<?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 Daniel Kioko on Medium]]></title>
        <description><![CDATA[Stories by Daniel Kioko on Medium]]></description>
        <link>https://medium.com/@itsdanielkioko?source=rss-9544aee41bf6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*iSkqxQwnI5042NroAeQZ5w.png</url>
            <title>Stories by Daniel Kioko on Medium</title>
            <link>https://medium.com/@itsdanielkioko?source=rss-9544aee41bf6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 08:25:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@itsdanielkioko/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[AI Help Desk vs. Traditional Help Desk: What’s Actually Different?]]></title>
            <link>https://medium.com/thecodr/ai-help-desk-vs-traditional-help-desk-whats-actually-different-d81b00bbd111?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/d81b00bbd111</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[customer-support]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[client-support]]></category>
            <category><![CDATA[helpdesk]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Thu, 09 Jul 2026 07:40:49 GMT</pubDate>
            <atom:updated>2026-07-09T07:54:29.854Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lreTun3_KodVD-tjSd27Hg.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@fedeprax?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Federico Passi</a> on <a href="https://unsplash.com/photos/people-walking-on-sidewalk-near-building-during-daytime-975KTV6a0m4?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>If you’ve spent any time evaluating help desk software recently, you’ve noticed that every vendor now claims to be “AI-powered.” The word is everywhere — in the headlines, the feature grids, the sales decks. And yet the actual experiences these platforms deliver can be worlds apart.</p><p>Some of them have genuinely reimagined how support works. Others have a chatbot bolted onto a ticket queue and a marketing team that found the right vocabulary.</p><p>The question worth asking isn’t “Does this platform have AI?” Almost all of them do, in some form. The question is: <em>what is the AI actually doing, and how does that change the experience for your team and your users?</em></p><p>This article breaks down the real differences — not in marketing language, but in architecture, workflow, cost, and outcomes. By the end, you’ll have a clear framework for distinguishing genuine AI help desks from traditional platforms with AI features, and you’ll know which approach fits your team’s actual situation.</p><h3>First: What Is a Traditional Help Desk, Really?</h3><p>To understand what’s different, it helps to be precise about what the traditional model actually is.</p><p>A traditional help desk is built around the ticket. A user experiences a problem and submits a request — via email, a web form, phone, or chat. That request becomes a ticket. The ticket enters a queue. An agent picks it up, reads through it, opens whatever other systems they need to find context or execute a resolution, responds to the user, and closes the ticket when the issue is resolved.</p><p>The traditional helpdesk is structured and reliable — but it’s also manual and reactive.</p><p>That model has powered IT and customer support for over two decades. The platforms built on it — Zendesk, Freshdesk, ServiceNow, Jira Service Management in its earlier forms — generated billions in revenue and handled billions of tickets. The model works.</p><p>But it has three structural limitations that become increasingly costly as organizations scale:</p><p><strong>It scales linearly.</strong> More tickets require more agents. There’s no architectural mechanism for absorbing volume growth without proportional headcount growth.</p><p><strong>It’s inherently reactive.</strong> Someone has to report a problem before the system responds. There’s no proactive capability built into the model itself.</p><p><strong>It’s manually intensive.</strong> Every step — triage, routing, context-gathering, response drafting, resolution — requires human time and attention, regardless of how routine the request is.</p><p>These aren’t bugs. They’re how the system was designed, for an era when they were acceptable trade-offs. In 2026, they’re not.</p><h3>What Makes an AI Help Desk Actually Different</h3><p>Here’s where the distinction matters most: a genuine AI help desk doesn’t just speed up the traditional workflow. It changes the workflow at an architectural level.</p><p>Traditional helpdesks transform support from manual document tracking into autonomous, intelligent resolution — and the gap between legacy systems and AI-first platforms has widened into a chasm in 2026. While legacy helpdesks struggle with linear scaling (where more tickets require more people), AI agents offer non-linear growth. You can handle a 10x spike in volume without hiring a single additional person.</p><p>Let’s break down exactly where and how the model changes.</p><h3>The 8 Real Differences</h3><h3>1. Resolution vs. Routing</h3><p>The most fundamental difference is what the system does with an incoming request.</p><p>A traditional help desk routes tickets to humans. The human resolves them. Every request, regardless of complexity, requires human involvement to reach resolution.</p><p>An AI help desk resolves directly. When a request arrives — a password reset, a software access question, a VPN configuration issue — the AI reads it, understands the intent, and takes action. For a password reset, that means verifying identity and executing the reset, end-to-end, without a human ever seeing the ticket.</p><p>AI-capable helpdesks now resolve 40–70% of inbound tickets autonomously, compared to just 14% of issues that fully resolve through traditional self-service tools. That gap isn’t a feature comparison. It’s a fundamentally different philosophy about what the system is supposed to do.</p><p>The distinction matters because routing is a cost. Every ticket that reaches a human agent costs time and money, even if the resolution takes 30 seconds. Resolution before the ticket enters the queue eliminates that cost entirely.</p><h3>2. Response Time: Minutes vs. Seconds</h3><p>The performance gap in response time between traditional and AI help desks is stark — and it compounds in ways that matter deeply to user experience.</p><p>Traditional help desks typically operate with response time goals measured in minutes or hours. Even well-run operations target first response times of 5–60 minutes during business hours. After hours, the model either breaks down entirely or requires expensive staffing arrangements to maintain.</p><p>AI-powered agents typically respond in under 30 seconds — and this near-instant response significantly improves customer satisfaction scores because the user never feels like they are waiting in a queue.</p><p>For many routine requests, AI doesn’t just respond faster — it resolves completely before a traditional team would have assigned the ticket. The user’s problem is gone before they’ve refreshed their inbox.</p><h3>3. Availability: Business Hours vs. Always On</h3><p>Traditional help desks are staffed by people, which means they’re constrained by when people work. Providing genuine 24/7 support in a traditional model requires either a follow-the-sun staffing model across multiple time zones, expensive on-call arrangements, or accepting degraded service after hours.</p><p>An AI agent is always on. It provides the same level of service at 3:00 AM on a holiday as it does on a Tuesday morning, and does so without any additional staffing costs.</p><p>For organizations with global teams, remote workforces, or users in multiple time zones, this difference alone represents significant value. The Friday evening outage no longer waits until Monday morning to get addressed.</p><h3>4. Scalability: Linear vs. Non-Linear</h3><p>This is the economic argument that matters most for growing organizations.</p><p>In a traditional help desk, the relationship between ticket volume and cost is essentially linear. Double the tickets, double the agents. This model works at low volumes. It becomes increasingly painful — and increasingly expensive — as the organization scales.</p><p>AI agents offer non-linear growth. You can handle a 10x spike in volume without hiring a single additional person. A product launch, a system-wide outage, a sudden influx of onboarding requests — these create volume spikes that break traditional support capacity planning. An AI help desk absorbs them without breaking a sweat.</p><p>The implications for cost structure are significant. Rather than staffing to peak demand, teams can staff to genuine complexity — the requests that actually require human judgment — and let AI handle everything else regardless of volume.</p><h3>5. Cost Per Ticket: $22 vs. Under $3</h3><p>The financial comparison between traditional and AI help desks is where organizational leaders often have their “why haven’t we done this already” moment.</p><p>The industry standard for a traditional human-led support interaction sits between $17 and $22 per ticket in 2026. When you factor in the hidden costs of traditional support — recruitment, training cycles, office space, and the high cost of agent burnout — the true expense is even higher. AI agents, by contrast, handle a typical interaction for around $3.</p><p>Cost per resolution falls to $1–3 when AI handles routine requests end-to-end, down from $8–12 with human agents.</p><p>For a team handling 500 tickets per day, this difference — at scale, over a year — is the difference between a support function that hemorrhages budget and one that operates efficiently. The math isn’t subtle.</p><h3>6. Knowledge: Static FAQs vs. Dynamic Intelligence</h3><p>Traditional help desks typically pair with a static knowledge base — a library of articles that agents reference and users can search. The knowledge base is only as useful as its maintenance. In most organizations, it goes stale. Articles get outdated, new issues don’t get documented, and users quickly learn that searching the knowledge base is usually a dead end.