<?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 Nasir Mahmood, PhD on Medium]]></title>
        <description><![CDATA[Stories by Nasir Mahmood, PhD on Medium]]></description>
        <link>https://medium.com/@nasirmaan?source=rss-1aaa392ab512------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*9a_Noty87V4oZhxsR3p4KQ.jpeg</url>
            <title>Stories by Nasir Mahmood, PhD on Medium</title>
            <link>https://medium.com/@nasirmaan?source=rss-1aaa392ab512------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 11:28:07 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nasirmaan/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[Too Many Doors to the Same Room]]></title>
            <link>https://medium.com/@nasirmaan/too-many-doors-to-the-same-room-ebd01a0d7c46?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/ebd01a0d7c46</guid>
            <category><![CDATA[product-strategy]]></category>
            <category><![CDATA[enterprise-technology]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[databricks]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Sat, 20 Jun 2026 03:41:39 GMT</pubDate>
            <atom:updated>2026-06-20T03:41:39.166Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LH5AVzcZgsVagr8S_XjdEA.png" /><figcaption>Too Many Doors to the Same Room</figcaption></figure><h3>I spent three days at the Databricks summit watching one company make a mistake and its cure at the same time, and the cure is the bigger story.</h3><p>A while back I wrote that <a href="https://www.nasirmaan.com/blog/Leading-AI-Transformation/companies-dont-die-from-doing-too-much/">big companies don’t die from doing too many things</a>. They die from doing the same thing twice: from building two, three, four products that overlap so much that no one person inside the company can explain, on the back of a napkin, when you’d use one and not the others. Think of it as too many doors to the same room. The customer stands in the hallway, sees four doors, and can’t tell which one is theirs. I used Microsoft as the example. Ask someone there to draw the line between Copilot Studio, Foundry, Agent Builder, and the agent SDK. They start strong, hit the seams within a minute, and land on “it depends on the customer” soon after.</p><p>I thought that was a story about one big, old company that had reorganized its product line one too many times. Last week I spent three days at the Databricks Data + AI Summit in San Francisco, and I watched a much younger, much sharper company do two things at once: build the same confusing set of doors at the product level, and, one floor up, build the exact thing that fixes it. Not just for itself. For the whole industry. That second move is the one worth your attention, and almost nobody is writing about it.</p><h3>What the floor looked like</h3><p>First the scale, because it matters. This was not a sales event in a hotel ballroom. Databricks put more than thirty thousand people on the ground at Moscone, with tens of thousands more watching online from over a hundred and fifty countries. The keynote stage had the founders, a recorded conversation with Satya Nadella, and people from OpenAI and Anthropic. The theme of the whole event was, more or less, “build apps and agents that actually work.” So when I say everyone was doing agents, I don’t mean it was background noise. It was the headline.</p><p>Let me be honest about what this piece is and isn’t. I attended, I walked the floor, I sat in keynotes, and I’ve built on Databricks before. But I saw a handful of booths out of hundreds, and only some of the sessions. Some of what follows is a read on strategy, not a fact from inside the company. I’ll keep the strong claims tied to things anyone can check (product pages, the open-source license, the announcements) and keep my guesses clearly labeled as guesses.</p><h3>Up close: too many doors</h3><p>At the product level, Databricks now has several overlapping ways to do the one thing the conference was about: build an agent on your data. There’s Agent Bricks, the more automated path: describe the task, let the system optimize it. There’s the Mosaic AI Agent Framework, the full developer toolkit with your hands on every control. There’s the governance layer. There’s the model serving and the foundation-model APIs. Each one touches the agent story from a slightly different angle.</p><p>Run the napkin test, the simple test of whether a salesperson can draw the difference on a napkin. Walk up to a booth and ask: I want to build an agent on my data, do I use Agent Bricks or the Agent Framework, and why not both? It’s a fair question. The honest answer is that one is the low-setup path and one is the full-control path, and that line is perfectly clear to Databricks and genuinely fuzzy to the buyer standing there. Same hallway, same four doors, same hesitation I described at Microsoft, just in a building ten times the size.</p><p>If I stopped here, I’d be writing the same essay again, and you’d be right to be bored. Plenty of people can tell you a big platform has confusing overlapping products. That’s true and it’s old. The interesting thing happened one floor up.</p><h3>One floor up: building the cure</h3><p>Look at the other things Databricks put on stage, though, and a different kind of product comes into focus: not more doors, but the thing that makes a building full of doors navigable. Three of them, and you should look at what they actually are, not just what they’re called.</p><p>Genie. A way for any employee to ask questions of the company’s governed data in plain language. The clever part isn’t the chat box. It’s that Genie deliberately runs inside other companies’ tools. It now plugs natively into Microsoft’s Copilot Studio and Foundry, so a Microsoft-built agent can pull trusted answers from Databricks data without the user ever opening Databricks. Read that twice. Databricks isn’t fighting Microsoft for the agent. It’s making itself the trusted data underneath Microsoft’s agent.</p><p>Unity AI Gateway. A control plane, the layer that sits above everyone’s models and agents and decides what they’re allowed to do: cost caps, routing between models on price and quality, runtime policy, monitoring. The key detail is that it governs other vendors’ models and agents, not only Databricks’ own. It’s a toll booth that sits above the whole road, regardless of who built the car.</p><p>Omnigent. An open-source “meta-harness”: a layer that sits on top of separate agent tools (Claude Code, Codex, and others) and gives them one shared interface to be combined, controlled, and governed. Databricks describes it as doing for agents what Kubernetes did for servers: a common standard above the mess. And they released it under an open license: Apache 2.0.</p><p>Notice what these three have in common. None of them is another door. Every one of them is a way to make a building full of doors walkable: a layer above the clutter. The product floor has the problem. The strategy floor is selling the cure.</p><h3>The mistake and its cure, from one company</h3><p>Here’s where it stops being a contradiction and starts being the actual story. My first essay was about a mistake. Watching Databricks, I think I’m watching a company that has made that mistake at the product level and learned its cure at the strategy level, at the same time, on the same floor.</p><p>Because the cure isn’t “have fewer products.” It never was. The cure is the move my first essay quietly praised when it pointed to Amazon’s cloud as the company that did breadth right. Amazon won not by having an opinion about every application a customer might build, but by being the ground everyone else builds on: the substrate, the layer underneath. You can have two hundred services and still be navigable if the customer can safely ignore most of them and trust the foundation. Open-sourcing Omnigent is the same instinct. You don’t give away a product you’re trying to sell. You give away a standard you’re trying to own. Kubernetes was Google’s gift that became Google’s position. Apache 2.0 is not generosity. It’s strategy wearing generosity’s clothes.</p><p>So the right way to read the two floors is not “sprawl versus focus.” It’s one ambition seen from two heights. Confusing up close, at the booth, where a buyer can’t tell which of four agent-building doors is theirs. Coherent once you step back, at the altitude where the plan is clear: let everyone else fight over models and agents (the parts that are quickly becoming cheap and swappable) and quietly own the two things that stay valuable underneath all of it: the trusted data, and the control plane that governs whatever runs on top.</p><h3>Why this is the real game now</h3><p>Step back from Databricks for a second, because the same shift is happening everywhere. Two years ago the prize was the best model. That prize is disappearing fast, not because models stopped improving, but because they’re becoming interchangeable. Every one of the three announcements above assumes you’ll run many models, from many vendors, and swap them as prices move. When the model is the cheap, swappable part, the value moves down the stack: to whoever holds the proprietary data the models need, and whoever runs the control plane that decides what the models are allowed to do.</p><p>That is the competition of the next couple of years, and it will be quieter and less glamorous than the model launches that get the headlines. Not “whose AI is smartest” but “whose foundation does everyone else end up standing on.” Databricks sees it clearly. So do the big cloud players. So, in its own corner, does a company like Snowflake. Substrate layers don’t support many winners. They tend to settle on one or two, plus an open standard that everyone agrees to share. Which is exactly why you give the standard away early, before someone else’s becomes the default.</p><p>And there’s a casualty class worth naming, because it’s the one most people reading this are closest to. The squeezed middle is the vendor who built a single agent product: no data foundation beneath it, no governance layer around it. In a world organizing itself into substrates and control planes, that product becomes a checkbox inside someone else’s platform. The deep specialists survive on real depth. The substrate owners survive by being the ground. The un-anchored middle gets absorbed.</p><h3>Before we believe the keynote</h3><p>One more thing, because I don’t want to sound like I bought everything from the stage. Most grand platform layers underdeliver against the slide that announces them. Omnigent is an alpha: early, not production-ready. “Govern every model and agent in one place” is a beautiful sentence and a multi-year, partly-broken reality in practice. The conservative bet is that the strategy is right and the timeline is two or three times longer and messier than the keynote implies.</p><p>Which, if you think about it, is a second and quieter version of the same problem. Up close, the buyer can’t tell which agent-building door is theirs. One floor up, the platform vision is genuinely coherent, but it’s a promise, not a finished floor you can walk on yet. The clutter is real today. The cure is real on paper. The gap between them is where the next two years of enterprise AI actually gets fought, and where a lot of money gets spent learning which parts of the keynote were true.