{"id":264139,"date":"2026-03-06T12:40:45","date_gmt":"2026-03-06T12:40:45","guid":{"rendered":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/?page_id=264139"},"modified":"2026-03-24T20:18:43","modified_gmt":"2026-03-24T20:18:43","slug":"ai-journal-2026","status":"publish","type":"page","link":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/ai-diaries\/ai-journal-2026\/","title":{"rendered":"AI Journal \u2014 2026"},"content":{"rendered":"\n[et_pb_section fb_built=&#8221;1&#8243; admin_label=&#8221;Section (Some CSS &#8211; Lineheight, FontSize)&#8221; module_id=&#8221;Section_AIJournal2026&#8243; module_class=&#8221;page-content-re-fouc&#8221; _builder_version=&#8221;4.27.5&#8243; _module_preset=&#8221;default&#8221; module_alignment=&#8221;center&#8221; custom_margin=&#8221;||||false|false&#8221; custom_padding=&#8221;0px||0px||false|false&#8221; custom_css_free_form=&#8221;selector p, li {||  line-height:1.25 !important;||  font-size: 16px;||}||||p.small {||  line-height:1.25 !important;||  font-size: 14px !important;||}||||||||&#8221; global_colors_info=&#8221;{}&#8221;][et_pb_row admin_label=&#8221;Row&#8221; module_id=&#8221;Row_AIJournal2026&#8243; _builder_version=&#8221;4.27.5&#8243; _module_preset=&#8221;default&#8221; max_width=&#8221;1440px&#8221; custom_margin=&#8221;|0px||0px|false|false&#8221; custom_padding=&#8221;||||false|false&#8221; global_colors_info=&#8221;{}&#8221;][et_pb_column type=&#8221;4_4&#8243; _builder_version=&#8221;4.27.5&#8243; _module_preset=&#8221;default&#8221; custom_padding=&#8221;||||false|false&#8221; custom_css_main_element=&#8221;width:100%;&#8221; custom_css_free_form=&#8221;||&#8221; global_colors_info=&#8221;{}&#8221;][et_pb_code admin_label=&#8221;Code: Title and Intro&#8221; _builder_version=&#8221;4.27.5&#8243; _module_preset=&#8221;default&#8221; hover_enabled=&#8221;0&#8243; global_colors_info=&#8221;{}&#8221; locked=&#8221;off&#8221; sticky_enabled=&#8221;0&#8243;]<div class=\"back-link back-link-hover-blue back-link-offset-up back-link-smaller\"><a href=\"https:\/\/daviddurlach.com\/writingsPP\/droplets\/ai-diaries\/\">( &larr; Back to AI Diaries)<\/a><\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->  <div class=\"aidiary-posts-container\"><!-- [et_pb_line_break_holder] -->    <article class=\"aidiary-blog-post aidiary-post-full\"><!-- [et_pb_line_break_holder] -->      <div class=\"aidiary-post-content\"><!-- [et_pb_line_break_holder] -->        <!-- Date Header --><!-- [et_pb_line_break_holder] -->        <!--<!-- [et_pb_line_break_holder] -->        <div class=\"aidiary-date-header\"><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-date-badge\">March 6, 2026<\/div><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-last-updated-date\">Last updated March 20, 2026<\/div><!-- [et_pb_line_break_holder] -->        <\/div><!-- [et_pb_line_break_holder] -->        &#8211;><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->        <!-- Title --><!-- [et_pb_line_break_holder] -->         <div class=\"aidiary-post-title aidiary-post-title-frame lh-sm\"><!-- [et_pb_line_break_holder] -->          <h2 id=\"its-just-math\">2026 Journal Of AI Tidbits<\/h2><!-- [et_pb_line_break_holder] -->            <!-- Subtitle - two separate elements for spacing control --><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-subtitle\"><!-- [et_pb_line_break_holder] -->              <pee class=\"aidiary-subtitle-1\">For items not quite meeting the \u201cfull separate post page\u201d threshold.<\/pee><!-- [et_pb_line_break_holder] -->              <!-- <pee class=\"aidiary-subtitle-2\">.<\/pee> --><!-- [et_pb_line_break_holder] -->            <\/div><!-- [et_pb_line_break_holder] -->         <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->        <\/article><!-- [et_pb_line_break_holder] -->  <\/div><!-- [et_pb_line_break_holder] -->[\/et_pb_code][et_pb_code admin_label=&#8221;Code: All&#8221; module_id=&#8221;Code__All&#8221; _builder_version=&#8221;4.27.5&#8243; _module_preset=&#8221;default&#8221; global_colors_info=&#8221;{}&#8221;]<!-- MC AIDiariesJournal2026-All --><!-- wp-remote-header<!-- [et_pb_line_break_holder] -->  post-id:         264139<!-- [et_pb_line_break_holder] -->  post-title:      AI Journal \u2014 2026<!