{"id":630527,"date":"2026-07-05T20:20:00","date_gmt":"2026-07-05T20:20:00","guid":{"rendered":"https:\/\/buglecall.org\/?p=630527"},"modified":"2026-07-05T20:20:00","modified_gmt":"2026-07-05T20:20:00","slug":"the-biggest-problem-with-ai-today-2","status":"publish","type":"post","link":"https:\/\/buglecall.org\/?p=630527","title":{"rendered":"The Biggest Problem With AI Today"},"content":{"rendered":"<p><span class=\"field field--name-title field--type-string field--label-hidden\">The Biggest Problem With AI Today <\/span><\/p>\n<div class=\"clearfix text-formatted field field--name-body field--type-text-with-summary field--label-hidden field__item\">\n<p><em>By Christopher Penn, of <a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-the-biggest-problem\">Almost Timely News<\/a><\/em><\/p>\n<p>What\u2019s the biggest problem in AI today? Is it cost, with token budgets being blown out of the water by agentic AI? Is it sustainability, with AI consuming electricity and fresh water? Is it ethics, with tech companies cramming AI into everything?<\/p>\n<p>I think it\u2019s deeper than that. Those are all symptoms of a much deeper-rooted problem: nobody\u2019s making decisions.<\/p>\n<p>Or more correctly, <strong>we\u2019ve abdicated far too much of our executive function to AI. We\u2019ve surrendered our thinking<\/strong>.\u00a0<\/p>\n<p>Let\u2019s dig in.<\/p>\n<h2 class=\"header-anchor-post\">Part 1: Where This Issue Came From<\/h2>\n<p>On Friday afternoon, I was mulling over what I wanted to cover in this week\u2019s issue. It\u2019s a holiday weekend here in the USA, so not as many folks will be reading, and that\u2019s okay. (I appreciate that YOU are) And I\u2019ve covered a ton recently:<\/p>\n<ul>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-how-to-improve-6db?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">How to improve advertising with AI<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-how-i-proved-listicles?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">Why listicles may cause more harm than good<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-4-angles-on-local?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">Setting up private, local models<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-how-ai-detection?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">How AI detection works<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-a-better-mental?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">AI for GEO mental models<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-ai-and-geo-advice?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">AI for retail GEO<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-18-ways-to-save?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">18 ways to save token budgets<\/a><\/li>\n<li><a href=\"https:\/\/almosttimely.substack.com\/p\/almost-timely-news-how-to-force-ai?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">How to make AI write better<\/a><\/li>\n<\/ul>\n<p>So on a whim, I set up a NotebookLM with the last 180 days of conversations from over 40 different subreddits, like r\/marketing, r\/chatgpt, etc. &#8211; everything around marketing, business, and AI. I connected it to Claude Code with the NotebookLM command line tool (the most token\u2014efficient way for Claude to talk to NotebookLM), and then put all of my 2026 newsletters year to date into an input folder.<\/p>\n<p>I asked Claude to compare what I\u2019ve written about thus far this year with what folks are finding their hardest problems are with AI. Claude spit out a list of 10 major things derived from over 800,000 words of foaming at the mouth on Reddit that it thought might be good newsletter topics:<\/p>\n<ul>\n<li>AI Visibility challenges<\/li>\n<li>Agentic oversight is degrading<\/li>\n<li>AI deployment is broken<\/li>\n<li>40-60% of company budget is wasted on the wrong models<\/li>\n<li>AI is a rental<\/li>\n<li>AI sycophancy is screwing up synthetic focus groups<\/li>\n<li>AI detectors don\u2019t work<\/li>\n<li>AI is hollowing out corporations and no one\u2019s hiring junior staff<\/li>\n<li>People measure AI by tokenmaxxing<\/li>\n<li>Marketers are basically unpaid labor for AI companies training data<\/li>\n<\/ul>\n<p>Claude was REALLY pushing for me to write about how measurement is broken in marketing and AI today, and I might do that at some point, but that\u2019s not what I see when I look at this laundry list. Yes, there are measurement issues in many of them, data issues in many of them, but&#8230; measurement being broken is the symptom of what I said earlier &#8211; we\u2019ve abdicated executive function.