</p><h3>Three questions, whichever floor you’re on</h3><p>I should name my own bias. I run a consulting firm. A world where the platform is too big to navigate alone is good for an expert: confused buyers hire people to read the map for them. So take the argument knowing the person making it benefits from the confusion he’s describing. I still think the pattern is real, and I think it’s the most useful thing I saw in three days on that floor.</p><p>If you’re building something and wondering which floor you’re on, three questions usually settle it:</p><ol><li><strong>Can a customer ignore this new thing and still get full value from what they already have?</strong> If yes, it’s breadth they can walk past: the Amazon pattern, two hundred services you never have to think about. If no, you’ve added another door to the same room, and the buyer pays for it in confusion.</li><li><strong>Are you adding it to serve a buyer, or to be the layer everyone else stands on?</strong> Both are real strategies. They’re just very different ones, and pretending you’re doing the first when you’re doing the second is how product lines turn into hallways full of doors.</li><li><strong>If your own salesperson can’t draw the difference on a napkin, what makes you think the buyer can?</strong> The napkin isn’t a gimmick. It’s the test the customer is silently running on you, every time they stand in the hallway and try to choose a door.</li></ol><p>Microsoft built its hallway by letting products drift apart over the years. Databricks is building its hallway and, one floor above it, building the elevator that’s supposed to make hallways like that obsolete: for everyone, as an open standard, with itself as the ground underneath. That’s not the same mistake repeated. That’s a company making the mistake and selling the cure in the same building. The cure is the part to watch. It’s also the part that isn’t finished yet.</p><p><em>Originally published at </em><a href="https://www.nasirmaan.com/blog/Leading-AI-Transformation/too-many-doors-to-the-same-room/"><em>https://www.nasirmaan.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ebd01a0d7c46" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Chatbot vs. AI Agent vs. Conversational AI: A Field Guide for Small Business Owners]]></title>
            <link>https://medium.com/@nasirmaan/chatbot-vs-ai-agent-vs-conversational-ai-a-field-guide-for-small-business-owners-9ac13f8c8063?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ac13f8c8063</guid>
            <category><![CDATA[conversational-ai]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[small-business]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Thu, 04 Jun 2026 12:31:01 GMT</pubDate>
            <atom:updated>2026-06-04T12:31:01.093Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Gu_XPWTcWMGwdmVrb1xoPg.png" /></figure><p>You’re not confused because you’re not technical. You’re confused because vendors profit from it. A buyer’s map of the five generations of business AI.</p><h3>TL;DR</h3><ul><li>Most “AI” sold to small businesses is one of five generations of technology — knowing which (and which blend) is how you know what you’re buying.</li><li>Three questions cut through any pitch: where you’re losing customers, how autonomous it actually needs to be, and what’s the cost of being wrong.</li><li>The terminology will keep changing — anchor to the capability curve, the job to be done, and your integration graph.</li><li>Buying principle: optimize for graceful obsolescence — pick the vendor most likely to carry you to 2028, not today’s best.</li></ul><h3>The confusion is manufactured</h3><p>In the last three months, a small business owner I know was pitched by five different companies selling roughly the same thing.</p><p>One called it a chatbot. One called it conversational AI. One called it an AI agent. One called it an AI employee. The fifth called it an AI receptionist. Five names, five sales decks, five pricing models that didn’t line up — and underneath at least three of them, the same core technology in different clothes.</p><p>She did what any reasonable person does when five confident strangers use five different words for the same thing. She froze, bought nothing, and went back to missing calls she’ll never get back.</p><blockquote>She froze, bought nothing, and went back to missing calls she’ll never get back.</blockquote><p>She didn’t freeze because she isn’t smart enough to understand AI. She froze because the people selling it have no incentive to make it clear. Vague categories and futuristic names work in the vendor’s favor: the more confused you are, the easier you are to sell to, and the harder it is to compare one quote to the next.</p><p>This piece is the map they don’t hand you. By the end you’ll have a simple model for the five kinds of business AI, the question that exposes each label on a sales call, three questions that cut through any pitch, and a way of buying that survives the next time the industry renames everything — which it will.</p><h3>The five generations of business AI</h3><p>Business AI didn’t arrive all at once. It came in waves, and each wave built on the last instead of replacing it. That matters, because most products on the market are blends of these waves — and the blend is where the marketing hides.</p><p>Think of phones: smartphones didn’t kill landlines, and plenty of businesses still run a landline behind a modern system. AI is the same. The older generations are still out there, still being sold, often with a fresh coat of paint.</p><p>One caveat. This isn’t an official classification; no standards body ratified “Generation 4.” It’s a practical buyer’s framework — a way to see what’s under the hood regardless of the label on the box. A map, not a law of physics.</p><h3>Generation 1: Scripted</h3><p>What it is. The decision tree. “Press 1 for billing.” On a website, the bot with four buttons that panics if you type anything else.</p><p>What it can do. Predictable paths, fast and cheap. Route a call, answer a fixed menu.</p><p>What it can’t. Anything off-script. Step an inch off the path and it breaks.</p><p>Who it’s right for. Genuinely simple, repetitive routing — and not much else.</p><p>The tell. If the demo only works when you say exactly the right thing, it’s Generation 1, whatever it’s called.</p><h3>Generation 2: Intent-matching</h3><p>What it is. The first “smart” chatbots, roughly 2016–2020 — Dialogflow, Lex, early Watson. It matches your words to a list of predefined intents.</p><p>What it can do. Handle variations of expected questions. “When do you close?” and “are you open late?” map to the same answer.</p><p>What it can’t. Cope outside its trained intents — flexible inside the box, useless outside it. This is the generation behind most of the “I hate chatbots” feeling.</p><p>Who it’s right for. Far from dead: banks, airlines, telecoms, and regulated call flows still run on it, because predictability beats flexibility there. For most new SMB deployments, it’s been overtaken.</p><p>The tell. It handles rephrasings but loops back to “I didn’t quite get that” on anything genuinely novel.</p><h3>Generation 3: Generative</h3><p>What it is. The ChatGPT moment, 2022 on. A large language model that writes fluent answers instead of retrieving canned ones.</p><p>What it can do. Hold a real conversation, understand nuance, sound human.</p><p>What it can’t. Stick to the facts about your business on its own. Raw, it’ll confidently invent your hours, prices, or a policy you never had. Fluency isn’t accuracy.</p><p>Who it’s right for. Almost no small business should point raw Generation 3 at customers. It needs a leash — which is Generation 4.</p><p>The tell. A smooth talker that occasionally says something untrue about your business. If a demo dazzles but you can’t see where it gets its facts, be careful.</p><h3>Generation 4: Grounded generative (RAG)</h3><p>What it is. Generation 3’s fluency, anchored to your real information. The technical name is RAG; the idea is simple — before answering, it looks up the actual facts (your hours, services, documents) and answers from those.</p><p>What it can do. Speak naturally and accurately. With a few integrations, book appointments and capture leads.</p><p>What it can’t. Reason through complicated multi-step problems, or recover gracefully when things get strange. Grounded, not clever.</p><p>Who it’s right for. The sweet spot for most small businesses today: accurate, natural, affordable, and good enough for the vast majority of conversations.</p><p>The tell. The one to look for. Most genuinely good SMB products are Generation 4 — even the ones marketed as something fancier.</p><h3>Generation 5: Agentic</h3><p>What it is. The frontier, 2024 on. The clean definition: a system that independently plans and carries out a multi-step task, picking its own tools and adjusting when something fails. Working a goal, not just answering a question.</p><p>What it can do. Complex, multi-step work with judgment. Chain actions, adapt mid-task.</p><p>What it can’t. Be cheap or fully mature yet. Real agentic capability is expensive and still being figured out, even by the biggest labs.</p><p>Who it’s right for. Specific high-complexity workflows — and businesses that actually have them. Most small businesses don’t, and shouldn’t pay for capability they’ll never use.</p><p>The tell. The most over-claimed word in the market. An “agent” at a small-business price is almost certainly Generation 4 with a few tools attached — fine, as long as you know that’s what you’re buying.</p><h3>The thing to take away</h3><p>New generations layer; they don’t replace. The product you’re sold is rarely one clean generation — it’s a blend, maybe a Generation 4 brain with Generation 1 fallback rules and a Generation 5 word on the box. Knowing the blend is how you know what you’re buying.</p><p>And the lines move. The gap between Generation 4 and 5 is already narrowing; within a year or two, “grounded chatbot with tools” and “agent” may stop being a useful distinction. That’s fine — the borders aren’t the point. Today, while you’re the one being sold to, knowing roughly where a product sits tells you what you’re paying for.</p><h3>The decoder ring</h3><p>Here’s how the words vendors use map onto the five generations — and the one question that cuts through each label on a live call. Keep this open on your next demo.</p><p>TermWhat it usually isThe question to askChatbotAnything from a primitive script to a capable Gen-4 system; the word only tells you it’s text.<em>Does it answer from my actual business information, or from a fixed script?</em>Conversational AIA fancier umbrella term, usually Gen 3–4, often with voice. Sounds more advanced than “chatbot.” Frequently isn’t.<em>What can this do that a good chatbot can’t?</em>AI agentShould mean Gen 5. In small-business sales, usually Gen 4 with one or two integrations. The most inflated word in the category.<em>What specific actions can it take on its own, without a human stepping in?</em>AI assistant / copilotTypically Gen 4, built to help a human rather than replace them.<em>Does it act on its own, or only suggest, while a person stays in control?</em>AI employee / AI receptionistDescribes the job, not the technology. The tech underneath varies enormously.<em>Forget the name — walk me through exactly what happens when a customer calls.