-- [et_pb_line_break_holder] -->  block-id:        AIDiariesJournal2026-All<!-- [et_pb_line_break_holder] -->  site:            https:\/\/daviddurlach.com\/writingsPP\/droplets<!-- [et_pb_line_break_holder] -->  context:         AI Journal 2026 \u2014 all entries<!-- [et_pb_line_break_holder] -->  pulled:          2026-03-20T11:32:45<!-- [et_pb_line_break_holder] -->  remote-modified: 2026-03-20T14:43:16<!-- [et_pb_line_break_holder] -->&#8211;><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] --><div class=\"blog-singlepost-multicolumn-topfullwidth\"><!-- [et_pb_line_break_holder] -->  <div class=\"multicolumnpost-content\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->    <!-- *********************************** --><!-- [et_pb_line_break_holder] -->    <div class=\"aidiary-posts-container\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->      <!-- ========================================<!-- [et_pb_line_break_holder] -->             EXAMPLE 1: Post WITH Image<!-- [et_pb_line_break_holder] -->             ======================================== &#8211;><!-- [et_pb_line_break_holder] -->      <article class=\"aidiary-blog-post\"><!-- [et_pb_line_break_holder] -->        <!-- Featured Image (Optional - remove section if no image) --><!-- [et_pb_line_break_holder] -->        <!-- <div class=\"aidiary-featured-image-wrapper\"><!-- [et_pb_line_break_holder] -->                <a href=\"#\" class=\"aidiary-featured-image-link\"><!-- [et_pb_line_break_holder] -->                    <img decoding=\"async\" src=\"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-content\/uploads\/2025\/10\/AIDiariesThreeStrip2x-1080x364.webp\" <!-- [et_pb_line_break_holder] -->                         alt=&#8221;AI Diaries&#8221; <!-- [et_pb_line_break_holder] -->                         class=&#8221;aidiary-featured-image&#8221;><!-- [et_pb_line_break_holder] -->                <\/a><!-- [et_pb_line_break_holder] -->            <\/div><!-- [et_pb_line_break_holder] -->            &#8211;><!-- [et_pb_line_break_holder] -->        <!-- Post Content --><!-- [et_pb_line_break_holder] -->        <div class=\"aidiary-post-content\"><!-- [et_pb_line_break_holder] -->          <!-- Date Header --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-date-header\"><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-date-badge\">Feb 7, 2026<\/div><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-last-updated-date\">Last updated March 17, 2026<\/div><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Title --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-post-title aidiary-post-title-frame lh-sm\"><!-- [et_pb_line_break_holder] -->            <h2>AI Impact On Coder Skills<\/h2><!-- [et_pb_line_break_holder] -->            <!-- <pee class=\"aidiary-tagline\">xxxx.<\/pee> --><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Excerpt\/Description --><!-- [et_pb_line_break_holder] -->          <div class=\"mt-0 mb-0 aidiary-post-excerpt\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->            <pee><span class = \"ai-chat-reference\">Quick Feb chat with Claude: Topic \u2014 Jan 2026 Anthropic paper \u201c<em><a class=\"link-against-light-black-underlined-color-inherited\" href=\"https:\/\/www.anthropic.com\/research\/AI-assistance-coding-skills\">How AI assistance impacts the formation of coding skills<\/a><\/em>\u201d<\/span> by <a class=\"link-against-light-black-underlined-color-inherited\" href=\"https:\/\/heyyjudes.github.io\">Judy Hanwen Shen<\/a> and <a class=\"link-against-light-black-underlined-color-inherited\" href=\"https:\/\/www.alextamkin.com\">Alex Tamkin<\/a>. I suggest \u201cstepping up a level\u201d when considering what kinds of skills (will) matter.<\/pee><!-- [et_pb_line_break_holder] -->            <pee>(Paper originally forwarded to me by a friend.)<\/pee><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Editorial-style accordion wrapper for post content --><!-- [et_pb_line_break_holder] -->          <div class=\"accordion-container-ai-impact-parent\"><!-- [et_pb_line_break_holder] -->            <div class=\"ac ac-ed-parent\"><!-- [et_pb_line_break_holder] -->              <button type=\"button\" class=\"ac-trigger ac-trigger-editorial\"><!