</p><p>Modern AI helpdesks use Retrieval-Augmented Generation (RAG) to sync with your entire knowledge base — reading Confluence pages, Notion docs, and Google Drive files to find the exact answer to a specific question. They don’t just point the user to a link; they synthesize the answer into a helpful, specific response.</p><p>But the deeper change is how knowledge grows. An AI help desk continuously analyzes resolved tickets, identifies questions that didn’t have documented answers, and generates draft knowledge base articles to fill those gaps. AI answers are only as good as the docs and past tickets feeding them — but tools that learn from your real ticket history tend to hit significantly higher resolution rates over time.</p><p>The knowledge base shifts from a static reference library to a living intelligence layer that improves with every ticket the system handles.</p><h3>7. Agent Experience: Tab-Switching vs. AI-Assisted Context</h3><p>For the tickets that do reach human agents, the day-to-day experience is dramatically different in an AI-augmented environment.</p><p>In a traditional help desk, agents open a ticket and start investigating. What’s the user’s history? Have they had this issue before? What did the last agent try? What systems is the user on? Answering these questions means switching between applications, searching ticket history, checking asset records, and reading through email threads. This context-gathering takes time — often as much time as the actual resolution.</p><p>An AI help desk compresses all of this into what the agent sees when they open the ticket. The AI has already summarized the relevant history, identified the probable cause based on similar past tickets, and suggested resolution steps. In a study of 256,934 customer service chats, agents using AI-based suggestions saw a 22% drop in response times — and that’s just from suggested responses, not the full suite of AI assistance available in modern platforms.</p><p>The agent’s job shifts from investigation to decision-making. They validate the AI’s context, exercise judgment on anything requiring nuance, and focus their expertise where it actually matters.</p><h3>8. Insight Generation: Reporting vs. Intelligence</h3><p>Traditional help desks generate reports. Ticket volume by category, average resolution time, SLA compliance rates, agent performance metrics. This data is useful for operational management, but it’s backward-looking — it tells you what happened, not what’s coming.</p><p>AI help desks generate intelligence. Pattern detection across ticket categories reveals emerging issues before they become widespread. Sentiment analysis flags users who are repeatedly frustrated before they escalate. Predictive analytics surface tickets likely to miss SLA thresholds before they do. Anomaly detection identifies unusual volume spikes that might indicate a system problem worth investigating proactively.</p><p>Modern platforms bring in sentiment tracking, knowledge recommendations, and real-time insights so support leaders can actually improve the experience — not just measure it.</p><p>The shift is from a function that reports on what happened to one that helps leaders understand what’s happening and anticipate what’s coming.</p><h3>Side-by-Side: The Honest Comparison</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OcBDgUid-Yx4XYFGTtu8FQ.png" /></figure><h3>What AI Help Desks Are NOT Better At</h3><p>An honest comparison requires acknowledging where traditional models still have advantages — or where AI introduces new challenges.</p><p><strong>Complex, emotionally sensitive situations.</strong> More than half of employees deliberately avoid the help desk because of poor experiences. That’s what happens when you try to automate humans out of the loop rather than design them in from the start. An AI that handles a difficult situation poorly — a user who’s frustrated, a complaint with emotional weight, a scenario that requires genuine empathy and judgment — can damage trust in ways that take significant time to repair. The best AI help desks know where to escalate, not just how.</p><p><strong>Edge cases and novel problems.</strong> AI resolves what it has seen before, or what closely resembles patterns it’s encountered. Genuinely novel problems — new software conflicts, unusual configurations, issues that fall outside historical ticket patterns — require human judgment. The risk of an AI confidently providing a wrong answer is real, and it’s worse than no answer at all. As one practitioner put it in a 2026 discussion on enterprise AI support: “A bot that confidently fails is worse than no bot, because now people don’t trust it.”</p><p><strong>Knowledge base quality dependency.</strong> AI is only as intelligent as the information it draws from. A poorly maintained, incomplete, or inconsistent knowledge base doesn’t become better just because AI is searching it. Teams that invest in AI without investing in documentation quality often see underwhelming results and blame the AI, when the real issue is the foundation it’s built on.</p><p><strong>Implementation investment.</strong> Traditional help desk software can be set up quickly with minimal configuration. A genuine AI help desk requires thoughtful implementation — training the model on your ticket history, mapping your knowledge base, defining escalation paths, validating AI accuracy on your specific request types before broad deployment. That investment is worth it, but it’s real.</p><h3>The False Binary: AI vs. Traditional</h3><p>Here’s the insight that gets lost in most comparisons: the best-performing support operations in 2026 aren’t purely AI or purely traditional. They’re thoughtfully hybrid.</p><p>The most effective support operations use AI to handle the front line — every incoming message hits the AI first, which responds instantly, resolves what it can, and collects context for what it can’t. Messages that AI can’t resolve get routed to the right human based on topic, language, customer value, or complexity. And AI assists human agents even on the tickets they handle.</p><p>The goal isn’t to replace human support with AI. The goal is to ensure that human support is focused exclusively on the problems that require human judgment — complex troubleshooting, sensitive situations, nuanced decisions, relationship-critical interactions — while AI handles everything else with speed and consistency no human team can match.</p><p>74% of organizations now use AI in at least one service management function, and the gap between teams that have adopted AI-driven service desks and those still running traditional ticketing systems is widening fast.</p><p>The teams that are falling behind aren’t the ones that chose AI over traditional support. They’re the ones that haven’t yet made a deliberate choice about where AI fits — and are therefore getting neither the efficiency of full automation nor the relationship depth of a human-centered model.</p><h3>How to Evaluate What’s Real vs. What’s Marketing</h3><p>Given that every vendor now claims AI capabilities, here’s a practical framework for evaluating whether a platform is genuinely AI-powered or traditionally-built with AI features attached.</p><p><strong>Ask about autonomous resolution rate.</strong> What percentage of incoming requests does the platform resolve without any human involvement? If the answer is vague or below 30%, the AI is assisting humans — not replacing manual triage.</p><p><strong>Test on your actual ticket types.</strong> Demo environments are curated. The real test is whether the platform can complete your top 10 customer requests end-to-end without human intervention. If it cannot process an access request or execute a password reset during the trial, it will not do it in production either.</p><p><strong>Ask where the AI lives in the workflow.</strong> Does it operate before tickets are created (true deflection), or only after tickets enter the queue (assistance)? Platforms that only touch tickets after creation are fundamentally still routing-first, regardless of how capable their AI is.</p><p><strong>Evaluate the escalation design.</strong> How does the AI decide what to escalate? How does it hand off context to humans when it does? Poor escalation design is where AI help desks lose user trust fastest — and the best platforms invest heavily in making the human handoff seamless and well-informed.</p><p><strong>Check the knowledge base architecture.</strong> Is the knowledge base static (manually updated) or dynamic (AI-updated based on resolved tickets)? The difference determines whether the system gets smarter over time or slowly becomes less useful as your environment evolves.</p><h3>The Bottom Line</h3><p>The difference between an AI help desk and a traditional help desk isn’t a set of features. It’s a different answer to a fundamental question: <em>who — or what — is responsible for resolving a support request?</em></p><p>Traditional help desks answer: humans, assisted by software that helps them manage their workload.</p><p>AI help desks answer: intelligent automation, supervised by humans who handle what the AI can’t.</p><p>That difference in architecture produces all the downstream differences in response time, cost, scale, and experience. And in 2026, with ticket volumes growing, user expectations rising, and the talent required to staff traditional models harder to find and retain than ever, the architecture you choose isn’t just a technology decision. It’s a strategic one.</p><p>The good news: you don’t have to choose between them cold. The best path is to start with the highest-volume, most automatable workflows — where AI delivers immediate, measurable results — and expand from there as your team builds confidence in the system and the system learns your environment.</p><p>That’s not a future-state aspiration. It’s what leading IT and support teams are doing right now.</p><h3>See What the Difference Looks Like for Your Team</h3><p>Meridesk is built as a genuine AI Help Desk — not a traditional ticketing system with AI features, but a platform designed from the ground up around autonomous resolution, intelligent escalation, and continuous learning.</p><p>If you want to understand what the difference looks like in practice for your specific environment, <a href="https://meridesk.live/">start a conversation at meridesk.live</a>.</p><p><em>Follow for more practical guides on AI-powered support, IT automation, and building a modern help desk that scales.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d81b00bbd111" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/ai-help-desk-vs-traditional-help-desk-whats-actually-different-d81b00bbd111">AI Help Desk vs. Traditional Help Desk: What’s Actually Different?</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Reduce L1 Support Tickets by 60% With AI Automation]]></title>
