Someone Had Explained to Me About Agentic AI]]></title>
            <link>https://medium.com/@ab6714462/6-things-i-wish-someone-had-explained-to-me-about-agentic-ai-737d8b276592?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/737d8b276592</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[upgrad]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Fri, 21 Aug 2026 05:51:29 GMT</pubDate>
            <atom:updated>2026-08-21T05:51:29.837Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Agentic AI concepts explored through an Agentic AI programme via upGrad." src="https://cdn-images-1.medium.com/max/1024/1*zWuvvGJBOsk6uPDB8CTtOA.png" /><figcaption>Understanding Agentic AI through an Agentic AI programme via upGrad.</figcaption></figure><p>When I first heard the term “Agentic AI,” I assumed it was just a new name for a smarter chatbot.</p><p>It took a while to understand why people were so excited about it.</p><p>The difference is actually quite significant.</p><p>Here are six things that helped me understand it.</p><h3>1. Agents Aren’t Just Chatbots</h3><p>A chatbot generally responds to a prompt.</p><p>An AI agent can potentially take a goal, break it into steps, use tools, and execute tasks.</p><p>That’s a completely different level of interaction.</p><h3>2. Agents Can Work Through Tasks</h3><p>Imagine asking an AI system to research competitors.</p><p>Instead of simply giving you a response, an agent could potentially search for information, organise it, compare findings, and produce a report.</p><p>The AI becomes part of the workflow.</p><figure><img alt="AI workflow and applications explored through an Agentic AI programme via upGrad." src="https://cdn-images-1.medium.com/max/1024/1*4h6h6AbtMwxjgS7TCRaaqA.png" /><figcaption>Exploring practical Agentic AI concepts through upGrad.</figcaption></figure><h3>3. Autonomy Creates New Problems</h3><p>Giving AI more autonomy also creates more responsibility.</p><p>What happens if the agent makes the wrong decision?</p><p>How do you monitor its actions?</p><p>What information should it be allowed to access?</p><p>These aren’t just technical questions anymore.</p><h3>4. Human Oversight Still Matters</h3><p>Agentic AI isn’t about removing humans from every process.</p><p>In many situations, the most sensible approach is to let AI handle repetitive or well-defined tasks while humans remain responsible for important decisions.</p><h3>5. The System Around the Model Matters</h3><p>Building an agent isn’t simply about choosing a powerful language model.</p><p>You need to think about tools, data, memory, workflows, evaluation, security, and integration with existing systems.</p><h3>6. This Is Why Structured Learning Matters</h3><p>I’ve been exploring <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/"><strong>Agentic AI programmes via upGrad</strong></a> because I wanted to understand the technology beyond simply experimenting with AI assistants.</p><p>The interesting part is learning how agents can be designed and applied to real workflows.</p><p>Agentic AI still has plenty of unanswered questions.</p><p>But one thing seems increasingly clear: the future of AI may involve systems that don’t just answer our questions — they help us accomplish the tasks behind them.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=737d8b276592" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Abstract AI intelligence navigating multiple information sources, tools, decisions, and obstacles…]]></title>
            <link>https://medium.com/@ab6714462/abstract-ai-intelligence-navigating-multiple-information-sources-tools-decisions-and-obstacles-44d83733a5c3?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/44d83733a5c3</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[upgrad]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Wed, 19 Aug 2026 12:13:15 GMT</pubDate>
            <atom:updated>2026-08-19T12:13:15.687Z</atom:updated>
            <content:encoded><![CDATA[<h3>Abstract AI intelligence navigating multiple information sources, tools, decisions, and obstacles while working toward a defined goal.I Stopped Thinking of AI Agents as Just Smarter Chatbots</h3><p>When I first came across Agentic AI, I assumed it was simply another way of describing more advanced AI assistants.</p><p>The more I explored it, the more I realised there was a fundamental difference.</p><p>A chatbot responds.</p><p>An AI agent can potentially <strong>reason through a goal, decide what steps are needed, use tools, and execute tasks</strong> with limited human intervention.</p><h3>From Prompts to Workflows</h3><p>This changes how I think about AI.</p><p>Instead of asking an AI system to summarise a report, imagine giving it the broader task of analysing the report, identifying important findings, comparing them with previous data, and preparing a recommendation.</p><p>The difference isn’t just better prompting.</p><p>It’s about giving AI a workflow to manage.</p><h3>Why This Is Harder Than It Looks</h3><figure><img alt="Abstract intelligence moving beyond a simple conversation to interact with information, tools, decisions, and tasks across a larger digital environment." src="https://cdn-images-1.medium.com/max/1024/1*QtkAs14NOR4MZ3GNB70dBQ.png" /></figure><p>The exciting part of Agentic AI is also what makes it challenging.</p><p>What happens when an agent makes the wrong decision?</p><p>How do you evaluate its output?</p><p>What information should it be allowed to access?</p><p>When should a human take over?</p><p>These questions become critical when agents move from experiments into real business processes.</p><h3>Learning Beyond Generative AI</h3><p>This is why I’ve been exploring <a href="https://www.upgrad.com/applied-ai-and-agentic-ai-executive-pgp-certification-iiitb/"><strong>Agentic AI programmes via upGrad</strong></a>.</p><p>The interesting part for me isn’t simply learning how to interact with AI models.</p><p>It’s understanding how intelligent systems can be designed to work with data, tools, applications, and people.</p><h3>The Next Stage of AI</h3><p>I think the biggest shift will be moving from AI that helps us complete individual tasks to AI that can participate in entire workflows.</p><p>That’s why Agentic AI feels less like another AI trend and more like a change in how we think about software itself.</p><figure><img alt="Abstract AI intelligence navigating multiple information sources, tools, decisions, and obstacles while working toward a defined goal." src="https://cdn-images-1.medium.com/max/1024/1*abDYHrZiGT86V7bsfdCiQg.png" /></figure><p>#AgenticAI #ArtificialIntelligence #AI #GenerativeAI #upGrad</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=44d83733a5c3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Meeting Where My Manager Said “Just Automate the Whole Thing”]]></title>