-- [et_pb_line_break_holder] -->                <span class=\"ac-ed-accent\"><\/span><!-- [et_pb_line_break_holder] -->                <span class=\"ac-ed-body\"><!-- [et_pb_line_break_holder] -->                  <span class=\"ac-ed-icon\"><span><\/span><span><\/span><span><\/span><\/span><!-- [et_pb_line_break_holder] -->                  <span class=\"ac-ed-text\"><!-- [et_pb_line_break_holder] -->                    <span class=\"ac-ed-title\">Continue reading<\/span><!-- [et_pb_line_break_holder] -->                    <span class=\"ac-ed-subtitle\">~10 min read<\/span><!-- [et_pb_line_break_holder] -->                  <\/span><!-- [et_pb_line_break_holder] -->                  <span class=\"ac-ed-chevron\"><!-- [et_pb_line_break_holder] -->                    <svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2.5\"<!-- [et_pb_line_break_holder] -->                         stroke-linecap=&#8221;round&#8221; stroke-linejoin=&#8221;round&#8221;><!-- [et_pb_line_break_holder] -->                      <polyline points=\"6 9 12 15 18 9\"\/><!-- [et_pb_line_break_holder] -->                    <\/svg><!-- [et_pb_line_break_holder] -->                  <\/span><!-- [et_pb_line_break_holder] -->                <\/span><!-- [et_pb_line_break_holder] -->              <\/button><!-- [et_pb_line_break_holder] -->              <div class=\"ac-panel\" id=\"ai-impact-parent-panel\"><!-- [et_pb_line_break_holder] -->                <div class=\"ac-body-background ac-body-background-transparent ac-body-border-none\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->                  <pee class=\"chat-id\"><strong>Me:<\/strong><\/pee><!-- [et_pb_line_break_holder] -->                  <div class=\"chat-human\"><!-- [et_pb_line_break_holder] -->                    <div class=\"render-markdown\"><!-- [et_pb_line_break_holder] -->I use AI in coding contexts for hours daily. It is making possible things I would otherwise most certainly not be able to accomplish in my lifetime. I am creating complex environments written in languages, and using frameworks, about which I know little if anything. That said, I have decades of technical experience broadly, with programming being part of that. So I fall at the 80% speedup for many tasks &#8212; and sometimes even more like 99% speedup. (That said, I have wasted days on simple problems where AI gets stuck, going in circles etc.)<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->When I read the article, I had the sense that &#8220;abstraction levels\/barriers&#8221; were being implicitly traversed, and that that was causing &#8220;trouble&#8221;. More specifically:<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->a) In the way I code with AI, if I ever have to look at actual code at all, I define that as a failure.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->b) I read an article by well-known AI author Ethan Mollick about AI functioning as &#8220;Wizard&#8221; &#8212; and another regarding getting things done functionally, regardless of the path. What I want you to notice is the abstraction levels \/ information-hiding. Both discuss &#8220;wrapping complexity inside a container&#8221;.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->c) This is what I thought was &#8220;missing&#8221; in the recent AI-assistance-coding-skills article. It felt to me a bit like saying &#8220;Writing C++ code fails to teach you adequately about compiler design or give you the ability to debug assembler&#8221; which is true, but not, in my view, a criticism.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->d) I would like to take the perspective that in working with AI, the goal is to learn how to effectively talk to it about coding, and to create good testing of final output for proper operation, but *not* to even look &#8220;under the hood&#8221; at the code itself.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Can you comment on my thinking, make a case for it, include lots of analogies, be honest about its limitations, discuss where you think the future is heading, etc.<!-- [et_pb_line_break_holder] -->                    <\/div><!-- [et_pb_line_break_holder] -->                  <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->                  <pee class=\"chat-id\"><strong>Claude:<\/strong><\/pee><!