<\/p>\n<p>For those who aren\u2019t analytics nerds, you know that measurement is a trailing indicator. It\u2019s not a leading indicator.<\/p>\n<h2 class=\"header-anchor-post\">Part 2: Executive Function Recap<\/h2>\n<p>As a reminder, I bucket executive function into four categories that I call PODS:<\/p>\n<ul>\n<li>Plan: you think about achieving something in the future and make a plan to get there from here<\/li>\n<li>Organize: you take what you have and try to make sense of it<\/li>\n<li>Decide: you take what you have and make decisions about it<\/li>\n<li>Solve: you solve the problems you have<\/li>\n<\/ul>\n<p>Yes, there is more nuance to executive function than this, but this handy, short list is an easy way to see what our brains are doing. That\u2019s critical thinking, one of the worst-named practices we have.<\/p>\n<p>Why? Because critical thinking isn\u2019t about being critical, per se. It\u2019s about metacognition &#8211; the definition of which is thinking about thinking. When you\u2019re thinking about how you think, you open the door to improvements, to growth.<\/p>\n<p>Thinking about thinking means asking questions and reflecting &#8211; is this the best way to do something? How could I do this better? How could I derive more enjoyment from this thing I\u2019m doing? It\u2019s not criticizing yourself as much as it is recognizing what you\u2019re doing and whether it\u2019s working or not.<\/p>\n<p>When you\u2019re planning, organizing, deciding, and solving, you\u2019re inherently thinking about thinking. Every time you plan, every time you bring order to chaos, you have to check in with your own brain to see if what you\u2019re doing is moving you closer to the goal posts.<\/p>\n<p>Executive function is one of the things that defines our sentience as living creatures. Every sentient creature from a mouse to us does these tasks. You\u2019ve read or heard stories about crows fashioning tools from wire to solve problems, you\u2019ve watched dogs and cats make decisions and plan. I\u2019ve watched my own cat measure optically whether or not she can make a particular jump.<\/p>\n<p>Properly prompted, today\u2019s AI tools are superb at executive functions as well. Given the right frameworks, harnesses, and data, they can plan, organize, decide, and solve better than we can at most language-based tasks.<\/p>\n<p>And therein lies the actual problem.<\/p>\n<h2 class=\"header-anchor-post\">Part 3: The Tale of the Tape<\/h2>\n<p>Let\u2019s look at each of the 10 topics Claude suggested to see the threads that connect them.<\/p>\n<p>AI Visibility challenges: when you read the verbatims of what people are saying about AI visibility measurement, you can tell they\u2019re pretty much making it up. This is especially true of software vendors that are offering and peddling solutions that have very little grounding in reality &#8211; and yet, stakeholders eat this stuff up because they\u2019d rather have certainty about a wrong number than accept uncertainty or no number at all. they are not thinking about their thinking.<\/p>\n<p>Agentic oversight is degrading: the commenters on Reddit focused on the fact that as agents get more sophisticated, it\u2019s harder and harder to follow along to see what they\u2019re doing. So we just hit OK all the time &#8211; if we\u2019re even thinking about a human in the loop. We\u2019ve forfeit our authority here. In fact, some AI tools have this built in as a feature. Claude calls it dangerously skip permissions. Qwen calls it YOLO mode.<\/p>\n<p><span>AI deployment is broken: here, the discussion is about stakeholders telling their stakeholders that the organization has deployed AI without any sense of the impact that it\u2019s had. One poster cited a statistic that 29% of companies see significant ROI from AI, even though individual employees are claiming 5x productivity increases. The math doesn\u2019t math. Here, people don\u2019t want to think and reflect about what deployment even means. <\/span><a href=\"https:\/\/www.trustinsights.ai\/newsletter?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">Katie\u2019s been writing a lot about this in the Trust Insights newsletter the last few weeks<\/a><span>. At its heart, we are confusing using AI with getting results out of AI.