            <link>https://medium.com/@ab6714462/the-meeting-where-my-manager-said-just-automate-the-whole-thing-82f4ed5b5a74?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/82f4ed5b5a74</guid>
            <category><![CDATA[online-learning]]></category>
            <category><![CDATA[online-learning-platform]]></category>
            <category><![CDATA[upgrad]]></category>
            <category><![CDATA[ai-course]]></category>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Mon, 17 Aug 2026 13:12:30 GMT</pubDate>
            <atom:updated>2026-08-17T13:12:30.522Z</atom:updated>
            <content:encoded><![CDATA[<p>A few months ago, my manager asked me to “just get AI to handle the ticket triage” in a planning meeting, said it like it was a five minute task. I nodded, went back to my desk, and realised I genuinely did not know what that meant in practice. I had used ChatGPT plenty, wrote emails faster, summarised documents, the usual. But “handle the ticket triage” meant something else entirely, reading an incoming ticket, checking priority rules, deciding who it should go to, maybe even drafting a first response, without me sitting there approving every single step.</p><p>That’s when it actually hit me, there’s a real difference between AI that responds to something you type, and AI that goes and does something on its own. I had been using the first kind for over a year and calling myself AI-savvy. Turns out I hadn’t touched the second kind at all.</p><p><strong>Why “AI that responds” and “AI that acts” are different animals</strong></p><p>Most people, myself included until recently, think of AI as something you ask a question and it answers. Generative AI does exactly that, you prompt, it responds, conversation over. Agentic AI works differently. Give it a goal instead of a question, and it breaks that goal into steps, decides which tools it needs, pulls in outside information, takes actions, and adjusts along the way, often without you supervising each individual move.</p><p>A simple way to think about it: a generative AI model can summarise a customer’s complaint. An agentic system can actually resolve it, reading the issue, checking the order history, applying policy, drafting a response, and updating the record, start to finish.</p><p>That’s a meaningfully different job description for a piece of software.</p><figure><img alt="Generative AI response compared with autonomous Agentic AI workflows in an upGrad Agentic AI programme." src="https://cdn-images-1.medium.com/max/1024/1*iXxM5cyIUgBB8UmUvid_5A.jpeg" /></figure><p><strong>Where I hit the wall trying to learn this properly</strong></p><p>Once I got curious about actually building something more serious than a bossy to-do app, I ran into the usual beginner problem, tutorials that teach you to write clever prompts but never actually explain how an agent plans, remembers context across steps, or decides when to call an external tool versus when to just respond in text.</p><p>Prompt engineering alone doesn’t get you there. You need to understand things like multi-step task decomposition, tool and API integration, memory systems that persist information between actions, and increasingly, multi-agent setups where several AI agents divide up a task and coordinate with each other, one drafting, one reviewing, one executing.</p><p>This is roughly where I stopped self-teaching and started looking at structured programmes, mainly because I wanted the RAG pipelines and multi-agent orchestration parts explained properly instead of pieced together from scattered blog posts and half-finished YouTube series.</p><p><strong>What actually stood out once I looked into upGrad’s </strong><a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/"><strong>Agentic AI programmes</strong></a></p><p>A few things surprised me. First, the project list isn’t theoretical, you’re building things like a conversational shopping assistant, an AI powered image generation pipeline for e-commerce catalogues, and a knowledge retrieval system that answers questions by scanning an organisation’s actual documents, using tools like LangChain, Hugging Face, and vector databases like ChromaDB and Pinecone. That’s a very different experience from watching someone else build an agent on a screen recording.</p><p>Second, the curriculum actually separates generative AI skills (prompting, fine-tuning, embeddings) from agentic AI skills (planning, tool use, multi-agent collaboration) instead of blurring them together, which honestly made the “what’s the actual difference” question I opened with finally click for me properly.</p><p>Third, and this mattered more than I expected, there’s real emphasis on deployment and safety, guardrails, bias reduction, human-in-the-loop checks for sensitive actions. An agent that can take real actions without any of that is, frankly, a liability waiting to happen, and it was reassuring to see that treated as core curriculum rather than an afterthought slide.</p><p><strong>Where it actually landed for me</strong></p><p>I haven’t built anything as dramatic as an autonomous trading system, some of the more advanced capstone examples floating around look genuinely intimidating. What I have built, as a course project, is a basic version of that exact ticket triage system my manager casually asked for months ago, one that reads incoming requests, checks them against priority rules, and routes them to the right person, flagging anything unusual for a human to look at instead of guessing.</p><figure><img alt="Agentic AI ticket triage workflow with human oversight from an upGrad Agentic AI programme." src="https://cdn-images-1.medium.com/max/1024/1*RmuAjTWLk00M9ftv6eSavg.jpeg" /></figure><p>It’s not fully in production at work yet, still going through the usual approvals, but I finally walked into the next planning meeting and explained, properly, what “automate the ticket triage” would actually involve, the tools, the tradeoffs, where a human still needs to stay in the loop. That alone felt like a different kind of promotion, not on paper, but in how seriously the conversation was taken.</p><p>Funny how one offhand instruction in a meeting can end up being the thing that sends you back to actually learning something properly.</p><p>If your manager asked you to “just automate it” tomorrow, would you actually know where to start, or would you be nodding the way I was?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=82f4ed5b5a74" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How Generative AI Became the Foundation Layer for Agentic Systems]]></title>