</em>RAGNot a product, a technique: it grounds the AI in your real information so it stops making things up.<em>How does it know facts about my business, and what stops it from inventing answers?</em>Fine-tuningRetraining the model on your data. Different from RAG, and rarely necessary for a small business.<em>Why do I need this instead of just feeding it my information?</em>Self-learningAt small-business prices, almost always means “you can retrain it manually.” A human in a loop, not an AI improving on its own.<em>Does it improve on its own, or does someone have to retrain it — and who?</em>Voice AI vs. Chat AIOften sold as separate products by separate companies. The plumbing genuinely differs; the knowledge behind it shouldn’t.<em>Is this one agent across phone and chat, or two products with two setups and two bills?</em></p><p>That covers their product. Before any vendor call, though, three questions about your own business matter more.</p><h3>The three questions that cut through any pitch</h3><p>You could spend a weekend reading about RAG and agentic frameworks. Or you could ask three questions and land somewhere better than most vendors will take you in an hour.</p><h3>Question 1: What channels are you actually losing customers on?</h3><p>Not where customers reach you — where you lose them. For most small businesses that’s two or three channels, not one: the dentist loses patients to voicemail during procedures and Instagram DMs that sit for days; the plumber misses calls on the job and the website chat at 9pm. Most products solve only one — voice vendors do the phone, chat vendors do the website — each with a separate bill and a separate brain that doesn’t know what the other said. If you’re losing customers in more than one place, knowing that up front changes the whole answer.</p><h3>Question 2: How autonomous does it really need to be?</h3><p>Three tiers, and vendors will push the top one regardless of what you need. Tier 1 — answer questions (hours, services, insurance); a grounded chatbot does this cheaply. Tier 2 — take actions: book, qualify, update the CRM. This is where most small-business value lives, and it needs real integrations. Tier 3 — make judgment calls: triage an urgency, decide a custom price, recover a conversation going sideways. Expensive, still maturing, rarely necessary. Pay for the tier you’ll use; a Tier 3 product running a Tier 1 job is an expensive way to answer “what are your hours?”</p><h3>Question 3: What’s the cost of being wrong?</h3><p>A bot that misquotes your hours costs a phone call. One that books the wrong appointment costs a customer. One that mishandles a medical intake or a legal question costs something harder to price. Match the technology — and the spend — to the failure cost, not the marketing copy. A restaurant can shrug off the occasional reservation slip; a law firm running first intake can’t. And in regulated work, “wrong” means liability, and you may have to prove later what the AI said and why — so data handling, consent, and a record of its conversations are part of the failure cost, not a feature to skim.</p><h3>The shortcut</h3><p>Three questions, four minutes of honest thinking — more clarity than any demo, and a filter that holds no matter what the categories get renamed to next year.</p><p>Here’s what those questions tend to produce by business type — a worked example, not a prescription:</p><p>Business typeChannels that matter mostAutonomy you actually needWhat “wrong” costsDental / medicalPhone + website chatTier 2 — book, qualify, routeA lost patient; insurance and clinical accuracy matterLaw firmPhone + chat intake, human-supervisedTier 2, edging toward 3High — confidentiality, liability, a record you can defendHome services (HVAC, plumbing)Phone, after-hours and overflowTier 2 — capture, dispatch, text the detailsA lost job; usually recoverable with a callbackRestaurantPhone + SMS, dinner-rush overflowTier 1–2 — reservations and hoursLow — a missed cover, recoverable</p><p>Notice the pattern: almost nobody lands on Tier 3, and almost everybody needs more than one channel. That gap — between what gets sold and what gets used — is where most of the overpaying happens.</p><h3>The uncomfortable truths</h3><p>Everything so far was the map. Here’s what it usually leaves out — true, decision-relevant, and not in any vendor’s interest to mention.</p><p>Take a real one. A friend works the front desk at a clinic with several doctors. They were pitched “AI” from every direction — some built into their practice-management software, some from outside — all promising to take over the inbox and referral work.</p><p>They picked one, deployed it, and it got roughly half of those referral and inbox tasks wrong. Her job didn’t shrink; it doubled, because every mistake had to be caught and redone by hand.</p><p>The vendor’s line never changed: it’ll learn from the corrections and improve. She logged every error and reported it. It never got better, the bill kept climbing, and after three or four months of doubled work and apologizing to the doctors, they dropped the vendor.</p><p>That clinic now believes AI is a scam and won’t look at a legitimate solution. The overpromise didn’t just cost them money — it poisoned the well for everyone who comes after.</p><p>And notice why it failed: not because the AI was fake, but because of the boring stuff. A knowledge base nobody kept current, a calendar integration that quietly broke, no path to hand a hard case to a human, no one inside the business owning it after the salesperson left.</p><blockquote>The model is usually the easy part. The discipline around it is the hard part, and it’s the part nobody demos.</blockquote><p>None of what follows is a vendor secret — it’s just what someone should have told that clinic before they signed.</p><h3>“AI agent” has become nearly meaningless</h3><p>Two years ago it meant a system that plans across steps, picks its own tools, and recovers when something fails. Now it’s a marketing word — most “agents” sold to small businesses are grounded chatbots with a couple of integrations. No insult; a chatbot that books appointments is useful. But you’re often paying agent prices for chatbot work.</p><h3>“Self-learning” is usually a human in a loop</h3><p>Ask what it means and it’s almost always: you can review transcripts and retrain it by hand. Useful, but not automatic. Fully autonomous, self-improving systems stay rare and immature at small-business prices; a vendor implying otherwise is describing a roadmap.</p><h3>The voice/chat split runs deeper than it should</h3><p>Voice and chat are genuinely different engineering problems — latency, people talking over each other, speech recognition, carrier integration. Separate plumbing makes sense. Separate brains don’t. There’s no good reason your phone AI and your website AI should run on disconnected knowledge — a different memory of your hours and policies depending on how a customer happened to reach you. That fragmentation isn’t a law of nature; it’s vendor history. Telephony companies built voice tools, chatbot companies built chat tools, and you inherited their org chart as your buying problem.</p><p>A friend runs a growing pest-control business on the side — too small for a full-time receptionist, drowning in repetitive calls. He added a website chatbot; it helped, but he still missed calls on the job, so he bought a separate AI phone receptionist from a known brand at a steep monthly fee. He now pays two vendors and is locked into both — not because he needed two products, but because no pitch told him voice and chat could be one system. The confusion didn’t make him freeze. It made him buy twice.</p><p>In fairness: unified isn’t automatically better. In heavily regulated or unusual workflows, a specialized tool can beat a generalist. Integration isn’t a virtue in itself. But for most small businesses, fragmentation costs more in daily friction than the specialization is worth.</p><h3>You’re probably quoted at last year’s prices</h3><p>Model and inference costs have dropped sharply; plenty of incumbents haven’t passed it on. When you compare quotes, you’re partly measuring how recently each vendor was forced to be honest about cost.</p><h3>The real lock-in is your knowledge base, not the AI</h3><p>The bot is replaceable; what isn’t is everything around it — your tuned answers, your integrations, the history of every conversation. That’s the asset, and it’s the vendor’s moat unless you ask one question up front: if I leave, what do I keep?</p><h3>The pattern underneath</h3><p>The root is the same each time: confusion is profitable. None of it survives a direct question — so ask direct questions.</p><h3>The half-life problem</h3><p>Here’s the catch with everything above: a lot of the vocabulary has a shelf life of about two years. “AI agent,” “conversational AI,” “copilot” — picture how they’ll sound in 2028, about like “expert system” sounds now. “Chatbot” was cutting-edge in 2017; today it’s the word for the annoying thing on a website. The names aren’t stable ground. So what do you anchor to?</p><p>Three things hold their shape while the words around them keep shifting:</p><ul><li>The capability curve. Cost per interaction keeps falling fast, the pause before an answer keeps shrinking, and voice quality has already crossed the line where most callers can’t tell. You don’t need to know what the tech will be called next year — only that it’ll be cheaper and better than today’s quote. Buy in a way that lets you benefit.</li><li>The job to be done. Answering customers, capturing leads, booking work — unchanged since you opened, and unchanged when the category gets renamed.</li><li>The integration graph. Your calendar, CRM, phone system, and the record of every conversation. The connections are the durable asset; the bot on top is replaceable by design.</li></ul><p>Which leaves one buying principle: optimize for graceful obsolescence. Don’t pick the best product for today; pick the vendor most likely to carry you to the 2028 version without making you start over. Ask how they handle model upgrades. Ask what happens to your data if you leave. Ask whether their pricing tracks the falling cost of the technology, or whether you’ll still pay today’s rate in two years while their costs drop.</p><p>That last one applies to us too — I build <a href="https://zalena.ai/">Zalena</a>, and “are we still charging what this actually costs to run?” is a question I’d rather you ask us than not. Put every vendor, including mine, through the same questions and the same uncomfortable truths. The ones worth your money won’t flinch.</p><p>The terminology will keep changing. The questions won’t. Learn the questions, and you’ll outlast every renaming.</p><p>Originally published at <a href="https://zalena.ai/">https://zalena.ai</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9ac13f8c8063" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Companies Don’t Die from Doing Too Much. They Die from Doing the Same Thing Twice.]]></title>
            <link>https://medium.com/@nasirmaan/companies-dont-die-from-doing-too-much-they-die-from-doing-the-same-thing-twice-eee51ab09b29?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/eee51ab09b29</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[aws]]></category>
            <category><![CDATA[microsoft-azure]]></category>