-- [et_pb_line_break_holder] -->                  <div class=\"chat-ai\"><!-- [et_pb_line_break_holder] -->                    <div class=\"render-markdown\"><!-- [et_pb_line_break_holder] -->## The Case for Your Position: Abstraction as the Point, Not the Problem<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your intuition is, I think, fundamentally correct&#8211;and the Anthropic study, while methodologically sound, may be asking a question that&#8217;s becoming less relevant by the month.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your C++ \/ assembly comparison isn&#8217;t just illustrative&#8211;it&#8217;s structurally identical to what&#8217;s happening. **Every one of those previous abstraction jumps was initially framed as &#8220;losing important skills.&#8221;** And every time, the correct response turned out to be: *those skills became specialized knowledge for a shrinking subset of cases, while the new abstraction level became the productive frontier.*<!-- [et_pb_line_break_holder] -->                    <\/div><!-- [et_pb_line_break_holder] -->                    <div class=\"mt-2 accordion-container-abstraction-detail\"><!-- [et_pb_line_break_holder] -->                      <div class=\"ac ac-ed-section-card\"><!-- [et_pb_line_break_holder] -->                        <button type=\"button\" class=\"ac-trigger ac-ed-section-trigger\"><!-- [et_pb_line_break_holder] -->                          <span class=\"ac-ed-accent\"><\/span><!-- [et_pb_line_break_holder] -->                          <span class=\"ac-ed-section-body\"><!-- [et_pb_line_break_holder] -->                            <span class=\"ac-ed-icon\"><span><\/span><span><\/span><span><\/span><\/span><!-- [et_pb_line_break_holder] -->                            <span class=\"ac-ed-section-text\"><!-- [et_pb_line_break_holder] -->                              <span class=\"ac-ed-section-title\">Claude continues&#8230;<\/span><!-- [et_pb_line_break_holder] -->                              <span class=\"ac-ed-section-reading-time\" data-auto-populate><\/span><!-- [et_pb_line_break_holder] -->                            <\/span><!-- [et_pb_line_break_holder] -->                            <span class=\"ac-ed-section-indicators\"><!-- [et_pb_line_break_holder] -->                              <span class=\"ac-ed-section-node\"><\/span><!-- [et_pb_line_break_holder] -->                              <span class=\"ac-ed-chevron\"><svg viewBox=\"0 0 24 24\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2.5\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><polyline points=\"6 9 12 15 18 9\"\/><\/svg><\/span><!-- [et_pb_line_break_holder] -->                            <\/span><!-- [et_pb_line_break_holder] -->                          <\/span><!-- [et_pb_line_break_holder] -->                        <\/button><!-- [et_pb_line_break_holder] -->                        <div class=\"ac-panel\" id=\"abstraction-detail-panel\"><!-- [et_pb_line_break_holder] -->                          <div class=\"ac-body-background ac-body-background-transparent ac-body-border-none ac-ed-section-panel-body\"><!-- [et_pb_line_break_holder] -->                            <div class=\"render-markdown\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### The Study Measures Skill Formation at the Wrong Level<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Here&#8217;s the crux: The study asked, &#8220;Did participants learn Trio?&#8221;<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->But what if the *actual skill being formed* when working with AI is:<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->1. **Prompt specification**: Learning to articulate intent precisely<!-- [et_pb_line_break_holder] -->2. **Output validation**: Learning to test whether the result actually works<!-- [et_pb_line_break_holder] -->3. **Failure recognition**: Learning to detect when AI is stuck or hallucinating<!-- [et_pb_line_break_holder] -->4. **Architectural thinking**: Learning to decompose problems into AI-delegatable chunks<!-- [et_pb_line_break_holder] -->5. **Iteration strategy**: Learning when to push back, rephrase, or abandon<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->These are *different skills* than &#8220;understanding the Trio library,&#8221; but they are not *lesser* skills. They&#8217;re the skills of someone operating at the next abstraction level up.