            <link>https://medium.com/@ab6714462/how-generative-ai-became-the-foundation-layer-for-agentic-systems-f548b5974bbc?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/f548b5974bbc</guid>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 09:23:40 GMT</pubDate>
            <atom:updated>2026-07-23T09:23:40.321Z</atom:updated>
            <content:encoded><![CDATA[<p>Artificial intelligence has evolved remarkably over the past few years. What began as systems capable of recognising images, analysing data, or generating text has now progressed to AI systems that can reason, plan, make decisions, and complete multi-step tasks with minimal human intervention.</p><p>These systems, commonly referred to as <strong>AI agents</strong> or <strong>agentic AI</strong>, are quickly becoming one of the most significant developments in enterprise technology.</p><p>While much of the attention today is focused on autonomous AI agents, it’s important to understand that they did not emerge in isolation. Their capabilities are built on the rapid progress made in <strong>Generative AI</strong>.</p><p>Without large language models, natural language understanding, and generative reasoning, modern agentic systems simply would not exist.</p><p>Understanding this relationship is becoming increasingly important for professionals looking to build careers in artificial intelligence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mY5K01VpEyUm2v9UcEKUDw.png" /></figure><h3>The Rise of Generative AI</h3><p>Generative AI marked a major shift in how machines interact with information. Unlike traditional AI systems that primarily classified or predicted outcomes, Generative AI introduced models capable of creating new content, answering questions, summarising documents, writing code, generating images, and engaging in natural conversations.</p><p>Large Language Models (LLMs) became the engine behind these capabilities. Instead of following fixed rules, they learned patterns from massive datasets, allowing them to understand context, generate coherent responses, and assist with increasingly complex tasks.</p><p>This represented a significant leap from automation toward intelligent assistance.</p><h3>Why Generative AI Was Only the Beginning</h3><p>Although Generative AI dramatically improved human-computer interaction, early models were largely reactive. They generated responses based on prompts but did not independently plan tasks, remember long-term objectives, or coordinate multiple actions.</p><p>For example, an LLM could write an email or summarise a report. Still, it could not independently retrieve additional information, verify facts, interact with external applications, and complete an entire workflow without continuous human guidance.</p><p>Businesses soon recognised that while Generative AI was powerful, greater value could be created by enabling AI to perform sequences of connected actions rather than isolated responses.</p><h3>Enter Agentic AI</h3><p>Agentic AI extends the capabilities of Generative AI by enabling systems to plan, reason, make decisions, use external tools, and execute multi-step workflows.</p><p>Instead of simply responding to prompts, AI agents can break complex objectives into smaller tasks, gather information from different sources, adapt their approach based on new inputs, and continue working until a goal is achieved.</p><p>This shift moves artificial intelligence from being an intelligent assistant to becoming an active problem solver.</p><h3>Why Generative AI Remains the Foundation</h3><p>Although agentic systems appear significantly more advanced, Generative AI continues to provide the core intelligence behind them.</p><p>Large language models remain responsible for interpreting user instructions, understanding context, generating reasoning, evaluating possible actions, and communicating with users.</p><p>Additional capabilities such as memory, planning, orchestration, tool usage, and workflow execution are layered on top of these foundational language models.</p><p>In other words, Generative AI provides the reasoning engine, while agentic architectures provide the decision-making and execution framework.</p><p>Without the language understanding and reasoning capabilities developed through Generative AI, modern AI agents would lack the flexibility that makes them valuable.</p><h3>How Businesses Are Using Agentic Systems</h3><p>Organisations are increasingly exploring agentic systems across multiple business functions.</p><p>Customer support teams are developing AI agents capable of resolving customer queries across several systems without human intervention.</p><p>Software engineering teams are using AI agents to assist with coding, debugging, testing, and documentation.</p><p>Marketing teams are experimenting with agents that analyse campaign performance, generate content, monitor trends, and recommend optimisation strategies.</p><p>Operations teams are deploying AI agents that automate repetitive workflows while coordinating information across multiple business applications.</p><p>These examples demonstrate why understanding both Generative AI and agentic systems is becoming increasingly valuable for technology professionals.</p><h3>Building Skills for the Next Generation of AI</h3><p>As businesses continue adopting intelligent automation, professionals need more than a basic understanding of AI tools. They need to understand how Generative AI models work, how agentic architectures are designed, and how intelligent systems can be deployed responsibly in real-world environments.</p><p>Programs like the <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/"><strong>Executive Post Graduate Certificate in Generative AI &amp; Agentic AI</strong></a><strong> via </strong><a href="https://www.upgrad.com/"><strong>upGrad</strong></a> are designed to help professionals build this understanding through structured learning.</p><p>The curriculum covers Generative AI, large language models, agentic systems, AI applications, and practical implementation, helping learners bridge the gap between AI theory and real-world deployment.</p><p>For professionals looking to build careers in AI, software engineering, product management, or technology leadership, developing expertise in both Generative AI and agentic systems is becoming increasingly important.</p><h3>The Future Belongs to Intelligent Systems</h3><p>Generative AI transformed how machines create and communicate. Agentic AI is transforming how machines act.</p><p>Rather than replacing Generative AI, agentic systems build upon it, adding planning, memory, reasoning, and execution capabilities that enable AI to complete increasingly sophisticated tasks.</p><p>As organisations continue investing in intelligent automation, professionals who understand this evolution will be better positioned to contribute to the next generation of AI-powered products and services.</p><p>The future of AI is unlikely to be defined by isolated models responding to prompts. It will be shaped by intelligent systems capable of understanding goals, making decisions, collaborating with humans, and executing complex workflows — built on the strong foundation that Generative AI has already established.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f548b5974bbc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Human-in-the-Loop vs Fully Autonomous Agents: When to Trust the Machine]]></title>
