{"id":34782,"topic":"ai","source":"Forbes","title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","url_hash":"b90bd08cdfb413f077848ec6b41b2fcb41003616","author":"","summary":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming Forbes","content":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. In the early nineties, that word was quality. Today it is AI. The vocabulary changes. The dynamic underneath it does not.\nMark Zbaracki noticed the dynamic thirty years ago. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management. What he found was not a story about quality. It was a story about language, about how a set of statistical practices got wrapped in a vocabulary that spread far faster than the practices themselves.\nI spoke with him recently because the gap looks familiar again. When we spoke, he had just reread his own paper to prepare a course on strategy implementation. Rereading your own work from three decades ago is its own kind of test. The dynamic held.\nThe Two Versions of Everything\nTQM existed in two versions at once. There was a technical version: control charts, statistical methods, the group practices that let people understand a production system in more depth. And there was a rhetorical version: the slogans, the presentations, the claim of transformation. The two were not the same, and they did not travel together. The rhetoric moved on its own.\nWhen I asked Zbaracki whether leaders knew they were performing a language they weren't practicing, his answer was no. He believed they wanted to believe. They didn't have the substance behind what they were doing, he told me, and in many cases had no way of accessing it. They were caught in the zeitgeist, using the vocabulary to establish an authority they weren't sure they had earned.\nThe gap between rhetoric and reality is rarely a gap between honest and dishonest people. It is a gap between people and structures. Words are cheap to adopt and practices are expensive to learn, so the language runs ahead, and the leader who has adopted it now has a stake in not being questioned about the substance.\nThe mechanism is plain. Jumping on the bandwagon is easy. Not jumping is hard. The incentives run one direction, cycle after cycle: TQM, then Six Sigma picking up the same themes under a different name, then the next term, and the next.\nWhen the Language Becomes Unquestionable\nA second move happens once the vocabulary is installed, and it is the more dangerous one. The language stops being a description and becomes a boundary. To question it is to mark yourself as someone who isn't keeping up.\nI put this to Zbaracki as a hypothesis: language gets installed partly to hide structural inadequacy, and once installed it becomes unquestionable. If an organization has staked itself on a word, raising your hand to ask what the word actually means reads as resistance, not curiosity. He didn't hesitate. \"Absolutely,\" he said. \"Without a doubt.\"\nThen he told me about the defense contractor. At the end of his study he gave a presentation telling one of the organizations, in effect, that they were not doing TQM. They never spoke to him again. He called it a rookie mistake. But inside that same organization, the true believers were glad to hear it said out loud. They had been frustrated for a long time. The refusal to talk to him wasn't a rebuttal. It was the preservation of a position.\nThat is the tell. When an outside question is met not with an argument but with silence and distance, the silence is the data. It signals that the language is doing load-bearing work the structure cannot do on its own, and that examining it too closely threatens something the organization has decided not to examine.\nWhy AI Is the Harder Case\nHere is where the parallel to AI both holds and breaks, and the break is the part leaders should consider.\nThe parallel holds on hype. Both arrived on a wave, and both waves were pushed. But the pushers are not the same. TQM had its gurus, and even among them a tension ran between the ones who understood the substance, Deming and Juran, and the ones riding the enthusiasm. Those people never held the kind of power now sitting behind AI. A figure like Sam Altman can generate hype at a scale the quality gurus could only have imagined.\nThe parallel breaks somewhere more important. With TQM, Zbaracki understood the tools well enough to walk into a room and quickly tell whether the people talking knew what they were talking about. He could see the substance. With generative AI, he told me, he cannot. He described sitting with capable people at large IT firms, enthusiastically describing how they set AI agents against one another, and having no way to judge the veracity of what they were saying. The tool no longer exposes its own substance to the person using it.\nHe drew a line most of the AI conversation skips. Predictive AI, to him, is not that different from TQM. There is a data scientist behind the model, the predictions can be tested against reality, and a domain expert can come to understand what the tool is doing. Netflix, dynamic pricing, much of medicine: predictive AI is legible to someone willing to do the hard work. Generative AI is the stranger case. The data used to build the model shapes the output profoundly, and we do not control that data, do not have access to it, and often cannot say what the tool is doing at all.