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->The study&#8217;s design is like testing whether calculator users understand long division. They might not&#8211;but they might also be solving differential equations that the long-division experts never reach.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### Mollick&#8217;s &#8220;Wizard&#8221; Framing Supports This<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Mollick&#8217;s insight is that we&#8217;re moving from **collaborators on process** to **curators of output**. The relevant competencies shift:<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->| Old Model (Co-Intelligence) | New Model (Wizard) |<!-- [et_pb_line_break_holder] -->|&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;-|&#8212;&#8212;&#8212;&#8212;&#8212;&#8212;-|<!-- [et_pb_line_break_holder] -->| Understand the steps | Define success criteria |<!-- [et_pb_line_break_holder] -->| Guide the process | Verify the result |<!-- [et_pb_line_break_holder] -->| Debug the internals | Recognize failure modes |<!-- [et_pb_line_break_holder] -->| Know the tools deeply | Know what to ask for |<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your statement&#8211;&#8220;If I ever have to look at actual code, I define that as a failure&#8221;&#8211;is precisely the wizard-model operating principle. You&#8217;re not *avoiding* competence; you&#8217;re *relocating* it to specification and validation.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### The Bitter Lesson Applied to Human Skills<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->The Bitter Lesson says: encoding human expertise into AI systems tends to lose to throwing compute at the outcome.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->But there&#8217;s a corollary for humans: **investing human effort in understanding AI internals may lose to investing human effort in outcome specification and testing.**<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### The Information-Hiding Principle Is Computer Science, Not Laziness<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->You&#8217;re invoking one of the foundational principles of software engineering: **information hiding** (Parnas, 1972). The entire point of good abstraction is that you *shouldn&#8217;t need to know* what&#8217;s inside the module. That&#8217;s not a bug; it&#8217;s the defining feature.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### Analogies That Support Your Position<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->**1. The Taxi Driver Analogy**<!-- [et_pb_line_break_holder] -->When GPS navigation arrived, London taxi drivers had spent years acquiring &#8220;The Knowledge&#8221;&#8211;memorizing 25,000 streets and 20,000 landmarks. GPS made this less necessary. The skill that *mattered* became: knowing where you actually want to go, recognizing when GPS is routing you badly, and knowing when to override.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->**2. The Calculator Analogy**<!-- [et_pb_line_break_holder] -->We don&#8217;t test engineers on manual logarithm table lookups. The skill shifted from *performing calculation* to *knowing what to calculate* and *recognizing wrong answers*.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->**3. The Radiology Analogy**<!-- [et_pb_line_break_holder] -->Radiologists who insist on reading every image without AI are like programmers who insist on writing assembly. The emerging skill is: knowing when to trust the AI, knowing what cases require human review, and integrating imaging into clinical judgment. Not *replacing* judgment&#8211;*relocating* it.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->&#8212;<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->## Honest Limitations of This Position<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### 1. The Abstraction Layer Is Still Leaky<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Unlike compilers (which are mature and predictable), AI is:<!-- [et_pb_line_break_holder] -->&#8211; **Stochastic**: Same input can give different outputs<!