            <link>https://medium.com/thecodr/how-to-reduce-l1-support-tickets-by-60-with-ai-automation-3d593e61b1b0?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d593e61b1b0</guid>
            <category><![CDATA[customer-support]]></category>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[customer-service]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[business]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Thu, 09 Jul 2026 07:06:01 GMT</pubDate>
            <atom:updated>2026-07-09T07:06:01.076Z</atom:updated>
            <content:encoded><![CDATA[<p><em>By the Meridesk Team · meridesk.live</em></p><figure><img alt="Photo by BaljkanN 4 on Unsplash" src="https://cdn-images-1.medium.com/max/1024/1*e_WrIGujCr6ZbwhyuT1jEw.jpeg" /></figure><p>If you manage a help desk, you already know what L1 support feels like from the inside: password resets before 9 AM, software access requests that sit in a queue for hours, the same onboarding questions arriving in slightly different forms every week.</p><p>These aren’t complex problems. They don’t require skilled engineers. They don’t need human judgment. But they consume the majority of your team’s time, day after day.</p><p>The good news: this is exactly the category of work that AI automation was built to eliminate. And organizations doing it right aren’t just shaving a few tickets off the pile — they’re cutting L1 volume by 40, 50, even 60 percent, while their agents focus on work that actually demands their expertise.</p><p>This article shows you how.</p><h3>What “L1” Actually Means — And Why It Matters</h3><p>L1 (Tier 1) support refers to the first line of contact for end-user issues. These are high-volume, low-complexity requests that follow predictable patterns and can be resolved with documented steps. Common L1 categories include:</p><ul><li>Password resets and account unlocks</li><li>Software installation and provisioning</li><li>Access requests and permissions changes</li><li>VPN and connectivity troubleshooting</li><li>Onboarding and offboarding tasks</li><li>Status inquiries (“Is this system down?”)</li><li>FAQ-level questions about policies or tools</li></ul><p>According to the Fixify 2026 Benchmark Report, software provisioning and onboarding/offboarding tasks alone drive over 50% of IT tickets. Most of these follow scripts. Most of them don’t need a human.</p><p>The cost of handling them manually is significant. Industry benchmarks put the average L1 ticket at $22 each — and for SaaS-heavy organizations, that figure runs higher. When you’re handling hundreds of these daily, the math adds up fast.</p><h3>Why Traditional Deflection Strategies Don’t Work Well Enough</h3><p>Before AI, the answer to L1 volume was simpler: build a knowledge base, add a FAQ page, hope users find it.</p><p>And those approaches help — to a point. A well-structured knowledge base does deflect a portion of tickets. But static self-service has real limits. Users still need to know what to search for, where to look, and how to interpret the answer they find. When those conditions aren’t met, they open a ticket anyway.</p><p>The second traditional approach is headcount. Hire more L1 agents. The problem: ticket volumes are growing faster than most organizations can staff. Internal service desk ticket volumes have climbed 16% since the pandemic — and that increase hasn’t been matched by proportional staffing growth. The structural capacity gap widens every year.</p><p>AI doesn’t just improve on these approaches. It changes the game entirely.</p><h3>The 5 AI Mechanisms That Actually Reduce L1 Volume</h3><p>Reducing L1 tickets by 60% isn’t a single strategy — it’s the combined result of several AI-powered mechanisms working together. Here’s how each one works.</p><h3>1. Conversational AI That Resolves Before a Ticket Is Created</h3><p>The most powerful deflection happens before a ticket ever enters the queue. When a user starts to describe a problem — in a chat interface, a Slack bot, or a support portal — an AI agent can understand the intent, pull from your knowledge base, and resolve the issue conversationally without any human involvement.</p><p>This isn’t a keyword-matching chatbot from 2015. Modern AI agents use Natural Language Processing to understand requests even when phrased informally, with typos, or with incomplete information. They learn from historical ticket data, improving their resolution accuracy over time.</p><p>The impact is real. AI agents are now capable of resolving 40–60% of incoming support requests automatically, according to Gartner’s 2024 research on B2B SaaS platforms. One organization, AssemblyAI, saw its AI resolution rate climb from the high 20% range to nearly 50% of all incoming chats — dramatically freeing its human team for complex issues.</p><p>The key point: when AI resolves an issue conversationally, that issue never becomes a ticket. Your deflection rate improves not because you resolved the ticket faster, but because the ticket was never created.</p><h3>2. Intelligent Triage That Eliminates Queue Bottlenecks</h3><p>For tickets that do enter the queue, AI triage ensures they reach the right destination instantly — no manual sorting, no misrouting, no billing questions sitting in the technical support queue for three hours waiting for someone to notice.</p><p>AI uses intent detection, sentiment analysis, and language understanding to classify incoming requests by topic, urgency, and appropriate team. It does this continuously and at scale, without the bias or fatigue that comes with human triage.</p><p>The performance numbers are striking. AI-driven ticket classification reduces response time by 50% compared to manual triage, while maintaining 98% accuracy in priority assignment. Without automated tools, support teams spend 12–17 hours per week on manual sentiment tracking and ticket hygiene alone. With AI, that drops to 1–2 hours.</p><p>That reclaimed time doesn’t disappear — it goes toward resolution.</p><h3>3. Self-Service That’s Actually Findable</h3><p>A knowledge base users can’t navigate is invisible. AI changes this by powering intelligent search that understands what users are actually trying to do, not just the keywords they typed.</p><p>When a user types “I can’t get into my email on my laptop” into an AI-powered portal, the system doesn’t return a generic list of articles. It surfaces the specific steps for their OS, their email client, their organization’s configuration — drawn from your knowledge base and past resolution data.</p><p>Self-service implementation at this level consistently reduces ticket volume by 20–40%. The content is the same. The discoverability — and therefore the usefulness — is transformed.</p><h3>4. A Knowledge Base That Writes Itself</h3><p>Here’s a compounding benefit that most teams don’t anticipate: AI doesn’t just help users find existing documentation. It identifies where documentation is missing — and fills the gaps.</p><p>By analyzing resolved tickets, AI can pinpoint the recurring questions that aren’t answered in your knowledge base. It then automatically generates draft articles based on how those tickets were resolved. Your team reviews and publishes. Over time, your self-service layer becomes comprehensive enough to handle a growing share of incoming requests without any human involvement.</p><p>This closes the loop that traditional knowledge bases can’t. Instead of documentation that goes stale and incomplete, you get a living knowledge base that improves as ticket volume grows.</p><h3>5. Proactive Issue Detection</h3><p>The cheapest ticket is the one that’s never created — not because the user found the answer themselves, but because the problem was caught and fixed before it affected them.</p><p>AI can monitor system signals and behavioral patterns to identify emerging issues before they generate a wave of tickets. A software update that’s failing silently. An access permission that’s about to expire for a group of users. A VPN configuration that’s causing intermittent failures.</p><p>When these are detected and resolved proactively, entire categories of L1 tickets simply don’t happen. This is the highest form of support efficiency: invisible resolution before the user even knew there was a problem.</p><h3>What a 60% Reduction Actually Looks Like in Practice</h3><p>Let’s make this concrete. Consider a mid-size IT team managing 400 L1 tickets per week, at an average cost of $22 per ticket. That’s $8,800 weekly — $457,600 annually — just in L1 handling costs.</p><p>A 60% reduction doesn’t mean 240 tickets disappear into thin air. It means:</p><ul><li><strong>~160 tickets</strong> are resolved automatically by AI before or immediately after creation — zero agent time</li><li><strong>~80 tickets</strong> are resolved faster because triage is instant and agents have AI-generated summaries and suggested resolutions ready before they open the ticket</li><li><strong>~160 tickets</strong> remain as genuine L1 work requiring human resolution — but agents are now handling this smaller, well-prioritized queue with dramatically less overhead</li></ul><p>The cost profile changes accordingly. At 60% automated resolution, you’re looking at roughly $180,000 in annual L1 cost savings for this team — not counting the value of what your agents do with the capacity they recover.</p><p>For organizations with higher ticket volumes, the numbers scale proportionally.</p><h3>The Implementation Roadmap: How to Get There</h3><p>A 60% reduction doesn’t happen overnight, and it shouldn’t. The most successful teams approach AI automation incrementally — building confidence with low-risk wins before expanding scope.