\nSo the danger does not sit in the tool. It lies with the choices we make around it. That distinction, between the technology and the choices surrounding it, is exactly the distinction the rhetoric is built to blur.\nThe Gap, Now Measured\nFor most of the last two years this was a pattern you could feel in a room before you could prove it. Now the proof has arrived, and it describes Zbaracki's gap almost to the decimal.\nWhen MIT's NANDA initiative studied enterprise AI in 2025, it found that despite roughly $40 billion invested, about 95% of generative AI pilots delivered no measurable impact on the bottom line. The researchers named it directly: the divide between high adoption and low transformation. One manufacturing executive gave them a sentence that could have come from the TQM study: the hype says everything has changed, but in their operations, nothing fundamental had shifted. That is rhetoric and reality, thirty years later, in a single line.\nMy employer Gallup’s data shows the same split from the other side of the desk. Half of U.S. employees now use AI at work at least occasionally. But asked whether it has changed how the organization actually works, only 12% strongly agree that AI has transformed how work gets done where they are. The practice is real where a person can hold it in their hands and rhetorical everywhere above that. Meanwhile only 22% of employees say their organization has communicated a clear AI plan. \"We are an AI company\" gets asserted from the top of organizations where most people cannot confirm it from where they sit. The word arrived. The structure that would make it true did not.\nPull the Rhetoric Toward the Experience\nNone of this is an argument against AI. Zbaracki was careful about that. Things necessarily come in as rhetoric, he told me, because you do not yet have the experience; the concept arrives before the practice can. The question is not how to keep the language out. It is how to use it to pull people toward the experience, in ways that let them confront the inadequacy of their own words rather than defend them.\nThe first discipline is to make the tool expose its own substance. Before any initiative names what it will deliver, it should name what it actually does. For a predictive system that is answerable. For a generative system the honest answer is often \"we cannot fully say,\" and that admission belongs on the table rather than buried under use-case language. MIT found the pilots that survived were integrated into a real workflow and able to admit what they did not know, while the ones that died were the demos and use cases that looked flawless in the boardroom and collapsed in the field. The discipline is the plain question asked out loud: walk me through what this is doing and where the data came from.\nThe second is to check whether anything but the vocabulary has changed. The transformations that failed ran on single-loop learning: people changing what they said while leaving what they did untouched. A team can adopt the language of AI fluency while the approval layers, the incentives, and the question of who is allowed to dissent stay exactly where they were. When that happens, the rhetoric is running ahead of the reality, and it will keep running ahead until the structure moves.\nHow to Call It Out\nNotice what happens when someone asks a plain question. When \"what does this actually do?\" is met not with an answer but with a certain distance, a sense that the questioner is lagging behind, not future-forward, the distance is the answer. Substance survives an honest question. Rhetoric cannot, which is why it defends itself with the implication that asking is the problem.\nI have watched what that costs when no one names it. A company rolls out its AI transformation across the front line. On paper the case is airtight: faster decisions, leaner operations, better insight. Six months in, adoption has stalled. The managers cannot translate the change into language their teams can use, because no one translated it into language they could use first. At a rollout meeting, someone finally asks the plain question, \"what does this actually do?\", and the silence that follows drains the energy from the room. That silence is the whole story. It is the defense contractor declining to talk, relocated to a conference room where the people paying for the strategy are the ones left holding the silence.\nThere is a discipline the AI moment asks of leaders, and it is not fluency. It is the willingness to be the person in that rollout meeting who asks what the word means, and then sits in the silence long enough for everyone else to hear it too. A house of cards stands only as long as no one is allowed to breathe near it.