-- [et_pb_line_break_holder] -->&#8211; **Opaque in failure modes**: When it fails, it often fails *confidently*<!-- [et_pb_line_break_holder] -->&#8211; **Inconsistently capable**: The &#8220;jagged frontier&#8221; means you can&#8217;t reliably predict what it can and can&#8217;t do<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your experience of &#8220;wasting days on simple problems where AI gets stuck, going in circles&#8221; is precisely this leakiness.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### 2. Testing Is Harder Than It Looks<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your approach requires robust testing of outputs. But:<!-- [et_pb_line_break_holder] -->&#8211; Many real-world systems have subtle correctness requirements<!-- [et_pb_line_break_holder] -->&#8211; Security vulnerabilities are often invisible to functional tests<!-- [et_pb_line_break_holder] -->&#8211; Performance characteristics may only emerge at scale<!-- [et_pb_line_break_holder] -->&#8211; Edge cases are what AI is worst at anticipating<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### 3. The Abstraction Level May Not Be Stable Yet<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Compiler abstractions stabilized over decades. The &#8220;AI as coding wizard&#8221; abstraction is months old. You may be learning skills that are themselves transient, as the AI interface changes rapidly.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### 4. The Study&#8217;s Concern About Oversight Is Real<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->The Anthropic study explicitly frames this around *AI safety*&#8211;the need for humans to supervise AI-generated code in high-stakes contexts. If no one understands the code, who catches the bug that causes the catastrophe?<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->&#8212;<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->## Where This Is Heading<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### Short-term (1-2 years)<!-- [et_pb_line_break_holder] -->The hybrid model dominates. People who can *both* work at the wizard level *and* drop down when needed will outperform either pure approach.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### Medium-term (3-5 years)<!-- [et_pb_line_break_holder] -->AI verification tools emerge. We&#8217;ll have AI systems that check other AI systems&#8217; code&#8211;not just for functional correctness but for security, performance, and maintainability.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### Long-term (5+ years)<!-- [et_pb_line_break_holder] -->If the trajectory continues, understanding code becomes *specialist knowledge* like understanding assembly is today&#8211;relevant for a shrinking set of cases, but not the general requirement for being &#8220;a developer.&#8221;<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->### The Key Uncertainty<!-- [et_pb_line_break_holder] -->Will AI get reliable enough, fast enough, that the leaky abstraction problem goes away? If yes, your position wins outright. If AI plateaus with meaningful error rates, the ability to debug will remain valuable longer.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->&#8212;<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->## My Honest Assessment<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Your approach is:<!-- [et_pb_line_break_holder] -->&#8211; **Pragmatically optimal** for your situation (experienced, working on projects that can tolerate iteration, not safety-critical)<!-- [et_pb_line_break_holder] -->&#8211; **Directionally correct** about where the field is heading<!-- [et_pb_line_break_holder] -->&#8211; **Intellectually coherent** as an abstraction-level argument<!-- [et_pb_line_break_holder] -->&#8211; **Vulnerable** to the unpredictability of current AI and the absence of mature verification tools<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] --><!\u2013- [et_pb_br_holder] -\u2013><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->The Anthropic study is:<!-- [et_pb_line_break_holder] -->&#8211; **Methodologically sound** for what it measures<!-- [et_pb_line_break_holder] -->&#8211; **Measuring something real** (people do learn less of the underlying system)<!