            <link>https://medium.com/@ab6714462/human-in-the-loop-vs-fully-autonomous-agents-when-to-trust-the-machine-5f2943561349?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/5f2943561349</guid>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Mon, 20 Jul 2026 10:42:04 GMT</pubDate>
            <atom:updated>2026-07-20T10:42:04.153Z</atom:updated>
            <content:encoded><![CDATA[<p>There is a moment that almost every company experimenting with AI eventually reaches.</p><p>The chatbot works. The workflow automation is running. Reports are being generated in seconds instead of hours. Someone in the meeting inevitably asks the obvious next question:</p><p><em>“If AI can already do this much, why are we still keeping humans in the loop?”</em></p><p>It sounds like a reasonable question. After all, the promise of AI has always been automation. Faster decisions. Lower costs. Less manual work.</p><p>But as organizations move beyond simple automation into <strong>AI agents</strong> capable of reasoning, planning, and taking actions independently, the conversation changes. The real challenge is no longer <em>whether</em> AI can perform a task. It’s deciding <strong>when the machine should make the final decision — and when a human still should.</strong></p><p>That’s becoming one of the defining leadership questions of the AI era.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zTi8si-eN9HV7C9ILiifBQ.png" /></figure><h3>Autonomous Doesn’t Mean Unsupervised</h3><p>One of the biggest misconceptions surrounding AI agents is that autonomy automatically means independence.</p><p>In reality, autonomy exists on a spectrum.</p><p>Some AI agents simply recommend actions.</p><p>Others execute predefined workflows with minimal oversight.</p><p>The most advanced systems can plan multi-step tasks, interact with software tools, retrieve information, generate outputs, and continuously adapt based on feedback.</p><p>These are often referred to as <strong>fully autonomous agents</strong>.</p><p>They’re incredibly capable.</p><p>They’re also not appropriate for every business problem.</p><h3>Why Human-in-the-Loop Still Matters</h3><p>Consider a bank evaluating loan applications.</p><p>An AI model can analyze thousands of variables far faster than any credit officer.</p><p>But should it reject a customer’s loan application entirely on its own?</p><p>Probably not.</p><p>Healthcare presents another example.</p><p>AI can identify abnormalities in medical scans with remarkable accuracy.</p><p>Yet most hospitals still require physicians to review recommendations before diagnoses are finalized.</p><p>The same pattern appears across hiring, legal services, insurance, cybersecurity, and government.</p><p>In each case, AI provides speed.</p><p>Humans provide accountability.</p><p>That’s the essence of <strong>Human-in-the-Loop (HITL)</strong> systems.</p><p>Instead of replacing people, AI becomes an intelligent collaborator that handles repetitive analysis while humans oversee judgment-heavy decisions.</p><h3>The Cost of Blind Trust</h3><p>One reason organizations remain cautious is that modern AI systems are exceptionally convincing — even when they’re wrong.</p><p>Large language models can generate confident explanations that contain factual errors.</p><p>Agentic systems may complete every step of a workflow while misunderstanding the original objective.</p><p>Automation can also amplify mistakes.</p><p>If a human makes one incorrect decision, the damage is usually limited.</p><p>If an autonomous agent repeats that same mistake thousands of times before anyone notices, the consequences become significantly larger.</p><p>That’s why mature AI strategies focus just as much on governance as they do on capability.</p><p>The question isn’t whether AI is accurate.</p><p>It’s whether the organization has built enough safeguards around it.</p><h3>Where Fully Autonomous Agents Make Sense</h3><p>That doesn’t mean autonomous agents should be avoided.</p><p>Far from it.</p><p>Many business processes are ideal candidates for complete automation.</p><p>Think about tasks like:</p><ul><li>Monitoring cloud infrastructure</li><li>Managing routine IT operations</li><li>Processing invoices</li><li>Scheduling workflows</li><li>Customer support triage</li><li>Software testing</li><li>Data synchronization across enterprise systems</li></ul><p>These activities involve structured rules, measurable outcomes, and relatively low risk if monitored properly.</p><p>In these environments, autonomous agents can dramatically improve productivity while allowing employees to focus on more strategic work.</p><h3>Leadership Is Becoming the Missing Skill</h3><p>Interestingly, the biggest challenge companies face isn’t building AI anymore.</p><p>It’s leading it.</p><p>Technical teams understand how models work.</p><p>Business leaders understand organizational priorities.</p><p>The difficult part is connecting those two worlds.</p><p>Questions like:</p><ul><li>Which decisions should remain human-led?</li><li>How do we evaluate AI risk?</li><li>What governance framework should we adopt?</li><li>How do we measure success beyond accuracy?</li></ul><p>These aren’t engineering questions.</p><p>They’re leadership questions.</p><p>That’s one reason executive education around AI has evolved so rapidly.</p><p>Programs offered through <a href="https://www.upgrad.com/"><strong>upGrad</strong></a>, in collaboration with institutions such as <a href="https://www.upgrad.com/iimk-ai-professional-certificate-programme/"><strong>IIM Kozhikode</strong></a>, increasingly focus not only on AI technologies but also on governance, responsible AI, decision-making, digital transformation, and strategic leadership. Rather than training every executive to become an AI engineer, these programs help professionals evaluate AI systems, build organizational trust, and lead AI adoption across business functions.