</p><h3>Phase 1: Audit and Baseline (Weeks 1–2)</h3><p>Before deploying anything, understand your current state clearly. Pull 90 days of ticket data and categorize it:</p><ul><li>What are your top 10 ticket types by volume?</li><li>Which categories have consistent, documented resolution paths?</li><li>Where do your longest handle times occur?</li><li>What percentage of your tickets are genuinely L1 vs. misclassified?</li></ul><p>This audit reveals where automation will have the fastest impact. High-volume, low-variance ticket types — password resets, access requests, software provisioning — are your starting point.</p><h3>Phase 2: Start With One High-Volume Workflow (Weeks 3–6)</h3><p>Pick a single ticket type and automate it end-to-end. Password resets are the classic starting point because the resolution path is completely consistent and the volume is universally high.</p><p>Deploy an AI agent that can handle this workflow conversationally: identify the user, verify their identity through your existing authentication layer, execute the reset, confirm completion. No ticket created. No agent involved.</p><p>Measure the results. Track deflection rate, user satisfaction, and any edge cases that required escalation. This data builds the case for expanding scope — and gives your team confidence in the AI’s performance before you hand it more responsibility.</p><h3>Phase 3: Expand to the Full L1 Category (Weeks 7–16)</h3><p>Once your first workflow is running reliably, systematically add the next highest-volume ticket types. At each expansion:</p><ul><li>Train the AI on your historical tickets and knowledge base for the new category</li><li>Run simulations on past ticket data to verify accuracy before going live</li><li>Define clear escalation triggers for edge cases</li><li>Build or update knowledge base articles to support the new AI scope</li></ul><p>By the end of this phase, the majority of your L1 volume should be running through automated resolution paths.</p><h3>Phase 4: Optimize and Compound (Ongoing)</h3><p>AI support automation is not a deploy-and-forget system. The teams that see the highest long-term deflection rates treat their AI as something that needs to be maintained and improved:</p><ul><li>Review AI resolution logs weekly for accuracy and gaps</li><li>Publish new knowledge base articles as the AI surfaces recurring unresolved questions</li><li>Adjust routing logic as your team structure evolves</li><li>Monitor user satisfaction scores on AI-resolved tickets to catch experience issues early</li></ul><p>The system should be getting better over time — and with a platform built to learn, it will.</p><h3>Measuring Success: The Metrics That Matter</h3><p>Once AI automation is running, track these indicators to understand and communicate its impact:</p><p><strong>Deflection Rate</strong> — Percentage of incoming requests resolved without creating a ticket. Aim for 40%+ in the first six months; 60%+ by year one.</p><p><strong>L1 Resolution Rate</strong> — Of tickets that do enter the queue, what percentage are resolved by AI without human involvement? This tracks autonomous resolution separately from pre-ticket deflection.</p><p><strong>First Contact Resolution (FCR)</strong> — Are issues resolved on first interaction, or are users coming back? AI-resolved tickets should have FCR rates of 85%+.</p><p><strong>Average Handle Time (AHT)</strong> — For tickets that do reach agents, how long does resolution take? AI-assisted agents — with summaries, context, and suggested resolutions — should see meaningful AHT reductions even on the tickets they handle.</p><p><strong>Cost Per Ticket</strong> — The ultimate financial metric. Track this monthly as automation scope expands.</p><p><strong>Agent Utilization</strong> — What percentage of agent time is spent on L1 vs. higher-complexity work? The goal is a steady shift toward higher-tier work as L1 volume is absorbed by automation.</p><h3>The Common Mistakes to Avoid</h3><p><strong>Automating everything at once.</strong> The teams that struggle with AI support automation usually tried to do too much too fast. Start narrow, validate, expand.</p><p><strong>Neglecting the knowledge base.</strong> AI is only as good as the information it draws from. If your knowledge base is incomplete or outdated, your AI will hallucinate or fail. Invest in documentation quality before expanding automation scope.</p><p><strong>Ignoring escalation design.</strong> Every automated workflow needs a well-designed escalation path for edge cases. Users who encounter an issue the AI can’t handle should reach a human quickly and without friction. The absence of a good escalation experience is where automation creates frustration instead of satisfaction.</p><p><strong>Treating deflection as the only metric.</strong> A ticket that reaches an agent but gets resolved in 30 seconds with AI assistance is still a win. Track the full picture — not just what was deflected, but how much faster and better everything else moved.</p><p><strong>Not involving your agents.</strong> The teams that successfully implement AI support automation bring their agents into the process early. Agents who understand how the AI works — and who feel like it removes their least interesting work rather than threatening their jobs — become advocates. Agents who feel excluded become obstacles.</p><h3>The Result: A Support Function That Scales Without Breaking</h3><p>The traditional model of IT support has a fundamental scaling problem: as your organization grows, so does your ticket volume — and the only way to handle it is to hire more agents. There’s a ceiling to how many agents you can afford, which means there’s a ceiling to the quality of support you can provide.</p><p>AI automation breaks that ceiling. It absorbs the growth in L1 volume without requiring proportional headcount growth. Your agents handle a smaller, better-prioritized queue of genuinely complex work. Your users get faster answers. Your costs stop growing linearly with your headcount.</p><p>A 60% reduction in L1 tickets isn’t a moonshot. It’s what organizations are achieving right now, with the right platform, the right implementation approach, and a willingness to start small and expand methodically.</p><p>The question isn’t whether this is achievable. It’s how quickly your team is ready to start.</p><h3>Ready to See What’s Possible for Your Team?</h3><p>Meridesk is built to make this kind of transformation practical — not just possible. If you want to understand what a 60% L1 reduction could look like for your specific environment, <a href="https://meridesk.live/">start a conversation at meridesk.live</a>.</p><p><em>Follow the Meridesk blog for more practical guides on AI-powered support, IT automation, and building a modern help desk that scales.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d593e61b1b0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/how-to-reduce-l1-support-tickets-by-60-with-ai-automation-3d593e61b1b0">How to Reduce L1 Support Tickets by 60% With AI Automation</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Claude Produces High-Quality Output: A Developer’s Guide to Token Efficiency and Hallucination…]]></title>
            <link>https://medium.com/thecodr/why-claude-produces-high-quality-output-a-developers-guide-to-token-efficiency-and-hallucination-dd93fc4ad901?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd93fc4ad901</guid>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[learning-to-code]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Fri, 05 Jun 2026 00:32:57 GMT</pubDate>
            <atom:updated>2026-06-05T00:37:28.238Z</atom:updated>
            <content:encoded><![CDATA[<h3>Why Claude Produces High-Quality Output: A Developer’s Guide to Token Efficiency and Hallucination Reduction</h3><p><em>Understanding what’s actually happening under the hood and how to work with it.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C5CQgiTiBy87h6wvelvRDA.avif" /></figure><h3>The Common Misconception: It’s Not About Compression</h3><p>When developers first notice that Claude tends to be concise and accurate, a reasonable assumption is that some novel compression mechanism is at work. A smarter way of packing information into fewer tokens. This is understandable, but it misses the real story.</p><p>The improvements in output quality come from <em>how the model was trained</em>, not from a fundamentally different architecture for compressing text. Understanding this distinction matters, because it changes how you should think about prompting and system design.</p><h3>Token Efficiency: What’s Actually Happening</h3><h3>The Tokenizer Is Only Part of It</h3><p>All large language models, including Claude, use a tokenizer to convert raw text into the numeric tokens the model actually processes. Tokenizers have evolved over time, Claude’s tokenizer handles code, multilingual text, and structured data more gracefully than earlier generations. But this accounts for a relatively small part of the efficiency story.</p><p>The bigger driver is <strong>instruction-following fidelity</strong>. Claude has been trained extensively to do what you ask, including being concise when you ask for conciseness, and verbose when detail is needed. It doesn’t pad responses by default, because padding was not rewarded during training.</p><h3>Practical Implications for Developers</h3><ul><li><strong>Be explicit about length.</strong> If you want a two-sentence summary, say so. If you want exhaustive coverage, say that instead. Claude will calibrate.</li><li><strong>Use system prompts to set a baseline tone.</strong> A system prompt like <em>“Respond concisely. Avoid unnecessary preamble.”</em> will consistently reduce token usage across a session.</li><li><strong>Don’t mistake brevity for capability.</strong> A short answer isn’t a less capable one. If you need more depth, ask for it explicitly.</li><li><strong>Structure your prompts to reduce redundancy.</strong> If your prompt repeats context the model already has, you’re spending tokens on noise. Pass only what’s needed.</li></ul><h3>Hallucination Reduction: The Deeper Picture</h3><p>Hallucinations (Confident, plausible-sounding responses that are factually wrong ) are one of the hardest problems in language model development. Here’s what has actually moved the needle.