            <category><![CDATA[microsoft-copilot]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Tue, 02 Jun 2026 14:01:59 GMT</pubDate>
            <atom:updated>2026-06-02T14:01:59.702Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*THmDZv6-oE4OTOgLy5J_9w.png" /></figure><h3>Microsoft can’t keep its AI products straight. AWS runs more than two hundred services and stays usable. The difference isn’t focus — it’s overlap. And the AI assistants buyers now ask for advice are about to make the bill come due.</h3><p>Try this experiment with a friend who works at Microsoft. Pick one who ships AI products — there are plenty. Ask them to explain, on the back of a coffee-shop napkin, the difference between Microsoft 365 Copilot, Agent Builder, Copilot Studio, and Azure AI Foundry. Which one is for which buyer. When you’d reach for one and not the others. And while they’re at it, where Agent 365 fits, since that’s a fifth thing now.</p><p>Watch what happens.</p><p>The good ones start strong. They own a piece of this and they know their piece cold. Within a minute they hit the seams between their piece and the next one. By minute three they’re hedging. By minute five there is usually a laugh, or a half-apology, and some version of <em>it depends on the customer — honestly, most companies end up using all of them.</em></p><p>That last sentence is the whole article. Hold onto it.</p><p>This is not a failure of intelligence or effort. It is a structural fact about a product line that no single person can hold in their head — the same fact a customer hits when trying to architect a solution, the same fact a salesperson hits when trying to qualify a deal, the same fact a partner hits when trying to recommend one.</p><p>It is tempting to call this a focus problem and move on. Microsoft does too much, the theory goes. Big companies always do. The cure is to narrow down — be more like a startup. The theory is satisfying, and it is wrong. The reason it’s wrong is sitting in plain sight inside the same industry.</p><h3>The obvious answer doesn’t survive a single counterexample.</h3><p>Amazon Web Services offers more than two hundred distinct cloud services. By any “doing too much” theory of corporate failure, AWS should be an incomprehensible mess, drowning in its own complexity, losing to whichever focused upstart everyone is currently celebrating.</p><p>It is, instead, the default substrate of the modern internet and one of the most profitable franchises ever built.</p><p>So the framing breaks on contact. If a company with two hundred products can stay usable, and a company with a handful of overlapping AI products can confuse its own engineers, then the <em>number</em> of things a company does is not the variable that matters. Something else is doing the work.</p><h3>AWS isn’t clean either — and the way it isn’t clean is the whole point.</h3><p>Before going further, let me concede the obvious objection, because anyone who actually uses AWS is already forming it. AWS has overlap. Plenty of it. You can run containers on ECS, on EKS, or skip servers entirely with Lambda. You can store relational data in RDS or in Aurora. You can query data with Athena or warehouse it in Redshift. You can call foundation models through Bedrock or build your own on SageMaker. These are not crisply separated products. Anyone claiming AWS is a perfectly orthogonal catalog has not filed an AWS bill.</p><p>But look at <em>how</em> AWS overlaps, because it is a different kind of overlap than Microsoft’s, and the difference is the entire argument.</p><p>AWS’s overlaps sit on a <strong>single, statable axis</strong>. ECS, EKS, and Lambda are points on a spectrum from “I want maximum control and Kubernetes portability” to “I never want to think about a server.” Aurora versus RDS is a question of how much performance and managed scaling you’re willing to pay for. Athena versus Redshift is ad-hoc querying versus a provisioned warehouse. In every case, you can write the rule for choosing in one sentence, and that sentence references <em>your</em> situation, not Amazon’s org chart. The choices are additive — more options along a clear dimension — not substitutive.</p><p>Microsoft’s AI overlaps don’t resolve to a clean axis, and the boundaries won’t hold still. Agent Builder is sold as the simple option — until you outgrow it, at which point you “upgrade” into Copilot Studio. Copilot Studio is the enterprise option — except you can also use it as a front end while Foundry does the actual reasoning underneath. Foundry is the custom option — except picking it too early is, by consultants’ own accounts, the single most expensive mistake teams make. And sitting above all of them is Agent 365, a separate product whose job is to govern the agents the other three produce. To choose, you don’t answer one question. You hold four or five intersecting questions at once — licensing, distribution channel, governance plane, who maintains it — and the products keep absorbing each other’s features, so the lines you drew last quarter have moved.</p><p>That’s the distinction the rest of this rests on. Not breadth versus focus. <strong>Additive overlap that sits on a clear axis, versus substitutive overlap where products compete to answer the same question and the boundaries blur.</strong> AWS leans hard toward the first. Microsoft’s AI portfolio is deep in the second.</p><h3>The framework, made portable.</h3><p>If you take one tool from this piece, take this. There are two independent variables, and only one of them is dangerous.</p><p>Breadth is the number of things you sell. Overlap is how much those things compete to solve the same problem. The bottom-left box — high breadth, low overlap — is <em>Platform</em>, and it’s where AWS lives and thrives. Breadth is not the failure mode. Breadth plus overlap is. The bottom-right box — high breadth, high overlap — is the <em>Product swamp</em>, and it’s where Microsoft’s AI portfolio currently sits. Notice that the top-right box is dangerous too: you can be a small company with only two products and still confuse everyone, if those two products fight over the same job.</p><p>The startup advice everyone repeats — “do one thing well” — is really advice to live in the top-left box. That works. But it quietly implies the only safe move is to stay small, and the <em>Platform</em> box proves that’s false. You can be enormous and clear. You just can’t be enormous and redundant.</p><h3>The product map is the org chart in a costume.</h3><p>There’s a structural reason redundancy happens, and it has a name. Melvin Conway noticed it in 1967: organizations ship their own communication structure. The shape of the software ends up mirroring the shape of the org chart. If you have four powerful groups all chasing the same opportunity, you will ship four powerful products that all chase the same opportunity. The redundancy on the outside is the redundancy on the inside, made visible.</p><p>Microsoft’s confused AI portfolio is not, primarily, a strategy failure. It is an org-chart failure rendered in product form. The Microsoft 365 group, the Power Platform group, and the Azure group each had a credible claim on AI, each had executive air cover, and each shipped. The customer was handed the job of assembling the result.</p><p>This is why “just hire a chief product officer” rarely fixes it. The overlap is downstream of organizational reality, and the organizational reality is that several powerful units cannot be told to stand down without political costs executives are unwilling to pay. The overlap survives because the org chart survives.</p><h3>“But some overlap is on purpose.” Sometimes. Rarely the way you think.</h3><p>Here is the strongest objection to everything above, and it deserves a real answer rather than a dismissal: not all overlap is a mistake. Large companies sometimes ship overlapping products <em>deliberately</em>, because different buyers, different sales motions, and different price points genuinely call for different products. Toyota and Lexus overlap enormously and it is not an accident. AWS offers both a managed Kubernetes service and a serverless one because the buyers really are different people with different problems. A little internal competition can even drive better products.</p><p>This is true, and Microsoft makes exactly this argument. The official framing is that these aren’t competing products at all — they are different <em>layers</em>. You consume M365 Copilot, you compose in Copilot Studio, you customize in Foundry. By this account the overlap is apparent, not real; it only looks redundant because everything wears the “Copilot” label. Several thoughtful consultants agree, and they have a point about the underlying architecture: the products do connect to one shared ecosystem.</p><p>So how do you tell healthy, intentional overlap from the swamp? There’s a clean test: <strong>intentional overlap is healthy only when buyers can self-select without expert help.</strong> A Toyota buyer and a Lexus buyer sort themselves in the parking lot. Nobody publishes a 3,000-word guide titled <em>Toyota vs. Lexus: How We Decide for Every Client.</em></p><p>Microsoft’s AI portfolio fails that test, and the evidence is the existence of the explanation industry around it. When the layers are genuinely distinct, you don’t need a dozen consultancies writing side-by-side comparisons, you don’t find the official documentation scattered across three separate sites, and the honest answer to “which one should I use” isn’t “all of them.” Distinct layers don’t generate a translation economy. Overlapping ones do. “A different team built it” and “it’s technically a different layer” are not the same thing as “a different buyer, with a different job, who can tell which door is theirs.” The first two are facts about Microsoft. Only the third would make the overlap intentional in the way that works.</p><h3>Startups have an unfair advantage they don’t even know they’re using.</h3><p>This is also where the usual focus cliché needs correcting. Chatbase, Neon, Clerk — the standard examples of focused startups beating bigger companies — aren’t winning because they <em>chose</em> focus. They’re winning because their size <em>forced</em> it. A six-person team can only build one coherent thing; it does not have the headcount to spin up three overlapping versions of it. Focus, in other words, is downstream of constraint. It’s a phase scarcity hands you for free.</p><p>Which is why the same founders sprawl exactly like the giants the moment they have abundance. OpenAI ships a consumer app, a developer API, an enterprise tier, an agent platform, and hardware ambitions. Nvidia long ago stopped being a chip company and became chips, networking, full systems, software stacks, and a cloud. Neither has obviously drowned — yet — and the reason is instructive: so far, both have sprawled more like a Platform than like a Product swamp, adding capabilities that compose rather than capabilities that compete. The lesson isn’t that focus is sacred. It’s that once scarcity lifts and you can finally afford to build the third overlapping thing, the discipline has to come from somewhere other than your bank balance. Almost nobody supplies it on time.</p><h3>The tax was invisible for decades. AI search is about to print the receipt.