</p><p>As AI moves into boardrooms, those capabilities are becoming just as valuable as technical knowledge itself.</p><h3>Trust Isn’t Binary</h3><p>Perhaps the biggest mistake organizations can make is thinking the choice is between complete automation and complete human control.</p><p>Most successful AI implementations operate somewhere in the middle.</p><p>Humans define objectives.</p><p>AI performs analysis.</p><p>Humans review exceptions.</p><p>AI executes routine actions.</p><p>Humans monitor outcomes and continuously improve the system.</p><p>In other words, trust isn’t something you either give or withhold.</p><p>It’s something you design.</p><p>As AI agents become more capable, the organizations that benefit the most won’t necessarily be the ones that automate everything. They’ll be the ones that know exactly <strong>where automation ends and human judgment begins.</strong> In the age of agentic AI, that distinction may become one of the most valuable leadership skills any professional can develop.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5f2943561349" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Rise of the “AI Orchestrator” Job Role: What It Is and How to Prepare]]></title>
            <link>https://medium.com/@ab6714462/the-rise-of-the-ai-orchestrator-job-role-what-it-is-and-how-to-prepare-58280e925c50?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/58280e925c50</guid>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Tue, 14 Jul 2026 11:08:48 GMT</pubDate>
            <atom:updated>2026-07-14T11:08:48.721Z</atom:updated>
            <content:encoded><![CDATA[<p>Artificial Intelligence is no longer limited to answering questions or generating text. Today’s AI systems are becoming increasingly capable of planning, reasoning, using external tools, and completing complex workflows with minimal human intervention. This shift is giving rise to a new kind of professional: the <strong>AI Orchestrator</strong>.</p><p>Rather than building every AI model from scratch, AI Orchestrators focus on designing, managing, and optimizing networks of AI agents that work together to solve real business problems. As organizations adopt Agentic AI, this role is quickly becoming one of the most valuable in the AI ecosystem.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0-PHMyrWLYTBmowaXOM9qA.png" /></figure><h3>What is an AI Orchestrator?</h3><p>An AI Orchestrator is someone who coordinates multiple AI systems, tools, and workflows to achieve a specific business objective.</p><p>Instead of writing prompts for a single chatbot, an AI Orchestrator creates processes where different AI agents collaborate. One agent may gather data, another may analyze it, another may generate reports, and another may integrate the results into business software.</p><p>The orchestrator ensures these systems work efficiently, securely, and accurately.</p><p>Think of it as conducting an orchestra. Each musician has a specific role, but without a conductor, the performance lacks coordination. Similarly, AI agents become significantly more powerful when someone designs their interactions and how they complement one another.</p><h3>Why This Role is Emerging</h3><p>Businesses are moving beyond experimenting with AI. They now want systems that can automate end-to-end workflows rather than isolated tasks.</p><p>Examples include:</p><ul><li>Customer support agents that resolve queries without human intervention.</li><li>Marketing agents that research trends, create content, and monitor campaign performance.</li><li>HR agents that screen resumes, schedule interviews, and generate reports.</li><li>Finance agents that analyze spending, detect anomalies, and prepare forecasts.</li></ul><p>Managing these intelligent systems requires professionals who understand both business processes and AI capabilities.</p><p>This is where AI Orchestrators add value.</p><h3>Skills Every AI Orchestrator Needs</h3><p>The role is highly interdisciplinary. Technical knowledge is important, but so is strategic thinking.</p><p>Some of the most valuable skills include:</p><ul><li>Understanding Large Language Models (LLMs)</li><li>Designing Agentic AI workflows</li><li>Prompt engineering</li><li>Workflow automation</li><li>API integrations</li><li>Python fundamentals</li><li>Data analysis</li><li>AI governance and ethics</li><li>Business problem-solving</li><li>Communication and stakeholder management</li></ul><p>Rather than focusing solely on coding, AI Orchestrators learn to integrate multiple technologies into intelligent systems that deliver measurable outcomes.</p><h3>Industries That Need AI Orchestrators</h3><p>The demand for AI Orchestrators extends across industries.</p><p>Healthcare organizations are automating patient support and documentation.</p><p>Banks are improving fraud detection and customer service.</p><p>Retail companies are personalizing shopping experiences using autonomous AI agents.</p><p>Manufacturing firms are optimizing supply chains through AI-driven decision-making.</p><p>Marketing teams are increasingly using multiple AI tools for content creation, customer segmentation, campaign optimization, and performance analysis.</p><p>Wherever businesses adopt AI at scale, orchestration becomes essential.</p><h3>How to Prepare for This Career</h3><p>The best way to prepare is by combining AI fundamentals with practical implementation.</p><p>Learning about machine learning and generative AI provides the foundation, but it’s equally important to understand how autonomous agents collaborate, access external tools, and execute multi-step tasks.</p><p>Hands-on projects, workflow design, and exposure to real-world business use cases help bridge the gap between theory and practice.</p><p>Many professionals are exploring specialized programs, such as <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/"><strong>Agentic AI courses offered via upGrad</strong></a> and developed in partnership with leading academic institutions, to build these skills.</p><p>These programs typically cover topics such as Generative AI, autonomous AI agents, LLM applications, workflow orchestration, and real-world AI implementation through projects and case studies.</p><p>For learners looking to move beyond simply using AI tools and toward designing intelligent AI systems, structured learning can provide a strong foundation.