</p><h3>1. Better Calibration</h3><p>Modern training techniques teach the model to distinguish between what it knows with high confidence and what it’s uncertain about. Rather than filling gaps with invented detail, a well-calibrated model will hedge: <em>“I’m not certain, but…”</em>or simply acknowledge a knowledge gap.</p><p>This is called <strong>epistemic calibration</strong>, and it’s a training objective, not a post-processing filter. The model is rewarded for expressing uncertainty accurately, not just for sounding confident.</p><p><strong>What this means for you:</strong> If Claude says it’s unsure, take that signal seriously. If you need high-confidence factual output, pair Claude with retrieval-augmented generation (RAG) so the model is grounding answers in documents you control.</p><h3>2. RLHF and Constitutional AI</h3><p>Reinforcement Learning from Human Feedback (RLHF) is a technique where human raters score model outputs, and those scores shape future training. Anthropic layers on Constitutional AI; a set of principles the model uses to self-critique and revise its own outputs before they’re finalized.</p><p>These techniques together push the model toward outputs that are not just fluent, but accurate and honest. They penalize overconfident wrong answers and reward appropriate hedging.</p><p><strong>What this means for you:</strong> Claude’s refusals and caveats aren’t bugs, they’re features of honest calibration. If you’re seeing excessive hedging for your use case, tighten your system prompt to establish the domain context clearly, which gives the model more confidence to answer directly.</p><h3>3. Training Data Quality</h3><p>The model learns from what it’s trained on. Higher-quality, more accurate training data means the model has better underlying representations to draw from. This doesn’t eliminate hallucination (the model is still generating probabilistic text), but it raises the floor significantly.</p><h3>4. Improved Reasoning Chains</h3><p>Claude is better at detecting when a line of reasoning has gone wrong mid-generation and correcting before producing the final output. This is partly a result of training on chain-of-thought data; examples where reasoning is made explicit and checkable.</p><p><strong>What this means for you:</strong> For complex tasks, ask Claude to reason step-by-step before producing a final answer. This externalized reasoning is easier for the model to self-check, and easier for you to audit.</p><h3>Practical Strategies for High-Quality Output</h3><h3>Ground truth-sensitive tasks: Use RAG</h3><p>If accuracy on specific facts matters, don’t rely on the model’s parametric memory. Feed it the documents, database excerpts, or API responses it needs to answer from. Claude’s context window is large, so use it.</p><h3>Reduce ambiguity in prompts</h3><p>Hallucinations often happen at the edges of ambiguous instructions. The more precisely you define the task, format, constraints, and expected output, the less room there is for the model to confabulate.</p><h3>Ask for reasoning, not just answers</h3><pre>Instead of: &quot;What is the best database for this use case?&quot;<br>Try: &quot;Walk me through the tradeoffs of PostgreSQL vs. MongoDB for this use case, then give a recommendation.&quot;</pre><p>The reasoning trace makes errors visible and correctable.</p><h3>Use temperature deliberately</h3><p>Lower temperature (0.0–0.3) reduces creative variation and keeps outputs closer to the model’s highest-confidence predictions. This is appropriate for factual or structured tasks. Higher temperature is better for brainstorming or creative generation where variation is valuable.</p><h3>Validate outputs programmatically where possible</h3><p>For structured outputs (JSON, code, classifications), add downstream validation. Claude supports structured output formats; use them, and build checks that catch malformed responses before they cause issues in your pipeline.</p><h3>What Developers Often Get Wrong</h3><p><strong>Assuming more tokens = better output.</strong> Longer prompts can actually degrade quality if they introduce contradictions or bury the key instruction. Keep prompts clear and prioritized.</p><p><strong>Treating caveats as failures.</strong> When Claude says “I’m not certain” or “this may vary,” that’s accurate calibration, not weakness. Build your UX to surface that uncertainty to end users appropriately.</p><p><strong>Not iterating on system prompts.</strong> Your system prompt is the highest-leverage place to improve output quality. Treat it like production code; version it, test it, and refine it.</p><p><strong>Expecting the model to have current information.</strong> Claude’s knowledge has a training cutoff. For anything time-sensitive, supplement with web search or retrieval tools.</p><p><em>For further reading, see Anthropic’s published research on Constitutional AI and their model cards for the latest Claude versions.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd93fc4ad901" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/why-claude-produces-high-quality-output-a-developers-guide-to-token-efficiency-and-hallucination-dd93fc4ad901">Why Claude Produces High-Quality Output: A Developer’s Guide to Token Efficiency and Hallucination…</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Is an AI Support OS? (And Why Your Team Needs One in 2026)]]></title>
            <link>https://medium.com/meridesk/what-is-an-ai-support-os-and-why-your-team-needs-one-in-2026-ea4b131e5c0e?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/ea4b131e5c0e</guid>
            <category><![CDATA[customer-experience]]></category>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[helpdesk]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[customer-support]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Sun, 24 May 2026 01:23:27 GMT</pubDate>
            <atom:updated>2026-05-24T01:23:27.126Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*efikDrEOvQAVewKmXrqivw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@joshhild?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Josh Hild</a> on <a href="https://unsplash.com/photos/a-city-skyline-with-a-bridge-RvTp824JZhU?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>There’s a conversation happening in every IT and support team right now. The tickets are piling up. The same L1 issues keep cycling through. Agents are burned out triaging requests that should never reach a human. And leadership is asking why costs keep climbing while resolution times stay flat.</p><p>The traditional help desk wasn’t built for this. It was built for a simpler era — one ticket, one agent, one resolution. But modern teams operate in a world of distributed workforces, SaaS sprawl, and users who expect answers in seconds, not hours.</p><p>That’s where the AI Support OS comes in.</p><h3>The Problem With the Traditional Help Desk</h3><p>Before we define what an AI Support OS is, it helps to understand what it’s replacing — and why that matters.</p><p>The conventional help desk model is fundamentally reactive. A user hits a problem, submits a ticket, and waits. An agent picks it up, triages it, and either resolves it or escalates. Multiply this across hundreds of daily requests, and you have a function that’s expensive, slow, and deeply unsatisfying for everyone involved.</p><p>The numbers back this up. The average North American help desk agent stays less than 2.5 years and costs $12,000 to replace. Turnover in the industry runs as high as 42%. Teams are stretched thin, and skilled specialists spend the majority of their time on tasks that require neither skill nor judgment — password resets, access requests, software installations.</p><p>Meanwhile, user expectations have fundamentally shifted. In their personal lives, people get instant answers from AI-powered assistants around the clock. When they show up to work and submit a ticket and wait 48 hours for a response, the contrast is jarring.</p><p>The help desk, as most organizations run it today, is a bottleneck masquerading as a solution.</p><h3>So, What Is an AI Support OS?</h3><p>An AI Support OS is more than a help desk tool with a chatbot bolted on. It’s a unified operating layer for your entire support function — one that uses artificial intelligence to automate, anticipate, and continuously improve how support is delivered across your organization.</p><p>Think of it this way: a traditional help desk is a <em>ticketing system</em>. An AI Support OS is the <em>intelligence infrastructure</em> that makes your entire support operation smarter.</p><p>Here’s what that looks like in practice:</p><h3>1. Intelligent Triage and Routing</h3><p>Instead of agents manually sorting and assigning tickets, an AI Support OS reads the content of each incoming request, understands its intent, assesses its urgency, and routes it to the right team or resolution path — automatically, in milliseconds.</p><h3>2. Tier 0 Automation</h3><p>The AI resolves the highest-volume, lowest-complexity requests without any human involvement at all. Password resets. Software access provisioning. Status updates. These are handled conversationally, often before the user has finished typing their request.</p><h3>3. Agent Augmentation</h3><p>For tickets that do require a human, the AI doesn’t step aside — it steps alongside. Real-time suggestions, auto-generated summaries of incident history, recommended resolution paths based on similar past tickets. Agents aren’t replaced; they’re made dramatically more effective.</p><h3>4. Proactive Issue Detection</h3><p>An AI Support OS doesn’t just wait for problems to be reported. It monitors systems and behavioral patterns to identify issues before they become tickets. A device showing early signs of failure. A user repeatedly struggling with the same workflow. A software conflict that’s about to affect a team. These are surfaced and addressed preemptively.</p><h3>5. A Living Knowledge Base</h3><p>Traditional knowledge bases go stale because no one has time to maintain them. An AI Support OS continuously identifies gaps — tickets that resolved without a documented solution — and drafts new articles automatically, keeping institutional knowledge current without adding to anyone’s workload.