</h3><p>For a long time the overlap tax was paid quietly. It happened inside confused buyers’ heads, inside stalled procurement cycles, inside the consulting fees companies paid firms whose entire business was translating a vendor’s product line back into plain English. Real money — but invisible, because it never showed up on an income statement as a line item called <em>confusion</em>.</p><p>Something is changing the visibility, and it’s worth looking at directly rather than predicting it. So I ran the test this article’s argument implies, in mid-2026, the way a buyer now would: I asked the question across the web and the AI assistants people actually consult before they ever talk to a vendor. <em>Which Microsoft product should I use to build an internal AI assistant?</em></p><p>The answers do not converge. They can’t, because the source material doesn’t. The consistent shape of the response — from Microsoft’s own documentation, from independent consultancies, from the assistants summarizing both — is <em>it depends, and most mature organizations end up using all of them.</em> One widely shared consultant guide lays out Agent Builder for quick wins, Copilot Studio as the cross-functional standard, and Foundry for the custom layer, then concludes the real question is never which tool is best. Another, walking through Copilot Studio versus Agent 365, opens by saying that being asked to choose between the two is itself the tell — that the question reflects confusion about what each one does. Microsoft’s own Learn pages split the guidance across separate sites for each product, which is why a cottage industry of “how we decide for every client” explainers exists at all.</p><p>And the cost is no longer hypothetical. The single most expensive mistake these consultants report is buyers picking the wrong overlapping product: a team hears “AI agent,” assumes it needs developers, commits to Foundry, spends three months building, and ends up with something the low-code option would have delivered in two weeks. That is the overlap tax with a number attached — months of wasted engineering, caused entirely by a product line a buyer couldn’t read.</p><p>Here’s the part the giants haven’t absorbed. AI assistants have <em>no more</em> ability to disambiguate an overlapping product line than a human buyer does — and arguably less, because they can’t pick up the phone and ask a partner. They pattern-match across everything written about a vendor, and when the vendor’s own materials are internally contradictory, the recommendation comes back hedged, conditional, or wrong. Clean primitives get clean recommendations. Overlapping suites get a paragraph that begins <em>it depends.</em> The buyer increasingly trusts that paragraph and never visits the vendor’s site to have the confusion cleared up in person.</p><p>So the companies that built tidy, composable product lines are about to discover an enormous accidental advantage in AI-mediated discovery. The companies that built swamps are about to watch the invisible tax become a visible one — recorded, every day, in the answers an assistant gives to a buyer they’re not in the room for.</p><p><em>(A fair caveat: assistant outputs shift week to week, so the specific responses above are a dated snapshot, not a fixed law. But the underlying mechanism — that an AI can only recommend as clearly as a vendor’s own materials allow — doesn’t move.)</em></p><h3>“Just break it up” sounds right. It usually isn’t.</h3><p>The intuitive endpoint of all this is to split the giant into ten focused independents, each with its own inch of wall to own. Decomposition is a real instrument, and occasionally the right one: eBay spinning off PayPal unlocked huge value precisely because the two shared almost no real economies and were clearer apart than together.</p><p>But decomposition is the crude instrument, and reaching for it first usually misreads the problem — because Amazon disproves the premise. Internally, Amazon runs as something close to a thousand startups: the two-pizza teams, the mandate that every team expose its work through clean service interfaces and communicate only that way. Each team behaves like a focused company. Yet Amazon never split into a thousand legal entities, because the unified balance sheet <em>is</em> the moat — AWS profits subsidized retail price wars no pure-play retailer could survive, and data flows across units no outsider would have built together.</p><p>So the honest rule is narrow: <strong>splitting solves overlap only when the units lack meaningful synergies.</strong> Microsoft’s pieces are not held together by mere inertia; they share genuine, valuable synergies — distribution, enterprise relationships, the cloud underneath. Splitting would trade a fixable problem (overlap) for a permanent loss (synergy). The fix that fits isn’t clarity by amputation. It’s clarity by architecture: a ruling on what each product is <em>for</em>, enforced hard enough that a customer can choose without a consultant, and a willingness to merge or kill the redundant ones. It’s unpopular because it means telling a powerful executive their product is the redundant one. It’s also the only move that actually works.</p><h3>The three questions worth asking of your own product line.</h3><p>This works whether you run a division of a giant, a Series B startup, or a single product. Don’t ask <em>are we doing too much?</em> AWS does an enormous amount and is fine; that question sends you the wrong way. Ask instead whether you’re in the swamp — and there are three quick diagnostics:</p><ol><li><strong>Can a new customer tell where one product ends and the next begins — without a sales call?</strong> If buyers can’t self-select, your “different layers” are overlap wearing a costume.</li><li><strong>Can your own salesperson explain why Product A still exists if Product B already does most of what it does?</strong> If the honest answer is “a different team owns it,” you’ve found redundancy, not strategy.</li><li><strong>Can an AI assistant recommend one of your products without a paragraph of caveats?</strong> This used to be a soft concern. It is now a distribution channel.</li></ol><p>If the answer to any of these is no, you don’t have a breadth problem. You have an overlap problem. You are paying for it today in stalled deals and confused buyers, and you are about to pay for it again — in front of a far larger audience — in the recommendations machines give to people who will never visit your website to sort out the confusion themselves.</p><p>You do not earn the right to be broad by being big. You earn it by being clear — and clarity, unlike size, is a choice you have to keep making.</p><p><em>Sources for the AI-discovery section (snapshot, mid-2026), for linking on publication:</em></p><ul><li><em>Microsoft Learn — “Choose between Agent Builder and Copilot Studio”: learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/copilot-studio-experience</em></li><li><em>“Agent Builder, Copilot Studio, or Azure AI Foundry: How We Decide for Every Client” (pnp.github.io)</em></li><li><em>“Agent 365 vs. Copilot Studio: Which One Does What” (valoremreply.com)</em></li><li><em>“Agent Builder vs Copilot Studio vs Foundry — 2026 Guide” (aguidetocloud.com)</em></li><li><em>“Microsoft 365 Copilot, Copilot Studio, and Azure AI Foundry: Understanding the Differences” (blog.chiffers.com)</em></li><li><em>“Copilot Studio vs Azure AI Foundry: Which Platform in 2026” (wrvishnu.com)</em></li></ul><p><em>Originally published at </em><a href="https://www.nasirmaan.com/blog/Leading-AI-Transformation/companies-dont-die-from-doing-too-much/"><em>https://www.nasirmaan.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eee51ab09b29" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Dimensionality Test: A Better Framework for Knowing Where AI Belongs in Your Business]]></title>
            <link>https://medium.com/@nasirmaan/the-dimensionality-test-a-better-framework-for-knowing-where-ai-belongs-in-your-business-e5886762ea05?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/e5886762ea05</guid>
            <category><![CDATA[tech-in-business]]></category>
            <category><![CDATA[generative-ai]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[dimensionality-test]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Thu, 14 May 2026 01:54:38 GMT</pubDate>
            <atom:updated>2026-05-25T17:30:20.829Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*biS-Y_x3pq_cNUgcQzLMDQ.png" /></figure><p>Walk into any SMB-focused subreddit and the same question shows up every week: <em>Do small businesses really need AI? Do brick-and-mortar shops need automation? Is this hype, or actually useful?</em></p><p>The question keeps recurring because most answers are bad. Vendors tell you yes. Skeptics tell you no. Both are arguing the wrong thing.</p><p>The real question isn’t whether your <em>business</em> needs AI. Businesses don’t have problems. <strong>Tasks</strong> have problems. And once you start looking at AI through the lens of tasks — not industries, not company size — the picture gets a lot clearer.</p><p>Here’s a framework I find useful.</p><p>Short on time?</p><p>Get the framework as a one-page PDF, plus a 26-page workshop guide. <a href="https://mailchi.mp/342813f116a4/datalya-dimensionality-test-lead-magnet">Get the PDFs →</a></p><h3>Two axes, not one</h3><p>Every task in your business can be plotted on two axes:</p><p><strong>Repetition</strong> — how often does this task happen? Once a year, once a month, twenty times a day?</p><p><strong>Dimensionality</strong> — how many variables meaningfully affect what the right output looks like?</p><p>This is not new ground. Economists have been studying which tasks suit automation since David Autor’s 2003 work on “routine vs non-routine” cognitive tasks, and in 2017 Erik Brynjolfsson and Tom Mitchell published a 23-criteria rubric for “Suitability for Machine Learning” that’s now been applied to over 18,000 work activities in the U.S. economy. What’s worth updating is the language and the implications — because LLMs have broken some of the predictions those earlier frameworks made.</p><p>Repetition is the axis everyone already talks about. “Automate repetitive tasks” has been the advice of every productivity blog since 2005. It’s true, but incomplete. If repetition were the only criterion, the right tool would still be Excel macros and Zapier — not AI.</p><p>The axis that actually distinguishes where AI shines is <strong>dimensionality</strong>.</p><h3>What “dimensionality” actually means</h3><p>A low-dimensional task is one you can write down as a flowchart. Take a customer’s name, email, and order number. Look up the order in the database. Send a tracking link. Done. There’s a finite number of branches and they all fit on a whiteboard.</p><p>A high-dimensional task is one where you <em>can’t</em> write a finite flowchart, because the right output depends on too many subtle variables interacting at once. Three quick tests will tell you which kind you’re looking at:</p><ul><li><strong>The Instruction Test.</strong> Could a competent stranger do this correctly with a single-page training document?</li><li><strong>The Edge-Case Test.</strong> Does roughly every fifth instance break the rules of the previous four?</li><li><strong>The Substitution Test.