\nThe organizations that will pay the highest price for this AI wave will not be the ones that moved too slowly. They will be the ones that mistook the vocabulary for the capability, staked themselves on a language they could not yet practice, and made sure no one was permitted to notice the difference, until the bill came due, and the words were all that was left.","image_url":"https://imageio.forbes.com/specials-images/imageserve/6a4c84b33b3cd30269e2a868/0x0.jpg?format=jpg&height=900&width=1600&fit=bounds","lang":"en","published_at":"2026-07-12T16:29:35+00:00","fetched_at":"2026-07-12T17:15:04+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management.","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}},"news_item":{"id":34782,"canonical_url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","source_url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","source_name":"Forbes","author":null,"published_at":"2026-07-12T16:29:35+00:00","locale":"en","topic":"ai","tags":[],"rss_summary":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming Forbes","full_text":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. In the early nineties, that word was quality. Today it is AI. The vocabulary changes. The dynamic underneath it does not.\nMark Zbaracki noticed the dynamic thirty years ago. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management. What he found was not a story about quality. It was a story about language, about how a set of statistical practices got wrapped in a vocabulary that spread far faster than the practices themselves.\nI spoke with him recently because the gap looks familiar again. When we spoke, he had just reread his own paper to prepare a course on strategy implementation. Rereading your own work from three decades ago is its own kind of test. The dynamic held.\nThe Two Versions of Everything\nTQM existed in two versions at once. There was a technical version: control charts, statistical methods, the group practices that let people understand a production system in more depth. And there was a rhetorical version: the slogans, the presentations, the claim of transformation. The two were not the same, and they did not travel together. The rhetoric moved on its own.\nWhen I asked Zbaracki whether leaders knew they were performing a language they weren't practicing, his answer was no. He believed they wanted to believe. They didn't have the substance behind what they were doing, he told me, and in many cases had no way of accessing it. They were caught in the zeitgeist, using the vocabulary to establish an authority they weren't sure they had earned.\nThe gap between rhetoric and reality is rarely a gap between honest and dishonest people. It is a gap between people and structures. Words are cheap to adopt and practices are expensive to learn, so the language runs ahead, and the leader who has adopted it now has a stake in not being questioned about the substance.\nThe mechanism is plain. Jumping on the bandwagon is easy. Not jumping is hard. The incentives run one direction, cycle after cycle: TQM, then Six Sigma picking up the same themes under a different name, then the next term, and the next.\nWhen the Language Becomes Unquestionable\nA second move happens once the vocabulary is installed, and it is the more dangerous one. The language stops being a description and becomes a boundary. To question it is to mark yourself as someone who isn't keeping up.\nI put this to Zbaracki as a hypothesis: language gets installed partly to hide structural inadequacy, and once installed it becomes unquestionable. If an organization has staked itself on a word, raising your hand to ask what the word actually means reads as resistance, not curiosity. He didn't hesitate. \"Absolutely,\" he said. \"Without a doubt.\"\nThen he told me about the defense contractor. At the end of his study he gave a presentation telling one of the organizations, in effect, that they were not doing TQM. They never spoke to him again. He called it a rookie mistake. But inside that same organization, the true believers were glad to hear it said out loud. They had been frustrated for a long time. The refusal to talk to him wasn't a rebuttal. It was the preservation of a position.\nThat is the tell. When an outside question is met not with an argument but with silence and distance, the silence is the data. It signals that the language is doing load-bearing work the structure cannot do on its own, and that examining it too closely threatens something the organization has decided not to examine.\nWhy AI Is the Harder Case\nHere is where the parallel to AI both holds and breaks, and the break is the part leaders should consider.\nThe parallel holds on hype. Both arrived on a wave, and both waves were pushed. But the pushers are not the same. TQM had its gurus, and even among them a tension ran between the ones who understood the substance, Deming and Juran, and the ones riding the enthusiasm. Those people never held the kind of power now sitting behind AI. A figure like Sam Altman can generate hype at a scale the quality gurus could only have imagined.\nThe parallel breaks somewhere more important. With TQM, Zbaracki understood the tools well enough to walk into a room and quickly tell whether the people talking knew what they were talking about. He could see the substance. With generative AI, he told me, he cannot. He described sitting with capable people at large IT firms, enthusiastically describing how they set AI agents against one another, and having no way to judge the veracity of what they were saying. The tool no longer exposes its own substance to the person using it.