-- [et_pb_line_break_holder] -->&#8211; **Potentially asking the wrong question** if the right skill to form is &#8220;working with AI&#8221; rather than &#8220;understanding Trio&#8221;<!-- [et_pb_line_break_holder] -->&#8211; **Valuable as a warning** for contexts where oversight matters<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->You&#8217;re not wrong to see this as analogous to &#8220;C++ doesn&#8217;t teach you assembler&#8221;&#8211;that framing is apt. The question is whether we&#8217;re at the point in history where assembler knowledge is truly optional, or whether we&#8217;re at the point where it&#8217;s *usually* optional but *critically* necessary in unpredictable moments.<!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->Given your background and your projects, betting on the former seems reasonable. Just keep those tests robust.<!-- [et_pb_line_break_holder] -->                            <\/div><!-- [et_pb_line_break_holder] -->                          <\/div><!-- [et_pb_line_break_holder] -->                        <\/div><!-- [et_pb_line_break_holder] -->                      <\/div><!-- [et_pb_line_break_holder] -->                    <\/div><!-- [et_pb_line_break_holder] -->                  <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->                <\/div><!-- [et_pb_line_break_holder] -->              <\/div><!-- [et_pb_line_break_holder] -->            <\/div><!-- [et_pb_line_break_holder] -->          <\/div> <!-- end accordion-container-ai-impact-parent --><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Footer with Read More and Meta Button --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-post-footer\"><!-- [et_pb_line_break_holder] -->            <!-- <a href=\"#\" class=\"aidiary-read-more-link\">More<\/a> --><!-- [et_pb_line_break_holder] -->            <pee><\/pee><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-meta-button\">\ud83d\udcad\ud83d\udcad\ud83d\udd28\ud83d\udd28\ud83d\udd28<\/div><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] -->        <\/div><!-- [et_pb_line_break_holder] -->      <\/article><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->      <!-- *********************************** --><!-- [et_pb_line_break_holder] -->      <!-- ========================================<!-- [et_pb_line_break_holder] -->             Word Documents Entry<!-- [et_pb_line_break_holder] -->             ======================================== &#8211;><!-- [et_pb_line_break_holder] -->      <article class=\"aidiary-blog-post\"><!-- [et_pb_line_break_holder] -->        <!-- Post Content --><!-- [et_pb_line_break_holder] -->        <div class=\"aidiary-post-content\"><!-- [et_pb_line_break_holder] -->          <!-- Date Header --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-date-header\"><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-date-badge\">Oct 14, 2025<\/div><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-last-updated-date\">Last updated March 20, 2026<\/div><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Title --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-post-title aidiary-post-title-frame lh-sm\" id=\"WorkingWithAIandWordDocs\"><!-- [et_pb_line_break_holder] -->            <h2><span class=\"text-secondary\" style=\"font-size:1rem;\">(<span data-tippy-content=\"Because the AIs themselves \u2014 certainly Claude anyway \u2014 have so improved as of March 2026.\" class=\"blog-tooltip-enabled-singlepost dd-tooltip-against-black\" data-tippy-theme=\"blog-popup-theme-singlepost\" aria-expanded=\"false\">May no longer bother.<\/span>)<\/span> AI and Customized Word Documents<\/h2><!-- [et_pb_line_break_holder] -->            <pee class=\"aidiary-tagline\">The formatting problem (and solution).<\/pee><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Excerpt\/Description --><!