</strong> Would two equally competent employees, given the same input, produce visibly different outputs?</li></ul><p>If any of these tip toward <em>no, one page isn’t enough</em> or <em>yes, edge cases are common</em> or <em>yes, employees would disagree</em> — you’re in high-dimensional territory.</p><p>These tasks aren’t repeatable in the rote sense. Each instance is a <em>little bit different</em> from the last. But the <em>pattern</em> is repeatable. That distinction is the whole game.</p><h3>What automation actually pays for</h3><p>There’s an underrated economic insight beneath the dimensionality framing:</p><p><strong>Traditional automation reduces volume cost.</strong> RPA, scripts, and macros let you handle more instances of the same task at lower marginal cost. They scale repetition.</p><p><strong>AI reduces variance cost.</strong> LLMs let you handle the unpredictable variation <em>between</em> instances without writing a rule for each variation. They scale judgment.</p><p>If your task is expensive because there are a lot of instances of the same simple thing — that’s a volume problem. Hire a script, an autoclave, or an offshore team. If your task is expensive because each instance requires interpretation, context, or a small judgment call — that’s a variance problem. That’s the cost LLMs were built to attack.</p><p>The “high-rep, high-dim” quadrant is just the geometric statement of <em>high volume and high variance, both at once</em>. That’s where AI has a unique edge that no prior tool offered.</p><h3>Why this distinction matters now</h3><p>For decades, automation only worked on the bottom-left of this 2×2:</p><p>If a task was low-dimensional and high-repetition, you could automate it with rules. If it was high-dimensional, you needed a person — and the more a task depended on context, judgment, or nuance, the more it stayed locked behind human attention. The economics literature in the early 2000s explicitly predicted this: non-routine cognitive tasks were supposed to stay safely human, because rules-based computers couldn’t handle them.</p><p>LLMs broke that prediction. They are <em>generalization engines over high-dimensional spaces</em> — they don’t follow rules; they learn the shape of a subspace from examples and produce sensible outputs for inputs they’ve never seen before. That capability is what unlocks the top-right quadrant for the first time.</p><p>That’s why “AI for small business” feels both overhyped and undersold at the same time. The hype is wrong about which tasks. The skepticism is wrong about whether <em>any</em> tasks fit.</p><h3>Want this on a single page?</h3><p>The framework in one visual page — handy for sharing with your team or pinning above your desk.</p><h3>A working example: a dental practice</h3><p>Take a dentist with a four-person staff. The Reddit question would ask: “Does this dentist need AI?” Wrong question. Let’s plot the tasks instead.</p><ul><li><strong>Sterilizing instruments.</strong> High repetition, low dimensionality. Don’t use AI here. Use a checklist and an autoclave.</li><li><strong>Performing a root canal.</strong> Low repetition per patient, extremely high dimensionality. Don’t use AI here either. This is where human expertise lives.</li><li><strong>Submitting insurance claims.</strong> High repetition, low-to-medium dimensionality. Traditional automation handles this well.</li><li><strong>Answering the phone.</strong> <em>High repetition, high dimensionality.</em> Every caller is a slightly different problem with a slightly different intent. This is the AI candidate quadrant.</li><li><strong>Choosing whether to buy a new CBCT scanner.</strong> Low repetition, high dimensionality. Human judgment, possibly informed by AI tools, but not delegable to AI.</li></ul><p>Notice that “the dentist’s business” doesn’t fit on the chart at all. Tasks fit on the chart. Some have a clear AI answer; most don’t.</p><h3>Why “candidate” and not “sweet spot”</h3><p>A first draft of this framework would stop here. It would conclude: high-rep + high-dim = AI sweet spot. Plug in a tool and go.</p><p>That conclusion is too strong. Living in the high-rep, high-dim quadrant makes a task an AI <em>candidate</em> — not an AI <em>fit</em>. Four filters determine whether a candidate actually flies. And critically: <strong>if any one of these filters fails catastrophically, the others can’t compensate. You don’t average your way around a veto.</strong></p><h3>1. Verification cost</h3><p>When the AI gets it wrong, can someone catch it cheaply?</p><p>Voice AI for a dentist works in part because callers self-correct (“no, I meant Tuesday”) and in part because someone at the practice glances at the day’s call log and notices anything odd. Mistakes surface quickly and are cheap to repair. AI-assisted medical diagnosis fails the same test in reverse: even if the model is 95% accurate, verifying <em>which</em> 5% is wrong costs as much as doing the diagnosis from scratch. The 5% you don’t catch is the lawsuit.</p><h3>2. Failure severity</h3><p>If the AI fails on a hard case, what is the worst plausible consequence?</p><p>Some errors are recoverable: a confused caller hangs up and calls back; a misrouted email gets re-sent. Some errors are catastrophic: a missed cancer indication, a botched contract clause, a wrong dose. AI deployed where the worst case is catastrophic is a liability regardless of average accuracy. The bar to ship isn’t “right most of the time” — it’s “wrong in a way you can survive.”</p><h3>3. The tacit-knowledge ceiling</h3><p>Michael Polanyi’s old observation, beloved of economists: <em>we can know more than we can tell.</em></p><p>A lot of expert work is high-rep and high-dim <em>and</em> still resists automation, because the relevant expertise has never been written down. A senior trial lawyer’s instinct for which witness to call first. A head waiter’s read of which table to seat a difficult regular at. A master plumber’s ear for which gurgling sound means the wax ring is failing rather than the trap. None of that is in any training dataset. The AI will sound right while being wrong.</p><p>There’s a useful second-order observation here: the process of <em>trying</em> to automate a tacit task often forces an organization to articulate it for the first time. Even AI projects that fail outright tend to leave behind a written workflow, a checklist, or a shared vocabulary that didn’t exist before. That residual value is real — but you should treat it as a consolation prize, not the win.</p><h3>4. Feedback loop density</h3><p>How quickly does the system find out it was wrong?</p><p>Phone AI gets fast feedback: callers correct it in real time, the calendar shows whether bookings landed, the receptionist hears about complaints same-day. Contract-review AI gets slow feedback: an error compounds for months before anyone notices, and by then five other contracts have been signed on the same flawed pattern.</p><p>Fast feedback loops let you catch and fix early — and the system gets better in production. Slow feedback loops let small errors silently scale into big ones. For a task in the AI candidate quadrant, the feedback loop is often what determines whether the deployment improves over time or quietly degrades.</p><h3>Subspace coverage: why vertical AI wins</h3><p>A related point worth flagging: LLMs only generalize well over the subspaces they’ve actually been trained on or grounded in. A general-purpose model is fluent in customer service language, calendar scheduling, basic medical and legal vocabulary, and a thousand other common subspaces. It is <em>not</em> fluent in your proprietary internal jargon, your specific clinical pathway, or the niche language of an industry that doesn’t generate much public text. This is the gap that vertical AI products fill — and it’s the practical reason a dental-specific voice agent will outperform a general one on a dental front desk.</p><p>The phone-answering example survives all four filters and lives in a well-covered subspace: callers self-correct, the relevant expertise is articulable, mistakes are recoverable, feedback is fast, and language models have seen enormous volumes of dental-adjacent customer service language. That’s why it’s the SMB use case that consistently delivers ROI rather than just demoware.</p><p>A 2023 study at the ACM UMAP conference (Salimzadeh et al.) found that human-AI team performance actually <em>decreases</em> as raw task dimensionality rises — a finding that contradicts the naive theory but aligns perfectly with the four-filter view. Dimensionality alone isn’t a green light. Verification, severity, articulability, and feedback are what turn a candidate into a fit.</p><h3>One more thing: workflow fit</h3><p>Even a task that passes all four filters can fail in deployment if the AI doesn’t fit how work actually happens. If staff have to switch systems, double-check every output, override the AI more often than they accept it, or paper over its errors with manual reconciliation — the organization has added friction rather than removed it. The MIT NANDA <em>GenAI Divide</em> report (2025) found that about 95% of enterprise GenAI pilots fail, and that success correlates not with task suitability alone but with deep workflow integration. A well-fitted candidate plus poor integration produces shelfware.</p><p>This isn’t a fifth filter so much as a precondition: before you invest in any of the surviving candidates, ask whether the AI’s output will drop into the existing workflow without being re-handled. If the answer is “we’ll figure it out later,” figure it out first.</p><h3>The exercise to actually do</h3><p>Forget industry reports. Forget vendor pitches. Spend thirty minutes doing this:</p><ol><li>List every recurring task someone in your business performs more than ten times a week.</li><li>Run the three dimensionality tests on each (instruction, edge-case, substitution). Any task that fails at least one is high-dimensional — a candidate.</li><li>For each candidate, score it on the four filters: verification cost, failure severity, articulability, and feedback density. <strong>If any single filter scores catastrophic, the task is out — regardless of how strong the others are.</strong></li><li>Check workflow fit. Will the AI’s output land in the existing workflow without being manually re-handled? If not, fix that before investing.</li><li>The candidates that survive — high rep, high dim, four filters clean, workflow fit confirmed — are your real AI investments.</li></ol><p>Most businesses, when they do this honestly, find one or two such tasks. Not twenty. Not zero. One or two. Those are the ones worth investing in.</p><h3>The framing worth holding onto</h3><p>The reason “should small businesses use AI?” generates so much circular debate is that it’s a category error. Businesses aren’t the right unit of analysis. Tasks are.</p><p>Repetition tells you whether automation is worth the setup cost. Dimensionality tells you whether <em>AI specifically</em> — variance-cost reduction — is the right kind of automation, versus a script, a checklist, or a hire. The four filters tell you whether an AI candidate actually deserves to ship. Workflow fit tells you whether it will survive contact with how your business actually runs.</p><p>The businesses that will get real value out of AI in the next five years aren’t the ones who adopt it broadly. They’re the ones who identify the small number of tasks in their operation that genuinely live in the high-repetition, high-dimensionality quadrant — survive all four filters cleanly — and then integrate the chosen one or two into how work actually gets done.</p><h3>Ready to actually do this exercise?</h3><p>Get the free 26-page workshop guide — built for a 2-hour leadership session. Plus the one-page summary.</p><p><strong>Everything else is theatre.</strong></p><p><em>Originally published at </em><a href="https://datalya.com/the-dimensionality-test-a-better-framework-for-knowing-where-ai-belongs-in-your-business/"><em>https://datalya.com</em></a><em> on May 14, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e5886762ea05" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The $126,000 line item that doesn’t exist]]></title>