</p><h3>The Future Belongs to AI Coordinators, Not Just AI Users</h3><p>Knowing how to use ChatGPT or another AI assistant is becoming a basic workplace skill.</p><p>The real opportunity lies in designing systems where multiple AI agents collaborate to solve complex problems with minimal human intervention.</p><p>Organizations are increasingly seeking professionals who can connect AI technologies with business strategy, automate workflows, and ensure reliable outcomes.</p><p>As Agentic AI continues to evolve, the AI Orchestrator is set to become one of the defining roles of the next decade.</p><p>Those who begin learning these skills today won’t simply adapt to the future of work — they’ll play a key role in shaping how intelligent systems transform businesses across every industry.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58280e925c50" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The AI Engineer Role: What It Involves and How It Differs from Data Scientist]]></title>
            <link>https://medium.com/@ab6714462/the-ai-engineer-role-what-it-involves-and-how-it-differs-from-data-scientist-a42b285544fd?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/a42b285544fd</guid>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Thu, 09 Jul 2026 12:19:00 GMT</pubDate>
            <atom:updated>2026-07-09T12:19:00.569Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>(Or: Why Two People at the Same Company Can Both Say “I Work With AI” and Mean Completely Different Things)</blockquote><p>Picture a dinner party. Someone asks, “So what do you do?” One guest says, “I’m a Data Scientist.” Another says, “I’m an AI Engineer.” The rest of the table nods politely, mentally files both answers under “computer wizard,” and moves on to talking about the appetizers.</p><p>But here’s the thing: these two people probably didn’t do a single overlapping task all week. One of them was elbow-deep in spreadsheets arguing with a p-value. The other was arguing with a production server at 2 a.m. because an API endpoint decided to have an existential crisis.</p><p>If you’re trying to figure out which of these careers is actually <em>for you</em>, or you’re a hiring manager who keeps writing job descriptions that ask for both roles in one posting (we see you, and we’re judging you gently), let’s actually untangle this.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Bpsqh5cFHUvxEJNlTalCmQ.png" /></figure><h3>The Short Version, For Those Who Just Want the Punchline</h3><p><strong>Data Scientists</strong> ask: <em>“What does this data tell us, and what should we predict?”</em></p><p><strong>AI Engineers</strong> ask: <em>“How do we take a working model and turn it into something a million real users can actually use without it falling over?”</em></p><p>One lives mostly in notebooks. The other lives mostly in production systems. Both will, at some point, silently question their life choices while staring at an error log. That part’s universal.</p><h3>What a Data Scientist Actually Does All Day</h3><p>Despite what LinkedIn thumbnails suggest, a Data Scientist’s day rarely looks like dramatically typing code while blue numbers scroll behind them like <em>The Matrix</em>. It looks more like:</p><ul><li>Cleaning data that someone entered in six different date formats, apparently out of spite</li><li>Running statistical analysis to find patterns humans would never notice</li><li>Building and testing machine learning models to predict outcomes: churn, fraud, sales, you name it</li><li>Translating “the model says X” into “here’s what X means for the business” for people who do not want to hear the word “regression” ever again</li><li>A truly staggering amount of exploratory data analysis, which is a fancy way of saying “staring at charts until they make sense”</li></ul><p>The Data Scientist’s core currency is <strong>insight</strong>. Their job succeeds when a decision-maker says, “Oh, that’s genuinely useful, let’s act on it.”</p><h3>What an AI Engineer Does All Day</h3><p>An AI Engineer, meanwhile, tends to enter the story <em>after</em> the model already sort of works. Their job is making sure it works reliably, at scale, in the real world, where users do unpredictable things like uploading a 4,000-page PDF or typing in Klingon just to see what happens.</p><p>A typical day might involve:</p><ul><li>Taking a trained model or a foundation model API and building the actual application around it</li><li>Designing pipelines that feed data in and get predictions out reliably, quickly, and without leaking anyone’s personal information</li><li>Building and orchestrating <strong>agentic workflows</strong>, where AI systems don’t just predict, but take multi-step actions</li><li>Managing latency, cost, and infrastructure (because a brilliant model that takes 45 seconds to respond and costs $12 per query is not a product, it’s a cautionary tale)</li><li>Testing for edge cases, hallucinations, and the special chaos that comes from letting an LLM interact with real users</li></ul><p>The AI Engineer’s core currency is <strong>reliability at scale</strong>. Their job succeeds when the system works so smoothly that nobody even thinks about the AI part; they just think, “huh, this app is really good.”</p><h3>A Slightly Petty But Accurate Analogy</h3><p>Think of building an AI product like opening a restaurant.</p><p>The <strong>Data Scientist</strong> is the person who develops the recipe. They test ingredients, taste constantly, tweak ratios, and eventually land on a dish that’s genuinely delicious in a controlled kitchen environment.</p><p>The <strong>AI Engineer</strong> is the person who figures out how to cook that exact recipe for 500 customers a night, with consistent quality, without running out of ingredients, without the kitchen catching fire, and ideally without anyone in the dining room ever knowing how chaotic it gets behind those doors.</p><p>Both roles are essential. Neither one can single-handedly run the restaurant. And if you’ve ever tried to do both jobs alone, you already understand why restaurants employ separate chefs and kitchen managers.</p><h3>Where the Two Roles Overlap (Because They Definitely Do)</h3><p>This isn’t a clean, drama-free divorce. Both roles need to understand:</p><ul><li>How machine learning models actually work under the hood</li><li>How to write solid, structured code (Python is the shared home turf)</li><li>How to evaluate whether a model’s outputs are actually good, not just technically functional</li><li>Enough about the business problem to know when a “successful model” is actually solving the wrong problem entirely.