</p><h3>6. Unified Context Across Every Channel</h3><p>Whether a user reaches out via chat, email, Slack, or a portal, the AI Support OS maintains a continuous thread of context. No more “as I mentioned in my previous ticket…” frustration. The system already knows.</p><h3>Why “OS” Is the Right Word</h3><p>The word “operating system” is intentional.</p><p>Your computer’s OS doesn’t do any one task for you. It provides the foundational layer — memory management, process coordination, resource allocation — that makes everything else possible. Application-level tools run on top of it and benefit from everything it provides.</p><p>An AI Support OS does the same thing for your support function. It doesn’t replace your existing tools. It provides the intelligent layer that connects them, learns from them, and makes them more effective together. Your ticketing system, your knowledge base, your communication channels — they all become part of a unified, intelligent whole.</p><p>This is a meaningfully different architecture than a help desk with AI features. Those products add intelligence as a layer on top of a fundamentally reactive system. An AI Support OS builds intelligence into the foundation.</p><h3>The Numbers That Make the Case</h3><p>If this still sounds abstract, consider what organizations are seeing in practice.</p><p>The global AI customer service market hit $15.12 billion in 2026, growing 25% year over year. That’s not speculative investment — that’s organizations seeing returns and doubling down.</p><p>Human agents cost between $6 and $8 per support interaction. AI handles the same interaction for $0.50 to $0.70. For a team managing 500 interactions per day, that’s a cost structure that changes entirely.</p><p>Support agents using AI tools handle 13.8% more inquiries per hour without additional headcount. ServiceNow reported that its AI agents now resolve 80% of support inquiries autonomously, cutting complex case resolution time by 52%.</p><p>Conversational AI is projected to reduce contact center labor costs by $80 billion in 2026 alone. For every $1 invested in AI support infrastructure, businesses see an average return of $3.50 — with top performers reporting 8x ROI.</p><p>By year three, average ROI on AI support investment exceeds 124%. The payoff doesn’t flatten — it accelerates.</p><h3>What This Means for Your Team Specifically</h3><p>The ROI data makes headlines, but the day-to-day impact is where teams actually feel the difference.</p><p><strong>For support agents:</strong> Instead of spending the first 20 minutes of every ticket reading through email threads and ticket history, they open a case and the AI has already summarized everything — context, what was tried, what the user actually needs. The cognitive load drops. The work gets faster and more meaningful.</p><p><strong>For IT managers:</strong> Instead of reporting on reactive metrics — how many tickets closed, average time to resolution — you’re managing toward outcomes. Uptime. User productivity. Prevention rates. The shift from SLAs to Experience Level Agreements (XLAs) isn’t just a terminology update; it’s a fundamentally better way to measure what support actually delivers.</p><p><strong>For leadership:</strong> Instead of a help desk that’s purely a cost center, you have a function that generates intelligence about your organization — where bottlenecks live, what tools are causing friction, where training gaps exist. That data has strategic value beyond support itself.</p><p><strong>For end users:</strong> They get answers immediately, often before a ticket is ever created. Their satisfaction goes up. Their productivity goes up. The persistent drag of waiting for IT is eliminated.</p><h3>The Objections Worth Taking Seriously</h3><p>It’s worth addressing the two most common concerns honestly.</p><p><strong>“Won’t this replace our support agents?”</strong></p><p>No — but it will change what they spend their time on. The agents who thrive in an AI-augmented support environment are the ones handling complex troubleshooting, nuanced escalations, and the work that genuinely requires human judgment. The repetitive L1 load that burns agents out disappears. What remains is more interesting and more impactful.</p><p>A 2026 Gartner analysis suggests organizations will replace 20–30% of agent <em>volume</em> with generative AI — meaning agents will handle fewer routine tickets, not that fewer agents will be employed. The goal is a higher ratio of meaningful work per agent, not headcount reduction.</p><p><strong>“Is our team ready for this?”</strong></p><p>This is a fair question, and the honest answer is: readiness is a spectrum. Most organizations don’t need to be fully mature before they start benefiting. The right approach is to start with clearly defined, high-volume workflows — L1 automation, intelligent routing, knowledge base maintenance — and expand from there as the system learns your environment and your team grows familiar with working alongside AI.</p><p>The teams that are <em>not</em> ready are the ones who wait until competitors have lapped them.</p><h3>The Shift That’s Already Happening</h3><p>“AI is transitioning from optional to operational,” as one 2026 industry analysis put it. The organizations that treated AI support tooling as a future-state initiative two years ago are now the ones playing catch-up.</p><p>The traditional help desk had a 30-year run. It served its purpose. But it was built for a world of lower volume, simpler systems, and more patient users. None of those conditions exist anymore.</p><p>The AI Support OS isn’t a replacement for your team. It’s what your team runs on.</p><h3>Where Meridesk Fits In</h3><p>Meridesk is built from the ground up as an AI Support OS for modern teams — not a legacy ticketing system retrofitted with AI, but a purpose-built platform designed around the principles above.</p><p>If your team is ready to move beyond the traditional help desk, <a href="https://meridesk.live/">start with a conversation at meridesk.live</a>.</p><p><em>Interested in more content like this? Follow Meridesk for weekly insights on AI-powered support, team productivity, and the future of IT operations.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ea4b131e5c0e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/meridesk/what-is-an-ai-support-os-and-why-your-team-needs-one-in-2026-ea4b131e5c0e">What Is an AI Support OS? (And Why Your Team Needs One in 2026)</a> was originally published in <a href="https://medium.com/meridesk">Meridesk</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why AI Isn’t Replacing Everyone (And Shouldn’t)]]></title>
            <link>https://medium.com/581-newsletter/why-ai-isnt-replacing-everyone-and-shouldn-t-a308d8d6c6d8?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/a308d8d6c6d8</guid>
            <category><![CDATA[anthropic-claude]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Sun, 08 Jun 2025 00:37:05 GMT</pubDate>
            <atom:updated>2025-06-08T00:37:05.289Z</atom:updated>
            <content:encoded><![CDATA[<p>AI is powerful — but it’s not a silver bullet. This article explores why AI still needs human oversight, the real limitations behind the hype, and how the future of work depends on learning with AI, not fearing it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NNVG89vGcLJ-mQes7KFshQ.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@wonderlane?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Wonderlane</a> on <a href="https://unsplash.com/photos/man-wearing-gray-long-sleeved-shirt-and-blue-denim-jeans-walking-on-beige-area-rug-FkHpejze0C4?utm_content=creditCopyText&amp;utm_medium=referral&amp;utm_source=unsplash">Unsplash</a></figcaption></figure><p>Here’s the thing… I’m not here to discredit AI. Far from it — I use GPT-4o and Claude Sonnet daily to engineer complex code, document systems better, and even explore edge-case solutions faster than I ever could alone. But despite how powerful these tools are, I don’t believe AI is ready to fully take over jobs in the market. Here’s why:</p><h3>1. Precision Matters — and Hallucinations Happen</h3><p>Even with the most advanced models, the quality of the output depends heavily on the clarity of the prompt. You can’t just “throw the problem at the AI” and expect gold. I’ve seen this firsthand. The more code I build with an agent, the more I realize how crucial it is to give it clear, focused instructions.</p><p>Large context windows help, but once the thread gets long, hallucinations creep in. Suggestions may become inconsistent or flat-out wrong — especially in technical domains. That’s not a model failure, that’s a design reality.</p><p>What’s the fix? You, the human, need to solve 85% of the problem mentally first — then use AI to accelerate, clarify, or validate parts of the solution. If you don’t understand what you’re building, relying on AI can lead to chaos: buggy systems, broken logic, and poor decisions. Unfortunately, that’s already happening with folks who treat AI like a shortcut, not a partner.</p><h3>2. The Ethics and Safeguards Aren’t There Yet</h3><p>Many people deploying AI at scale still haven’t put the right ethical checks or guardrails in place. Whether it’s data privacy, model bias, or simply transparency in decision-making, we’re still playing catch-up.</p><p>This matters because if you’re going to replace human decisions with machine-generated ones — in hiring, lending, diagnosis, or security — you’d better be able to explain how those decisions were made. Otherwise, we’re heading straight into accountability blind spots that no one wants to own.</p><h3>3. AI Makes Sense With People, Not Instead of Them</h3><p>The real power of AI isn’t in replacing jobs — it’s in enhancing people. I’ve seen teams do this brilliantly:</p><ul><li>Creating custom knowledge bases so employees don’t waste time digging for docs</li><li>Using agents as internal copilots to research, summarize, and cross-reference material</li><li>Automating repetitive admin tasks that slow down actual work</li><li>Delegating email drafts, support replies, and data analysis</li><li>Running deep, focused research without rabbit holes</li></ul><p>With multi-agent systems (MCPs) and more accessible tooling, AI is becoming part of the <em>ecosystem</em>, not the enemy.</p><h3>So What Should We Do?</h3><p>We shouldn’t panic about AI taking jobs — we should teach people how to work with it.</p><p>The future belongs to those who can:</p><ul><li>Learn with AI</li><li>Build with AI</li><li>Create with AI</li><li>Research with AI</li><li>Solve meaningful problems with AI</li></ul><p>This isn’t about being replaced. It’s about being <em>amplified</em>. The winners in this next wave of tech will be those who know how to think critically, move fast with tools, and communicate clearly — because AI is only as good as the person using it.</p><p>What’re your thoughts? Do you think filling positions with AI Agents will get us anywhere?</p><p>If you’d like to read more, please subscribe to <a href="https://581-group.com/newsletter">581 Newsletter</a>.</p><p>Thank you for reading.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a308d8d6c6d8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/581-newsletter/why-ai-isnt-replacing-everyone-and-shouldn-t-a308d8d6c6d8">Why AI Isn’t Replacing Everyone (And Shouldn’t)</a> was originally published in <a href="https://medium.com/581-newsletter">581 Newsletter</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Create A Subscription Service with Stripe in NodeJS & Express]]></title>