-- [et_pb_line_break_holder] -->          <div class=\"mt-0 mb-0 aidiary-post-excerpt\"><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->            <pee><span class = \"ai-chat-reference\">AIs are typically fantastic at scanning painfully long and complex email trails<\/span>, extracting out relevant stakeholder names, project details, final decisions on various points, etc. <span class = \"ai-chat-reference\">However, when it comes to creating new Word documents based on those details<\/span>, matched in visual appearance to an existing reference Word document, <span data-tippy-content=\"At least they were for the frontier models I tried \u2014 ChatGPT\/OpenAI, Claude\/Anthropic, and Gemini\/Google \u2014 Oct. 2025. UPDATE March 2026: Claude at least is much better \u2014 others may be improved as well.\" class=\"blog-tooltip-enabled-singlepost\" data-tippy-theme=\"blog-popup-theme-singlepost\" aria-expanded=\"false\"><span class = \"ai-chat-reference\">they have <em>historically<\/em> been pretty terrible<\/span>.<\/pee><!-- [et_pb_line_break_holder] -->            <pee>The solution: Have the AI do what it does best \u2014 extracting the relevant content from the email stream, and then use another (deterministic) application to take that data and populate an existing Word document template.<\/pee><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->          <!-- Footer with Read More and Meta Button --><!-- [et_pb_line_break_holder] -->          <div class=\"aidiary-post-footer\"><!-- [et_pb_line_break_holder] -->            <!-- <a href=\"#\" class=\"aidiary-read-more-link\">More<\/a> --><!-- [et_pb_line_break_holder] -->            <pee><\/pee><!-- [et_pb_line_break_holder] -->            <div class=\"aidiary-meta-button\">\ud83d\udd28\ud83d\udd28\ud83d\udd28\ud83d\udd28\ud83d\udd28<\/div><!-- [et_pb_line_break_holder] -->          <\/div><!-- [et_pb_line_break_holder] -->        <\/div><!-- [et_pb_line_break_holder] -->      <\/article><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->      <!-- *********************************** --><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] -->    <\/div><!-- [et_pb_line_break_holder] -->  <\/div><!-- [et_pb_line_break_holder] --><\/div><!-- [et_pb_line_break_holder] --><!-- [et_pb_line_break_holder] --><script><!-- [et_pb_line_break_holder] -->  \/\/ Initialize accordions: nested first, parent last<!-- [et_pb_line_break_holder] -->  initializeAccordions([<!-- [et_pb_line_break_holder] -->    'accordion-container-abstraction-detail',<!-- [et_pb_line_break_holder] -->    'accordion-container-ai-impact-parent'<!-- [et_pb_line_break_holder] -->  ], null, false);<!-- [et_pb_line_break_holder] --><\/script><!-- [et_pb_line_break_holder] --><!-- \/MC AIDiariesJournal2026-All -->[\/et_pb_code][\/et_pb_column][\/et_pb_row][\/et_pb_section]\n<span class=\"et_bloom_bottom_trigger\"><\/span>","protected":false},"excerpt":{"rendered":"<p><div class=\"back-link back-link-hover-blue back-link-offset-up back-link-smaller\"><a href=\"https:\/\/daviddurlach.com\/writingsPP\/droplets\/ai-diaries\/\">( &larr; Back to AI Diaries)<\/a><\/div> March 6, 2026 Last updated March 20, 2026 &#8211;> 2026 Journal Of AI Tidbits For items not quite meeting the \u201cfull separate post page\u201d threshold.<\/p>\n","protected":false},"author":3,"featured_media":0,"parent":259997,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"_acf_changed":false,"_et_pb_use_builder":"on","_et_pb_old_content":"<!-- wp:divi\/placeholder \/-->","_et_gb_content_width":"","inline_featured_image":false,"footnotes":""},"categories":[1],"tags":[45],"folder":[],"class_list":["post-264139","page","type-page","status-publish","hentry","category-uncategorized","tag-tag-no-header"],"acf":[],"_links":{"self":[{"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/pages\/264139","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/comments?post=264139"}],"version-history":[{"count":30,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/pages\/264139\/revisions"}],"predecessor-version":[{"id":264682,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/pages\/264139\/revisions\/264682"}],"up":[{"embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/pages\/259997"}],"wp:attachment":[{"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/media?parent=264139"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/categories?post=264139"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/tags?post=264139"},{"taxonomy":"folder","embeddable":true,"href":"https:\/\/daviddurlach.com\/writingsPP\/droplets\/wp-json\/wp\/v2\/folder?post=264139"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}