</li></ul><p>The split isn’t about intelligence or difficulty; plenty of people move between these roles over the course of a career. It’s about <em>what part of the AI lifecycle you enjoy living in</em>. If you get a dopamine hit from finding a hidden pattern in messy data, Data Science will feel like home.</p><p>If you get that same hit from watching a system handle 10,000 concurrent users without breaking a sweat, AI Engineering is calling your name.</p><h3>Why This Distinction Matters More Than Ever Right Now</h3><p>A few years ago, “AI Engineer” barely existed as a job title. Most companies just had Data Scientists and hoped someone on the team could also handle deployment. That approach has, politely put, stopped scaling.</p><p>With the explosion of generative AI and agentic systems AI that doesn’t just predict a number but actually <em>does things</em> on your behalf companies need people who can specifically build and maintain those systems in production.</p><p>That’s a genuinely different skill set from traditional data science, and the market has started paying accordingly. AI Engineering has quickly become one of the most in-demand, well-compensated tracks in tech, precisely because so few people have formal training in it yet.</p><p>Which, if you’re early to this space, is less “uh oh, competition” and more “oh good, arriving early for once in my life.”</p><h3>So Which One Should You Learn?</h3><p>If any of these sound like you, Data Science is your lane:</p><ul><li>You love asking “why” until you hit bedrock</li><li>Statistics doesn’t scare you (or at least, it scares you in a fun way)</li><li>You want your work to influence business strategy directly</li></ul><p>If these sound more like you, AI Engineering is your lane:</p><ul><li>You’d rather build the thing than analyze the thing</li><li>You like systems, architecture, and the satisfaction of something running smoothly at scale</li><li>You’re drawn to the idea of building AI agents that actually <em>do</em> work, not just describe it</li></ul><p>And if you’re sitting there thinking “honestly… both,” that’s a completely reasonable answer. The two disciplines increasingly blend at the edges, especially in generative and agentic AI, where understanding the model <em>and</em> knowing how to deploy it reliably has become the actual differentiator in the job market.</p><h3>Where to Learn This Properly</h3><p>You <em>could</em> keep doing what most of us do: opening seventeen browser tabs, watching half a YouTube video, getting distracted by a completely unrelated Reddit thread about sourdough starters, and calling that “self-study.” It’s a valid life choice. It is not, however, an efficient one.</p><p>If you’d rather skip the sourdough tangent and actually build the production-facing skills that define modern AI Engineering working with foundation models, designing agentic workflows, and deploying real AI systems- the AI programs on the <a href="https://www.upgrad.com/">upGrad</a> platform are worth a look.</p><p>No CS degree required, just a willingness to close the sourdough tab. And if you’d rather start at a lower level with core machine learning and AI fundamentals, both roles are built on the <a href="https://www.upgrad.com/artificial-intelligence-course/">Artificial Intelligence course</a> via upGrad. It’s a solid place to lay that groundwork first.</p><h3>The Real Takeaway</h3><p>Data Scientist and AI Engineer aren’t two names for the same job; they’re two different relationships with the same technology. One finds the insight. The other builds the thing that puts that insight to work at scale, reliably, for real people, at 2 a.m., while everyone else is asleep.</p><p>You don’t have to pick a side forever. Careers zigzag, titles blur, and the smartest people in this space usually end up fluent in both. But if you’re standing at the fork right now, wondering which door to walk through first, just ask yourself whether you’d rather taste the dish or run the kitchen. Either answer is correct. Only one of them involves being on call.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a42b285544fd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building Your First Agentic Workflow Without a Computer Science Degree]]></title>
            <link>https://medium.com/@ab6714462/building-your-first-agentic-workflow-without-a-computer-science-degree-0899566d2598?source=rss-891b48146094------2</link>
            <guid isPermaLink="false">https://medium.com/p/0899566d2598</guid>
            <dc:creator><![CDATA[Antonio D'Souza]]></dc:creator>
            <pubDate>Thu, 09 Jul 2026 12:06:52 GMT</pubDate>
            <atom:updated>2026-07-09T12:06:52.898Z</atom:updated>
            <content:encoded><![CDATA[<blockquote>(Spoiler: You Don’t Need to Know What a Recursive Function Is to Make an AI Do Your Job for You)</blockquote><p>Let’s get something out of the way immediately: I do not have a computer science degree. My most advanced technical achievement before this year was changing my Wi-Fi router password from “admin123.”</p><p>And yet, last month, I built an AI agent that reads my emails, drafts replies, checks my calendar for conflicts, and only interrupts me when something is <em>actually</em> on fire.</p><p>If you had told 2019 — me that 2026 — me would casually say the phrase “I built an agent,” I would have assumed I’d joined a spy agency or, worse, an MLM.</p><p>So here’s the truth nobody tells you loudly enough: <strong>agentic AI is not a coding skill. It’s a thinking skill.</strong> The coding part has been increasingly outsourced to the AI itself. Your job is to be the manager, not the intern typing semicolons.</p><p>Let’s break down how to build your first agentic workflow, why it’s less intimidating than it sounds, and where you can actually go to learn this properly instead of piecing it together from forty different YouTube tutorials at 1 a.m. (we’ve all been there).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*39uKK3moIJBx0zI_qizJiw.png" /></figure><h3>First, What Even Is an Agentic Workflow?</h3><p>Think of a regular chatbot as a very smart parrot. You ask it something, it says something clever back, and then it forgets you exist the moment the conversation ends. Charming, but ultimately a one-trick parrot.</p><p>An <strong>agentic workflow</strong>, on the other hand, is less parrot and more personal assistant who actually shows up to work. It doesn’t just answer; it <em>acts</em>. It can:</p><ul><li>Break a big task into smaller steps</li><li>Decide which tool or app to use for each step</li><li>Execute those steps in sequence</li><li>Check its own work</li><li>Ask you for help only when it’s genuinely stuck</li></ul><p>In other words, a chatbot talks. An agent <em>does</em>. And the gap between those two things is exactly where all the exciting (and slightly terrifying) job-market shifts are happening right now.