            <link>https://medium.com/@nasirmaan/the-126-000-line-item-that-doesnt-exist-6c274f8c07ab?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/6c274f8c07ab</guid>
            <category><![CDATA[small-business]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[customer-experience]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[voice-assistant]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Wed, 13 May 2026 19:42:07 GMT</pubDate>
            <atom:updated>2026-05-13T19:42:07.947Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AVkZ6e-zOQwI1cCL18SlHw.png" /></figure><h4>The average small business loses $126,000 a year to missed calls and unanswered messages. Here’s why the loss stays invisible — and what it takes to see it.</h4><p>The average small business loses around $126,000 a year to missed calls, <a href="https://www.getaira.io/blog/missed-business-calls-statistics">according to research aggregated by Aira from sources including 411 Locals, PATLive, and AMBS Call Center</a>. It doesn’t show up on any P&amp;L.</p><p>There’s no row in QuickBooks for callers who hung up. No GL code for the website visitor who closed the tab and went to a competitor. No expense category for the contact form that took 36 hours to answer. The loss is real, recurring, and measurable, but it sits entirely outside the systems most owners use to run their business. This post is about how big that loss actually is, why it stays invisible, and what it takes to finally see it.</p><h3>Where the $126K comes from</h3><p>The number isn’t a guess. It comes from straightforward math applied to publicly available benchmarks.</p><p>Start with the miss rate. A study by 411 Locals tracked 85 businesses across 58 industries and found that only 37.8% of inbound calls were answered by a live person. The remaining 62% went to voicemail or got no response at all.</p><p>Add the non-callback rate. According to PATLive, 85% of callers who don’t reach a live person never call back. Most of them don’t even leave a voicemail — about 80% hang up rather than record one. And 78% of customers buy from the first business that responds to their inquiry, which means the call that didn’t get through is usually a call that closed somewhere else.</p><p>Multiply those rates by a per-call value and the math does the rest. The standard formula used across the industry is:</p><p><em>Monthly missed calls × Average customer value × 12 × 0.85 = annual revenue loss.</em></p><p>A small service business missing 25 calls a month at a $400 average customer value loses about $102,000 a year. A slightly busier business missing 40 calls a month at the same value loses $163,000. The widely-cited $126,000 figure is the middle of that range — and it’s the floor for businesses big enough to track this. For most SMBs, who don’t track it at all, it’s almost certainly higher.</p><p>These benchmarks vary by industry, geography, and call intent, but even conservative assumptions produce meaningful loss. And the call data understates the real number, because every modern small business has a second front door — the website chat box, the contact form, the inquiry that comes in at 11pm and doesn’t get answered until Thursday. Those losses don’t show up in any phone log either.</p><h3>Why this stays invisible</h3><p>Three reasons explain why owners can spend years bleeding revenue through this channel without ever seeing the line item.</p><p>The first is that there is no accounting category for it. A missed call is the absence of a transaction, not a failed one. The marketing dollars that drove the caller to the business are booked. The call itself, and its outcome, are not. The same is true for a contact form that goes unanswered or a chat session that ends without resolution. Lost revenue from a conversation that never happened doesn’t reconcile against anything.</p><p>The second is that there’s no callback signal. With voice, 80% of callers who reach voicemail never leave a message. The owner literally never knows they were tried. The same blind spot exists on the web side, in a different form — a visitor who closed the tab and clicked the next Google result leaves no trace at all. The visitor counter doesn’t tell anyone they wanted to ask a question. In both channels, the lost prospect is silent by default.</p><p>The third is that marketing dollars hide the leak. A business spending $3,500 a month on Google Ads, with 62% of resulting calls unanswered and a meaningful share of form submissions answered too late, is paying full cost for a fraction of the value. The customer acquisition math looks fine on a dashboard while the actual conversion engine is leaking through the floor. The leak gets noticed only when ad spend goes up and bookings don’t follow.</p><p>In most SMB setups, this data is fragmented across a phone system, a CRM, a website analytics tool, and the owner’s memory. Tools like CallRail and HubSpot capture parts of it. None of them stitch the whole picture together by default. Most owners aren’t ignoring missed conversations — they’re looking at five different tools that each show one slice of the loss, and never the total.</p><h3>Where the loss is sharpest</h3><p>The $126K average smooths over enormous variation between industries. The verticals where voice and chat AI generate the highest measurable returns aren’t the ones with the most calls — they’re the ones where the <em>first contact</em> is the highest-stakes contact.</p><p>In dental practices, 20–38% of calls go missed, mostly because front desk staff are checking in another patient when the phone rings. Each missed new-patient call represents about $850 in lifetime patient value, according to Patient Prism data. A practice missing 10 new-patient calls a month loses over $100,000 a year on this channel alone. Dental websites also see meaningful after-hours traffic — patients comparing options at 9pm on a Sunday — and a static contact form converts at a fraction of the rate of an immediate response.</p><p>Law firms miss approximately 35% of incoming calls, according to Clio’s Legal Trends Report. The economics are worse per call: a single missed inquiry in personal injury or family law can represent $5,000 or more in lost fees. Legal matters are by definition acute. A caller who just got injured, just got served, or just got arrested doesn’t wait for a callback. They call the next firm on the list. A growing share of those inquiries now start as chat — someone Googling “personal injury lawyer near me” at 11pm wants to talk to <em>something</em>, not fill out a form and wait until morning.</p><p>Home service businesses miss around 27% of inbound calls (Invoca), with each missed call worth approximately $1,200 in immediate revenue — and the miss rate spikes during peak demand. A burst pipe in winter or a failed AC in summer is exactly when a contractor can’t pick up, and exactly when the caller is least patient.</p><p>Veterinary practices have a similar dynamic to dental, but with a sharper after-hours problem. Pet owners calling about a sick animal at 8pm need a real answer about whether to drive to the emergency clinic or wait until morning. A voicemail is the wrong answer at that moment, and the practice that fails to provide one tends to lose the client permanently.</p><p>The pattern across all four verticals is the same: the value of an instant response isn’t proportional to the volume of inquiries. It’s proportional to the stakes of the <em>first</em> contact — whether that contact is a phone call, a chat message, or a form submission.</p><h3>What this looks like in practice: Geek Square</h3><p><a href="https://geeksquare.ca/">Geek Square</a> is a Toronto-based computer repair business, and one of Zalena AI’s earliest chat deployments. Before the chat widget went live, the website had the standard SMB setup: a contact form, a generic email address, and a phone line that rolled to voicemail outside business hours. Customers with a broken laptop on a Saturday evening either filled out a form and hoped for a Monday reply, or moved on to the next shop on Google.</p><p><strong>Two patterns showed up in the deployment data quickly.</strong></p><p>About 30% of chat conversations come in outside business hours. That’s nearly a third of all inbound conversations that, before the widget, would have been a contact form submission with a 36-hour response time at best — or no submission at all. Computer repair, like most home services, is a category where the customer’s pain is acute. They want to know if their hard drive is recoverable, what it costs, and how soon you can look at it. They don’t want to wait until Monday.</p><p>The chat widget converts roughly 4–5 of those weekly conversations into qualified bookings, with full contact information, the device details, and a transcript that the team reviews before the customer arrives. The response is instant, the lead is scored, and the team’s first interaction with the customer is on a job that’s already been triaged. None of these conversations existed in any system before — no row in the CRM, no entry in the call log, no line item anywhere.</p><p>This is what the $126K problem actually looks like in a single small business. Not a dramatic catastrophe. Just a steady, weekly recovery of conversations that were previously invisible.</p><h3>The reframe: making the invisible visible</h3><p>The default pitch for AI assistants is that they answer when the team can’t. That’s true, but it undersells what the technology actually does.</p><p>The real value is that every inbound interaction gets logged, structured, and searchable. For the first time, an SMB owner can see exactly how many calls came in last Tuesday between 6pm and 9pm, what those calls were about, how many ended in a booking, and what objections came up. They can see how many website visitors started a chat, what they asked, and at what point the conversation ended without a resolution. They can see the question pattern by season, by day of week, by hour.</p><p>This is the data SMBs have never had access to. It required a person to answer every call and every chat and write everything down — which no SMB has ever done at scale.</p><p>Voice and chat aren’t substitutes for each other. They’re two halves of the same dataset. The customer who calls today is often the same customer who chatted yesterday or filled out a form last week. Most small businesses have no way to know this. Each channel is a black box, and the connections between them are invisible.