            <link>https://medium.com/thecodr/create-a-subscription-service-with-stripe-in-nodejs-express-7e81095b27c5?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/7e81095b27c5</guid>
            <category><![CDATA[nodejs]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[stripe]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Sun, 12 Sep 2021 18:22:11 GMT</pubDate>
            <atom:updated>2021-09-12T18:26:07.428Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Here’s how to create a backend service for recurring charges for your customers with Stripe.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fAgXMc2cx8JrjV9DpacyRA.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@alesnesetril?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Ales Nesetril</a> on <a href="https://unsplash.com/s/photos/credit-card?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>With subscriptions, customers can make recurring payments for access to a product. Subscriptions are not so different from one-time purchases — only that they are recurring and requires you to hold onto your customers’ information so you can charge them automatically.</p><p>This tutorial will guide you to setup a backend application in NodeJS and Express to setup a subscription service.</p><p>Navigate to your project folder and create a file named <strong>index.js. </strong>Once you’re done, install Express, HTTP, Body-Parser, and Stripe.</p><pre>npm i express<br>npm i http<br>npm i body-parser<br>npm i stripe</pre><p>We’ll next code for our server to use routes and body parser to handle data.</p><pre>const express = require(&quot;express&quot;);<br>const app = express();<br>const routes = require(&#39;./<em>stripe-module/route</em>&#39;);<br>const bodyParser = require(&#39;body-parser&#39;)<br>const http = require(&#39;http&#39;);</pre><pre>app.use(cors());<br>app.use(bodyParser.json())<br>app.use(bodyParser.urlencoded({ extended: true }));</pre><pre>const port = 8001;<br>app.set(&#39;port&#39;, port);<br>app.use(routes);</pre><pre>const server = http.createServer(app);<br>server.listen(port, () =&gt; {<br>console.log(`Express server listening on port ${port}`);<br>});</pre><p>To keep things organized, let’s create a folder to hold the service , controller, and route javascript files. Name it <strong><em>stripe-module.</em></strong></p><p>Create a file named <strong><em>service.js</em>. </strong>Inside we’ll write a function that sends a request to Stripe. The required parameters for this to work is a customer’s ID and an array of price items the customer is subscribing for.</p><p><strong>Note: Remember to add your stripe key for this work!</strong></p><pre>const stripe = require(&#39;stripe&#39;)(&#39;sk_test_key_here&#39;);<br>const service = {};</pre><pre>service.createSubscription = (customerId, items) =&gt; {<br>       return stripe.subscriptions.create({<br>              customer: customerId,<br>              items: items<br>       });<br>}</pre><pre>module.exports = service;</pre><p>We’ll next create a controller that acquires data from the request and sends it over to the service. Create a file named <strong><em>controller.js </em></strong>and write the following code.</p><pre>const service = require(&#39;./service&#39;);<br>const controller = {};</pre><pre>controller.createSubscription = async(req, res) =&gt; {<br>    const { customerId, items } = req.body;<br>    try {<br>      const subscription = <br>              await service.createSubscription(customerId, items);<br>      res.json({ response: subscription });<br>    } catch (error) {<br>      console.error(error);<br>      return res.status(400).json({ error: error.message });<br>    }<br>};</pre><pre>module.exports = controller;</pre><p>Lastly, let’s setup the route that trigger the subscription request. Create a file named <strong><em>router.js</em></strong> and write the following code in it.</p><pre>const router = require(&#39;express&#39;).Router();<br>const controller = require(&#39;./controller&#39;);</pre><pre>const subscriptionMiddleware = [<br>      controller.createSubscription<br>];<br>router.post(&#39;/subscribe&#39;, subscriptionMiddleware)</pre><pre>module.exports = router;</pre><p>That’s it! Run the app with <strong><em>npm run start</em></strong> or <strong><em>node index.js</em></strong> and send a request using Postman.</p><p>Remember to add the <strong>customerId</strong> and <strong>items</strong> — as an array — to the body of this post request. Below is an example of how I did that using JSON.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*tPxpcSMifLZFOIIk2R-xTA.png" /></figure><p>Thanks for reading &amp; Happy Coding! 🚀</p><ul><li><a href="https://medium.com/thecodr/how-to-save-payment-methods-on-stripe-nodejs-46fd564dac88">How to save Payment Methods on Stripe — NodeJS</a></li><li><a href="https://medium.com/thecodr/payments-on-ios-with-stripe-759507f892af">Payments on iOS with Stripe</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7e81095b27c5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/create-a-subscription-service-with-stripe-in-nodejs-express-7e81095b27c5">Create A Subscription Service with Stripe in NodeJS &amp; Express</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Apple can’t force developers to use its in-app purchasing system]]></title>
            <link>https://medium.com/thecodr/apple-cant-force-developers-to-use-its-in-app-purchasing-system-fc486ac741da?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/fc486ac741da</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[apple]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Sat, 11 Sep 2021 07:52:18 GMT</pubDate>
            <atom:updated>2021-09-11T07:52:18.297Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A federal judge in the Epic Games Inc. case issued a ruling that said Apple can no longer force developers to use its payment system.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qHpyve4R_CiyJ0McHGV_nw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@einfachlaurenz?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Laurenz Heymann</a> on <a href="https://unsplash.com/s/photos/apple?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Apple violated California’s unfair-competition law by forbidding developers from leading customers to other payment services. Apple must now allow developers to add links in their apps to other payment methods within 90 days.</p><p>According to an Apple representative, Apple has the legal option to seek to block the order before the 90-day period ends. It wasn’t all bad for Apple since the court did not find that the company is an antitrust monopolist in the submarket for mobile gaming transactions. The court however still finds that Apple’s actions in imposing anti-steering restrictions is anti-competitive.</p><p>Katherine Adams, Apple’s senior vice president, said in a conference call with reporters that Apple’s App Store business has been validated and that the company is confirmed not to be a monopolist under federal or state law.</p><p>Jessica Melugin, director of the Competitive Enterprise Institute’s Center on Technology and Innovation, said that the company will <strong>mostly</strong> continue to operate the App Store by their rules — Apple cannot be forced to allow external payment systems from developers.</p><p>Thanks for reading! Stay tuned for more.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fc486ac741da" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/apple-cant-force-developers-to-use-its-in-app-purchasing-system-fc486ac741da">Apple can’t force developers to use its in-app purchasing system</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to save Payment Methods on Stripe — NodeJS]]></title>
            <link>https://medium.com/thecodr/how-to-save-payment-methods-on-stripe-nodejs-46fd564dac88?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/46fd564dac88</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[nodejs]]></category>
            <category><![CDATA[javascript]]></category>
            <category><![CDATA[stripe]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Sat, 11 Sep 2021 06:23:32 GMT</pubDate>
            <atom:updated>2021-09-11T06:23:32.512Z</atom:updated>