</p><h3>The Myth That’s Been Scaring You Off</h3><p>Somewhere along the way, “AI agent” got mentally filed next to “quantum computing” and “blockchain consensus algorithms” technical wizardry reserved for people who dream in Python. This is, frankly, propaganda spread by people who want you to feel intimidated so they can feel special.</p><p>Here’s the actual skill breakdown for building a first agentic workflow:</p><ul><li>10% technical setup (mostly clicking buttons and connecting accounts)</li><li>30% understanding how tools chain together</li><li>60% clear thinking about <em>what problem you’re actually solving</em></li></ul><p>That last part, clear thinking, is not a CS skill. It’s a skill marketers, teachers, project managers, nurses, and your uncle who runs a small hardware store all already have. If you’ve ever written a to-do list, delegated a task, or explained a recipe to someone over the phone, you have already practiced the core skill of agent-building: <strong>breaking a goal into steps a less-informed party can follow.</strong></p><p>Congratulations, you’re basically already qualified.</p><h3>Step 1: Pick a Problem Annoying Enough to Solve</h3><p>Don’t start your agentic journey by trying to build “an AI that revolutionizes healthcare.” Start by building an AI that reminds you to reply to your landlord.</p><p>Good first agent projects are boring, personal, and repetitive, which is exactly the kind of task agents are built for:</p><ul><li>Summarizing long email threads into three bullet points</li><li>Pulling data from a spreadsheet and generating a weekly report</li><li>Monitoring a webpage for price drops</li><li>Drafting social media captions from a content calendar</li><li>Sorting incoming customer queries by urgency</li></ul><p>The magic ingredient isn’t complexity. It’s <em>annoyance</em>. Pick the task you most recently did, and you’ll stay motivated long enough actually to finish building the thing.</p><h3>Step 2: Give Your Agent a Job Description, Not Just a Vibe</h3><p>This is where most beginners trip up. They type something like “help me with my inbox” and then act shocked when the AI produces something vague and useless. That’s not the AI failing you — that’s you handing a new employee zero instructions and expecting them to read your mind.</p><p>Instead, write your agent a proper brief:</p><p>“You are an email triage assistant. Read incoming emails. If it’s from a client, flag it as urgent and draft a reply within two hours’ tone. If it’s a newsletter, archive it. If it’s ambiguous, ask me before acting.”</p><p>Notice this reads less like code and more like onboarding instructions for a new hire. That’s the whole trick. Agent-building is 80% about <em>writing really good instructions</em>, which, again, is not a computer science skill. It’s a communication skill. English majors, rejoice.</p><h3>Step 3: Connect the Tools (This Is the “Scary” Part That Isn’t)</h3><p>This is usually where beginners panic and Google “how to code an API integration,” find a 40-minute tutorial full of jargon, and quietly close the laptop forever.</p><p>You don’t need to do that anymore.</p><p>Modern agent-building platforms let you connect tools email, calendars, spreadsheets, Slack, CRMs through simple, visual, click-to-connect interfaces. You’re not writing integration code from scratch; you’re selecting from a menu, just like you’d add a filter on Instagram. The “wiring” happens behind the scenes.</p><p>Your job is to understand <em>what</em> should connect to <em>what</em>, in <em>what order</em>, and <em>why</em>. That’s workflow logic, not software engineering.</p><h3>Step 4: Let It Fail (On Purpose, in a Safe Sandbox)</h3><p>Here’s something nobody warns you about: your first agent will mess up. It might summarize the wrong email. It might send a draft reply that’s a little too enthusiastic about a customer complaint. This is normal, expected, and honestly kind of funny in hindsight.</p><p>Treat your early builds like training a very eager golden retriever. It wants to help. It just doesn’t fully understand boundaries yet. Test in a low-stakes environment, review outputs before anything goes live, and refine your instructions based on what goes wrong. Every failure is just a more specific instruction waiting to be written.</p><h3>Step 5: Scale Up Once You Trust It</h3><p>Once your simple agent is reliably doing its one boring job well, you can start chaining multiple agents together — one that researches, one that drafts, one that reviews, one that schedules. This is where agentic AI starts looking less like “a chatbot with extra steps” and more like a genuinely useful digital team working quietly in the background of your life.</p><h3>“Okay, But I Want to Actually Learn This Properly”</h3><p>Fair. Trial and error is fun until you’ve spent three weekends reinventing concepts someone could’ve explained to you in twenty minutes. If you want a structured path, one that goes beyond “watch a guy on YouTube build a Twitter bot” and actually teaches the underlying principles of prompting, tool orchestration, evaluation, and real-world agent design, it’s worth looking at a guided program rather than assembling knowledge from scattered blog posts (like, arguably, this one).</p><p>The <a href="https://www.upgrad.com/gen-ai-and-agentic-ai-programs/">Agentic AI</a> programs offered via <a href="https://www.upgrad.com/">upGrad</a> are built specifically for people who want to move from “I use ChatGPT sometimes” to “I design and deploy working AI agents,” without requiring a computer science background to get started. It’s a solid reference point if you want structured learning rather than duct-taping tutorials together at midnight.</p><h3>The Real Takeaway</h3><p>You don’t need a computer science degree to build an agentic workflow. You need curiosity, a genuinely annoying problem, and the willingness to write instructions as if you’re training a very capable, very literal new employee who takes everything you say at face value.</p><p>The tools have gotten dramatically easier. The gatekeeping hasn’t caught up yet. That gap is exactly where you should be standing right now — instructions in hand, ready to put your first agent to work.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0899566d2598" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>