<\/span><\/p>\n<p>40-60% of budget is wasted: here, folks are talking about how everyone just accepts the default model in AI tools, which is typically the most expensive one. Claude, for example, defaults to Opus 4.8, which is a much more expensive model than Sonnet 5 or Haiku 4.5. We\u2019re not thinking. We\u2019re not making decisions about cost trade-offs versus effectiveness. Another person pointed out that this is by design to create habits. It\u2019s about habit formation for the most expensive models so that when the subsidization of today\u2019s AI ends, we are accustomed to using the most expensive models. This is brain hijacking in a way.<\/p>\n<p>AI is a rental: in this particular topic, the discussion centers around what you actually own in AI, which is very little if you are using today\u2019s closed weights frontier models. Particularly Anthropic\u2019s on-again, off-again rollout of Fable 5, thanks to U.S. export controls, was a wake-up call to the entire industry that you don\u2019t own anything in SaaS, any more than you own music in Spotify or own videos in Netflix &#8211; but people think they do.<\/p>\n<p>Sycophancy in focus groups: even though we have good academic research showing that properly prompted AI models can emulate human purchase intent with about 90% accuracy, the level of sycophancy in AI models steers them towards confirmation bias in most situations. This is especially true of synthetic focus groups; when people use AI to simulate consumer intent, what they\u2019re really doing is reinforcing their own biases most of the time. There\u2019s no reflection or questioning the AI output.<\/p>\n<p>AI detectors don\u2019t work: A perpetual favorite topic of mine. This thread of conversation revolved around how companies are using AI detectors to identify the use of AI in situations where it\u2019s not appropriate, without recognizing that the detectors themselves are also broken. In testing I did 3 weeks ago now, AI detectors falsely flagged human outputs 1 out of 7 times. No one is thinking and reflecting enough about who\u2019s watching the watchers.<\/p>\n<p>AI is hollowing out companies: I really liked this quote from the agency owners subreddit:<\/p>\n<p><span>\u201c<\/span><em>What\u2019s strange is nobody decided this. There was no meeting where we discussed this. We automated one annoying task, then another, and one day the job had hollowed out from the inside.<\/em><span>\u201c<\/span><\/p>\n<p>This erosion of tasks is all about a lack of cognition, a lack of reflection, a lack of a plan. No one\u2019s making decisions &#8211; just leaving it up to the machines, a bit more each day.<\/p>\n<p>Tokenmaxxing: this was reflecting on Meta\u2019s most recent news story in which they were on track to spend several billion dollars in AI tokens because they measured AI productivity based on token spend, the dumbest possible way to measure AI.<\/p>\n<p>Marketers as unpaid trainers: this was a whole bunch of ranting about how marketers are effectively unpaid trainers for AI platforms. The more content we produce, the more AI has to train on while simultaneously competing for the tasks we\u2019re paid to do. Here, the thread was about how the average marketer isn\u2019t thinking or reflecting about their relationship to AI.<\/p>\n<p>And this laundry list of 10 items isn\u2019t everything, not by a long shot. Think about how else people use AI without thinking, without thinking about their thinking. Go on LinkedIn and look at the endless streams of comment-bots all paraphrasing the same template over and over again. Look at the workslop flooding your inbox, read the reports your agencies send you that are clearly copy paste jobs.<\/p>\n<p>When we put aside the direction that Claude wanted to nudge this issue of the newsletter, it becomes pretty apparent that it\u2019s really about how much we think about thinking. How self-aware are we? How well and accurately do we perceive our relationship with AI?<\/p>\n<p>Most of all, do we see the amount of executive function we\u2019ve ceded to AI?<\/p>\n<h2 class=\"header-anchor-post\">Part 4: The Antidote<\/h2>\n<p>\u201cNobody decided this\u201d is haunting me. When you hand off executive functions to AI, who is making the decisions? No one. There\u2019s no one accountable for a decision because the machine is making it for us. Whether it\u2019s building a PowerPoint deck, assembling a report for a client, creating content for a newsletter, when the machine does it, there\u2019s no accountability and there\u2019s no decision making on our part other than approving it.