            <content:encoded><![CDATA[<h3>How to save Payment Methods on Stripe — NodeJS</h3><p>Stripe is capable of storing payment methods for you. This is more secure than storing this kind of information on your own database.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0536-x-65MHWE2asStXP4Q.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@cardmapr?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">CardMapr</a> on <a href="https://unsplash.com/s/photos/payment?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>This is a brief tutorial to help you save a payment method and link it to a registered customer.</p><h3><strong>Installing Stripe</strong></h3><p>We’ll start by creating a js file, name it<strong> index.js. </strong>Install Stripe by running:</p><pre>npm i stripe</pre><h3><strong>Instantiate Stripe</strong></h3><p>Inside<strong> index.js</strong>, we’ll instantiate Stripe with our test secret keys</p><pre>const stripe = require(&#39;stripe&#39;)(&#39;sk_test_key_here&#39;);</pre><h3>Save A Payment Method</h3><p>We’ll next code to send a request with the card details as an object.</p><pre>const paymentMethod = await stripe.paymentMethods.create({<br>  type: &#39;card&#39;,<br>  card: {<br>    number: &#39;4242424242424242&#39;,<br>    exp_month: 9,<br>    exp_year: 2022,<br>    cvc: &#39;314&#39;,<br>  },<br>});</pre><pre>console.log(paymentMethod)</pre><p>Test the app by running it:</p><pre>node index.js</pre><p>You should the see a response with an “ID” property. The value of this ID is a unique identifier for this payment method.</p><p>Please note that the card is saved on your Stripe account but it is <strong>not</strong> associated with any customer — we’ll need to attach it.</p><h3>Attach A Saved Payment Method to a User</h3><p>You might prefer to allow users to re-use their payment methods for future transactions. Here’s how to link a payment method to a customer.</p><p>For this function, you’ll need to provide a <strong>Payment Method ID</strong> and a <strong>Customer ID.</strong></p><pre>const paymentMethod = await stripe.paymentMethods.attach(<br>  paymentId,<br>  {customer: customerId}<br>);</pre><p>You’ll need to re-write your code to pass these IDs as parameters. Here’s how I would do it:</p><pre>attachPaymentMethod = (customerID, paymentMethodID) =&gt; {<br>      return stripe.paymentMethods.attach(<br>             customerID,<br>             {customer: paymentMethodID}<br>);</pre><p>You can finally call this function right after the payment method is added.</p><pre>attachPaymentMethod(&#39;customer-id-goes-here&#39;, &#39;pm-id-goes-here&#39;)</pre><p>Thanks for reading &amp; Happy Coding!</p><p><a href="https://medium.com/thecodr/payments-on-ios-with-stripe-759507f892af">Payments on iOS with Stripe</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=46fd564dac88" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/how-to-save-payment-methods-on-stripe-nodejs-46fd564dac88">How to save Payment Methods on Stripe — NodeJS</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Selenium for a Web Scraper]]></title>
            <link>https://medium.com/thecodr/using-selenium-for-a-web-scraper-dccac1b95d1f?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/dccac1b95d1f</guid>
            <category><![CDATA[python]]></category>
            <category><![CDATA[sofware-development]]></category>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[sofware-engineering]]></category>
            <category><![CDATA[programming]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Fri, 10 Sep 2021 22:23:26 GMT</pubDate>
            <atom:updated>2021-09-10T22:23:26.788Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Here’s all the things you can do with </em><a href="https://pypi.org/project/selenium/"><em>Selenium</em></a><em> on your Web Scraper.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wc3MgXJ00mrdGO_-8vJXTw.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@euwars?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Farzad Nazifi</a> on <a href="https://unsplash.com/s/photos/website?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><h3>Finding Elements</h3><p>This is the most common use for <a href="https://pypi.org/project/selenium/">Selenium</a>. You can find specific data on a website and the extract it.</p><p>Selenium can find elements using:</p><ul><li>Tag name</li><li>IDs</li><li>Class name</li><li>X Path</li><li>CSS Selectors</li></ul><p>Here’s what it would look like on code for each case:</p><pre>by_class = driver.find_element_by_class_name(&#39;class name&#39;)<br>by_id = driver.find_element_by_id(&#39;id&#39;)<br>by_tag = driver.find_elements_by_tag_name(&#39;tag name&#39;)<br>by_xpath = driver.find_element_by_xpath(&#39;x-path&#39;)<br>by_css_selector = driver.find_element_by_css_selector(&#39;css selector&#39;)</pre><h3>Running Javascript</h3><p>You might need to make some actions on a web page, for example, scrolling. Here’s an example of how you would do that:</p><pre>js = &quot;window.scrollBy(0,1000);&quot;<br>driver.execute_script(js)</pre><h3><strong>Screenshots</strong></h3><p>You can also take a screenshot of a web page:</p><pre>driver.save_screenshot(&#39;webpage.png&#39;)</pre><h3>Waiting for elements to show up</h3><p>Web pages might take a while to load up. Sometimes it’s because of slow network speeds, sometimes the web application could be waiting for an API response.</p><p>In this case, we’d use WebDriverWait.</p><pre>try:<br>    element = WebDriverWait(driver, 3).until(<br>        EC.presence_of_element_located((By.ID, &quot;elementID&quot;))<br>    )<br>finally:<br>    driver.quit()</pre><h3>Proxy with Selenim Wire</h3><p>You can’t solely use Selenium for proxy authentication, you’ll need to install the <a href="https://pypi.org/project/selenium-wire/"><strong>Selenium Wire</strong></a><strong> </strong>package.</p><p><a href="https://pypi.org/project/selenium-wire/"><strong>Selenium Wire</strong></a> extends Selenium’s Python bindings to give you access to the underlying requests made by the browser.</p><p>I hope this was helpful! If you’d like to <a href="https://medium.com/thecodr/make-a-web-scraper-with-python-and-beautiful-soup-8d13108e237f"><strong>make a web scraper with Selenium &amp; Beautiful Soup</strong></a>, read this post:</p><p><a href="https://medium.com/thecodr/make-a-web-scraper-with-python-and-beautiful-soup-8d13108e237f">Make a Web Scraper with Python and Beautiful Soup</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dccac1b95d1f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/using-selenium-for-a-web-scraper-dccac1b95d1f">Using Selenium for a Web Scraper</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Make a Web Scraper with Python and Beautiful Soup]]></title>
            <link>https://medium.com/thecodr/make-a-web-scraper-with-python-and-beautiful-soup-8d13108e237f?source=rss-9544aee41bf6------2</link>
            <guid isPermaLink="false">https://medium.com/p/8d13108e237f</guid>
            <category><![CDATA[web-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Daniel Kioko]]></dc:creator>
            <pubDate>Fri, 10 Sep 2021 21:36:39 GMT</pubDate>
            <atom:updated>2021-09-10T22:24:56.616Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Here’s how to make your own web scraper in less than 10 minutes</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KTgOi6dScqsrKOz4dXubbg.jpeg" /><figcaption>Photo by <a href="https://unsplash.com/@filisantillan?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Filiberto Santillán</a> on <a href="https://unsplash.com/s/photos/data?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></figcaption></figure><p>Web scraping is the process of automatically collecting information from web pages. This tutorial will help you build a scraper from scratch using python.</p><h3>Setup your environment</h3><p>Make sure you have the latest version of python. Run this in terminal to check what version you’re on:</p><pre>python --version</pre><p>Create your project folder and a python file, name it <strong><em>app.py</em></strong></p><p>We’ll next create a virtual environment named <strong>env</strong>, just run:</p><pre>python -m venv env</pre><p>To start the environment, run:</p><pre>source env/bin/activate</pre><h3>Libraries</h3><p>To run this scraper, we’ll need to add a few libraries to interact with the browser and collect the information we need.</p><p><strong>Selenium</strong> — this is a web framework that allows you to execute cross-browser tests. To install selenium, run:</p><pre>pip install selenium</pre><p><strong>Web Driver Manager </strong>— this simplifies management of binary drivers for different browsers, instead of installing the exact driver, the web driver manager does it for you. To install webdriver-manager, run:</p><pre>pip install webdriver-manager</pre><p><strong>Beautiful Soup — t</strong>his is a package for parsing HTML and XML documents. To install it run:</p><pre>pip install beautifulsoup4</pre><h3>Let’s Code</h3><p>We’ll first import our libraries:</p><pre>from selenium import webdriver<br>from webdriver_manager.chrome import ChromeDriverManager<br>from bs4 import BeautifulSoup</pre><p>Then we’ll setup the driver, well use wiki how for this tutorial:</p><pre>url = &#39;<a href="https://www.wikihow.com/Main-Page">https://www.wikihow.com/Main-Page</a>&#39;<br>browser = webdriver.Chrome(ChromeDriverManager().install())<br>browser.get(url)</pre><p>Then onto the main function, we’ll need to define the element using an x-path. To get one, inspect the element on a page, then copy <strong>specifically</strong> the x-path.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/377/1*ZWKCN4D3pftge_WCmRDUvQ.png" /><figcaption>Copying the x-path (Safari)</figcaption></figure><p>We’ll code to collect the data by the x-path, parse it and print it out. I copied the x-path of the wiki-how’s homepage title for this example:</p><pre>try:<br>    element = browser.find_element_by_xpath(&#39;//*[<a href="http://twitter.com/id">@id</a>=&quot;hp_container_inner&quot;]/h1&#39;)<br>    element_data = element.get_attribute(&#39;innerHTML&#39;)<br>    parsed_info = BeautifulSoup(element_data, &#39;html.parser&#39;)<br>    print(parsed_info)<br>    browser.quit()<br>catch:<br>    break</pre><p>Now run the app:</p><pre>python app.py</pre><p>Congrats! You’re running your first web scraper. Let me know how it goes in the comments.</p><p>Stay tuned for more &amp; happy coding!</p><p><a href="https://medium.com/thecodr/using-selenium-for-a-web-scraper-dccac1b95d1f">Using Selenium for a Web Scraper</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8d13108e237f" width="1" height="1" alt=""><hr><p><a href="https://medium.com/thecodr/make-a-web-scraper-with-python-and-beautiful-soup-8d13108e237f">Make a Web Scraper with Python and Beautiful Soup</a> was originally published in <a href="https://medium.com/thecodr">TheCodr</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>