</p><p>When the conversations get captured, the accounting line item that didn’t exist starts existing. Once it exists, it can be managed.</p><p>Voice AI makes the loss visible for the first time. Chat AI does the same for the conversations that started somewhere other than the phone. Together they cover the full inbound surface.</p><h3>What to actually do about it</h3><p>For owners who want to act on this without buying anything, four steps come first.</p><p>Measure the current miss rate. Most VoIP systems already log call answer rates. Pull the last 90 days. The number is almost always higher than expected. For chat, check whether the website widget logs sessions that end without resolution — most don’t, by default, which is itself the problem.</p><p>Calculate the per-conversation value honestly. Use lifetime customer value, not first-transaction value. A missed dental patient isn’t an $850 loss, it’s a $7,500 loss. A missed legal inquiry isn’t a $5,000 loss, it’s a $5,000 loss <em>plus</em> the referrals that prospect would have generated.</p><p>Look at the time-of-day distribution for both channels. Calls peak at lunch, after 5pm, and on Mondays. Web traffic skews even more heavily toward evenings and weekends. If the team is structurally unavailable during those windows, no amount of training fixes it.</p><p>A caveat worth saying out loud: not every missed call is recoverable revenue. Some are price shoppers who would never have closed. Some are misdials. Some are existing customers calling about something that didn’t need a same-day response. But enough of them are high-intent first-time inquiries that the economics still work — and the only way to know the actual ratio in a given business is to start capturing the conversations.</p><p>Decide the threshold. If the combined miss-cost across calls and web inquiries is below $20,000 a year, this isn’t the highest-priority operational fix. If it’s above $50,000 — and for most SMBs that hit either threshold, it usually is — then it almost certainly is.</p><p>The point isn’t to install a tool. The point is to start seeing a number that’s been there all along.</p><p><em>I’m building </em><a href="https://zalena.ai/"><em>Zalena AI</em></a><em> — voice and chat assistants for the small businesses where missed conversations cost the most: dental practices, law firms, vet clinics, home services. Built in Ottawa, deployed across North America. If you want to see what your missed-conversation cost actually looks like, </em><a href="https://zalena.ai/"><em>chat with us</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6c274f8c07ab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Kids Should Be Coding Python Amid COVID-19 Crisis?]]></title>
            <link>https://medium.com/@nasirmaan/why-kids-should-be-coding-python-amid-covid-19-crisis-befeef3f7f1b?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/befeef3f7f1b</guid>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Wed, 22 Apr 2020 21:26:22 GMT</pubDate>
            <atom:updated>2026-06-01T03:08:04.452Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*i7oyACIWxUZmxOJwV-Jiew.png" /><figcaption>Learn Python: Kids, COVID-19 &amp; Coding</figcaption></figure><h3>Digital Transformation Is Almost Over</h3><p>Python has recently become the second most used programming language in the world. Globally, programmers are in very high demand, and it would be more so in the future due to the increasing digitization of the global economy. Computing technologies and data are the backbone of modern digital economies. Over the last two decades, most organization have transformed their manual business processes into rule-based and data-driven ones. Amid digital transformation, the compute and disk storage started becoming cheap — consequently, the organization began accumulating vast volumes of data.</p><h3>AI Transformation Is Up Next</h3><p>The next phase of transformation is about AI transformation. We are going to migrate into AI space where almost everything, whether it is businesses, education, healthcare, or vehicles, would be AI-enabled.</p><p>For a successful transformation, public and private organizations need to conquer the following challenges;</p><ul><li>process the vast amount of accumulated data,</li><li>make sense of it through exploratory analysis</li><li>train machine learning models, and</li><li>replace rule-based business processes with AI-powered smarter and automated processes</li></ul><p>Data science and machine learning are going to be critical ingredients of AI transformation. That is why most organizations are still hiring data scientists amid the COVID-19 crisis.</p><h3>Global Economy Needs Python Programmers</h3><p>Python is the programming language of data science and machine learning. Most organizations are struggling to hire and retain data scientists with strong programming skills. A good Python programmer can quickly learn important data science and machine learning skills. That is why the kids at school must start learning Python programming now. So that in the future, they can easily pick up data science and machine learning related courses at universities.</p><p>Governments across the world are in the process of introducing programming as a mandatory subject in schools. The new curriculums and degree programs will also be involving intensive computer programming.</p><h3>Python For Kids</h3><p>In summer 2019, from <a href="https://datalya.com/">Datalya’s platform</a>, I delivered Python for Kids course. In the class, we had students who were studying in grades 3, 4, and 5 (ages 8 to 11 years). We adapted the idea of parent-kid pair programming, as the kids in lower grades would need help with the implementation of programming instructions. We kept it a hands-on programming course, where students would be explained the concept first and then asked to do practice on the computer by writing Python code. At the end of the course, both parents and kids were pleasantly surprised and happy that later had picked up the core concepts of Python programming.</p><p>I am thankful to <a href="https://datalya.com/">Datalya</a> leadership that they agreed to make the course material available for free download amid the COVID-19 crisis.</p><p><a href="https://www.dropbox.com/scl/fi/562yed5fzcga4eoc12yme/Python-for-kids-free-download.zip?rlkey=fsou9fxrjxbuw60jfctzga0po&amp;st=0zqtd5pb&amp;dl=0"><strong>Download Free Learning Material</strong></a></p><p>Probably this is the best time for kids to learn Python programming skills, as they have enough time to learn something different.</p><p>The course has been split into three weeks (slides) as listed below:</p><h4>WEEK 1:</h4><ul><li>Installing and setting up the Python environment</li><li>First <strong>Turtle Program </strong>— a package in Python</li><li>What are <strong>Number Variables</strong>?</li><li>What are <strong>String</strong> Variables?</li><li>How <strong>Lists</strong> work in Python?</li></ul><h4>WEEK 2:</h4><ul><li>What is Python <strong>Dictionary</strong>?</li><li>How <strong>For Loop</strong> works for repetitive tasks?</li><li>How to used <strong>if-else</strong> Statements</li></ul><h4>WEEK 3:</h4><ul><li>What is Jupyter Notebook, and How to use it?</li><li>What Pandas dataframes and how to use them?</li><li>Visualization with Seaborn package</li></ul><p>I would suggest to cover a week’s material by spending 1–2 hours every day. This approach would let the kid(s) absorb core Python concepts and keep them in memory for a more extended period.</p><p>If you have any questions related to the course material, please do not hesitate to <a href="https://nasirmaan.com/">contact me</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=befeef3f7f1b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Future Prospects of Machine Learning in Remote Sensing]]></title>
            <link>https://medium.com/@nasirmaan/future-prospects-of-machine-learning-in-remote-sensing-8780bbc5a979?source=rss-1aaa392ab512------2</link>
            <guid isPermaLink="false">https://medium.com/p/8780bbc5a979</guid>
            <category><![CDATA[remote-working]]></category>
            <category><![CDATA[ideas]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[futuristic]]></category>
            <dc:creator><![CDATA[Nasir Mahmood, PhD]]></dc:creator>
            <pubDate>Mon, 12 Aug 2019 16:13:35 GMT</pubDate>
            <atom:updated>2019-08-12T16:13:35.731Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/660/0*nDnE5wNqq8DBpd2r.jpg" /></figure><h3>Public and private organizations are equally excited about future prospects of machine learning (ML) and artificial intelligence (AI).</h3><p>Over last few years, we have been seeing so many success stories of machine learning applications in marketing, image classification, natural language processing (NLP). However, remote sensing is a field has not come into machine learning focus yet. It has got a wealth of untapped data in form of highly available satellite imagery which can be integrated with socio-economic data in order to build environmental and social threat prediction models. In addition to web and social media data, remote sensing applications can hugely benefit by building their building their own user communities and encouraging them to contribute by verifying assessments.</p><p>Using these such models, one can then build demand-driven applications to help public and private organizations understand, prepare and respond to economic, social, and humanitarian losses in timely manner. These applications can be made near-real-time remote sensing applications by adding additional layer of response tools, like alerts and advisories etc. Ideally these applications should have the capability to be scaled from local to global level for disaster and risk management.</p><p>Recently a case study remote sensing application has been built for floodplain prediction in Senegal. It has been noticed that floods cause more economic, social, and humanitarian losses than any other threat. It leveraged satellite imagery data using Google’s cloud based Earth Engine and turned turn data into insight for decision-makers on the ground. Following are fewer of its objectives:</p><ul><li>Disaster risk reduction</li><li>Event prediction for timely response and recovery</li><li>Fill information gaps about exposure and vulnerability to flooding</li><li>Allow stakeholder to better target investments and protect assets.</li></ul><p>Official census data were one of key component of training set, as reported its variation is effectively explain by five key features: 1) a lack of basic informational resources, (2) age of population, (3) disabilities, (4) distance to main population centres and (5) population increase due to migration. These factors reportedly explain 69% of the variation in the census. These features combined with remotely sensed satellite data were used to train machine learning model for floodplain prediction. More about this study can be found under <a href="https://link.springer.com/chapter/10.1007/978-3-319-65633-5_16">https://link.springer.com/chapter/10.1007/978-3-319-65633-5_16</a></p><p><em>Originally published at </em><a href="https://datalya.com/blog/machine-learning/future-prospects-of-machine-learning-in-remote-sensing"><em>https://datalya.com</em></a><em>.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8780bbc5a979" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>