<\/p>\n<p>And this leads to a bunch of bad outcomes, everything from job loss to dissatisfaction with your own work. You know, when you use AI to offload a task, that you didn\u2019t do the work &#8211; and you take no pride in it, any more than you\u2019d take pride in the work that a contractor did on your behalf.<\/p>\n<p>Think about this in the context of parents. Go to any parent\u2019s house and you\u2019ll likely see art that the kids made when they were young. The art is generally, objectively, pretty bad. But the parent values it not because of the quality of the art, but because of the level of effort made by the child. They take pride in their child\u2019s efforts, and the child takes pride in what they did in their efforts. For good or ill, when people use AI, they themselves feel like they haven\u2019t made an effort, and the person on the receiving end also feels like they didn\u2019t make an effort.<\/p>\n<p>Sometimes, you don\u2019t even understand the work if you\u2019ve outsourced it. You present it to your stakeholders, and the first question they ask that isn\u2019t in the prepared materials leads to panic city because you can\u2019t answer it, like buying a cake at the store instead of baking it yourself and then having someone ask if a specific allergen is in it. And you\u2019re left scrambling, looking for the label to see what\u2019s actually in the cake.<\/p>\n<p>So my suggested antidote is this: for every task that matters, always start with someting you lead, and force the machines to educate you.<\/p>\n<p>For example, when I compile monthly reports for Trust Insights clients, I turn on my voice recorder and I review the data myself. I talk out loud what I see, what I think, what makes sense and what doesn\u2019t make sense, and then I have AI transcribe it. After the transcription is complete, I ask AI to review it and show me what I missed. I ask it to ask me questions, to record more information, to fish more information from me.<\/p>\n<p>I also ask it, especially around anything in my subject matter expertise, to find me resources to learn and read about its recommendations. Recently, I was asking it to choose from a catalog I\u2019d prepared of over 1,000 different analytical techniques, and it chose an interesting ensemble of 3 techniques, one of which I didn\u2019t know well. So I had it teach me that, so that instead of me passively accepting its recommendations, I learned something. I got better as a professional. I grew my subject matter expertise.<\/p>\n<p>If you think about it, this is not only rational from the perspective of delivering great quality work, it\u2019s also rational from the perspective of my value. If I\u2019m nothing more than a copy paste drone, a meat-based interface to an LLM, then why does my company need me? Why would my clients pay for me when they could just pay to ask ChatGPT or Claude the exact same things?<\/p>\n<p>What they\u2019re paying for is my expertise, my skills not only at using the technology, but the specific lens I direct it with, and the perspective that only I can bring. And if I\u2019m using AI to constantly improve that expertise, to improve that domain knowledge, then they should keep paying for me.<\/p>\n<p>Outside my subject matter expertise, I start with deep research, using AI tools to gather information and then having them create a synthesis. Once I\u2019ve got that, then I have it create a checklist of what constitutes quality in the domain I\u2019m working in. Finally, I sit down with the creations and I read and learn for myself. I have AI make infographics or podcast summaries to learn the domain so that I can connect it to my expertise.<\/p>\n<p><span>Agentic AI &#8211; tools like Claude Code, OpenCode, etc. &#8211; are phenomenal researchers, far better than the web-based deep research tools folks have become accustomed to in the past couple of years. When you use a research agent, it has a lot more latitude to gather up sources, to take the time to write down notes and observations, and to synthesize conclusions from the data it has. If you use something like the <\/span><a href=\"https:\/\/www.trustinsights.ai\/casino?utm_source=almost-timely-newsletter&amp;utm_medium=email&amp;utm_campaign=almost-timely-2026-07-05\">Trust Insights CASINO research framework<\/a><span>, you\u2019ll get some amazing results from the tools that tend to have fewer hallucinations than their web-based counterparts.<\/span><\/p>\n<p>Then with that research data in hand, you use it to become a better professional within your domain. You use it to level yourself up. You use it to add to your insights instead of substitute for your insights.<\/p>\n<h2 class=\"header-anchor-post\">Part 5: Wrapping Up<\/h2>\n<p>The biggest problem in AI today is the delegation of our executive function to machines. Whether it\u2019s accountability (machines have none), deskilling, or dissatisfaction with our work, the moment we forfeit executive function is the moment when AI becomes more problem than solution.<\/p>\n<p>We can boil it all down to a simple set of questions:<\/p>\n<ol>\n<li>\n<p>Does the use of AI make the output better?<\/p>\n<\/li>\n<li>\n<p>Does the use of AI make me better?<\/p>\n<\/li>\n<\/ol>\n<p>If the answer isn\u2019t yes to BOTH, then you\u2019re not using it well.<\/p>\n<p>Properly used, AI is one of the greatest professional development tools ever created.<\/p>\n<p>Improperly used, it\u2019s one of the most destructive forces your career has ever known, because the moment you offload a task to AI, your own skills at that task get rusty.<\/p>\n<p>And once something becomes rusty enough, it\u2019s cheaper and easier to replace it.<\/p>\n<p><em>More in the <a href=\"https:\/\/almosttimely.substack.com\/\">Almost Timely Newsletter<\/a><\/em><\/p>\n<p>* * *<strong><em>\u00a0Next-level <a href=\"https:\/\/store.zerohedge.com\/collections\/ranch-wagyu\">Wagyu<\/a>, now at ZeroHedge Store<\/em><\/strong><\/p>\n<p><a data-image-external-href=\"https:\/\/store.zerohedge.com\/collections\/ranch-wagyu\" data-image-href=\"https:\/\/store.zerohedge.com\/collections\/ranch-wagyu\" data-link-option=\"2\" href=\"https:\/\/store.zerohedge.com\/collections\/ranch-wagyu\"><img fetchpriority=\"high\" decoding=\"async\" data-entity-type=\"file\" data-entity-uuid=\"35c1f47b-212f-4f4a-bd24-5e7770dfab86\" data-responsive-image-style=\"inline_images\" height=\"333\" width=\"500\" class=\"inline-images image-style-inline-images\" src=\"https:\/\/assets.zerohedge.com\/s3fs-public\/styles\/inline_image_mobile\/public\/inline-images\/kc-cattle-tomahawk_80.png?itok=2-Mk4edG\" alt=\"\" \/><\/a><\/p>\n<\/div>\n<p>      <span class=\"field field--name-uid field--type-entity-reference field--label-hidden\"><a title=\"View user profile.\" href=\"https:\/\/cms.zerohedge.com\/users\/tyler-durden\" lang=\"\" class=\"username\" xml:lang=\"\">Tyler Durden<\/a><\/span><br \/>\n<span class=\"field field--name-created field--type-created field--label-hidden\">Sun, 07\/05\/2026 &#8211; 16:20<\/span><img decoding=\"async\" src=\"https:\/\/assets.zerohedge.com\/s3fs-public\/styles\/inline_image_mobile\/public\/inline-images\/kc-cattle-tomahawk_80.png?itok=2-Mk4edG\" title=\"The Biggest Problem With AI Today\" \/><\/p>","protected":false},"excerpt":{"rendered":"<p>The Biggest Problem With AI Today By Christopher Penn, of Almost Timely News What\u2019s the biggest problem in AI today? Is it cost, with token budgets being blown out of the water by agentic AI? Is it sustainability, with AI consuming electricity and fresh water? Is it ethics, with tech companies cramming AI into everything?&hellip; <a class=\"more-link\" href=\"https:\/\/buglecall.org\/?p=630527\">Continue reading <span class=\"screen-reader-text\">The Biggest Problem With AI Today<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":630519,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rop_custom_images_group":[],"rop_custom_messages_group":[],"rop_publish_now":"initial","rop_publish_now_accounts":[],"rop_publish_now_history":[],"rop_publish_now_status":"pending","footnotes":""},"categories":[18,19,10,21,12,11,9],"tags":[],"class_list":["post-630527","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cancel-culture","category-censorship","category-civil-liberties","category-election-integrity","category-equal-justice","category-free-speech","category-religious-freedom","entry"],"_links":{"self":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts\/630527","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=630527"}],"version-history":[{"count":0,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/posts\/630527\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=\/wp\/v2\/media\/630519"}],"wp:attachment":[{"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=630527"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=630527"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/buglecall.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=630527"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}