\nHe drew a line most of the AI conversation skips. Predictive AI, to him, is not that different from TQM. There is a data scientist behind the model, the predictions can be tested against reality, and a domain expert can come to understand what the tool is doing. Netflix, dynamic pricing, much of medicine: predictive AI is legible to someone willing to do the hard work. Generative AI is the stranger case. The data used to build the model shapes the output profoundly, and we do not control that data, do not have access to it, and often cannot say what the tool is doing at all.\nSo the danger does not sit in the tool. It lies with the choices we make around it. That distinction, between the technology and the choices surrounding it, is exactly the distinction the rhetoric is built to blur.\nThe Gap, Now Measured\nFor most of the last two years this was a pattern you could feel in a room before you could prove it. Now the proof has arrived, and it describes Zbaracki's gap almost to the decimal.\nWhen MIT's NANDA initiative studied enterprise AI in 2025, it found that despite roughly $40 billion invested, about 95% of generative AI pilots delivered no measurable impact on the bottom line. The researchers named it directly: the divide between high adoption and low transformation. One manufacturing executive gave them a sentence that could have come from the TQM study: the hype says everything has changed, but in their operations, nothing fundamental had shifted. That is rhetoric and reality, thirty years later, in a single line.\nMy employer Gallup’s data shows the same split from the other side of the desk. Half of U.S. employees now use AI at work at least occasionally. But asked whether it has changed how the organization actually works, only 12% strongly agree that AI has transformed how work gets done where they are. The practice is real where a person can hold it in their hands and rhetorical everywhere above that. Meanwhile only 22% of employees say their organization has communicated a clear AI plan. \"We are an AI company\" gets asserted from the top of organizations where most people cannot confirm it from where they sit. The word arrived. The structure that would make it true did not.\nPull the Rhetoric Toward the Experience\nNone of this is an argument against AI. Zbaracki was careful about that. Things necessarily come in as rhetoric, he told me, because you do not yet have the experience; the concept arrives before the practice can. The question is not how to keep the language out. It is how to use it to pull people toward the experience, in ways that let them confront the inadequacy of their own words rather than defend them.\nThe first discipline is to make the tool expose its own substance. Before any initiative names what it will deliver, it should name what it actually does. For a predictive system that is answerable. For a generative system the honest answer is often \"we cannot fully say,\" and that admission belongs on the table rather than buried under use-case language. MIT found the pilots that survived were integrated into a real workflow and able to admit what they did not know, while the ones that died were the demos and use cases that looked flawless in the boardroom and collapsed in the field. The discipline is the plain question asked out loud: walk me through what this is doing and where the data came from.\nThe second is to check whether anything but the vocabulary has changed. The transformations that failed ran on single-loop learning: people changing what they said while leaving what they did untouched. A team can adopt the language of AI fluency while the approval layers, the incentives, and the question of who is allowed to dissent stay exactly where they were. When that happens, the rhetoric is running ahead of the reality, and it will keep running ahead until the structure moves.\nHow to Call It Out\nNotice what happens when someone asks a plain question. When \"what does this actually do?\" is met not with an answer but with a certain distance, a sense that the questioner is lagging behind, not future-forward, the distance is the answer. Substance survives an honest question. Rhetoric cannot, which is why it defends itself with the implication that asking is the problem.\nI have watched what that costs when no one names it. A company rolls out its AI transformation across the front line. On paper the case is airtight: faster decisions, leaner operations, better insight. Six months in, adoption has stalled. The managers cannot translate the change into language their teams can use, because no one translated it into language they could use first. At a rollout meeting, someone finally asks the plain question, \"what does this actually do?\", and the silence that follows drains the energy from the room. That silence is the whole story. It is the defense contractor declining to talk, relocated to a conference room where the people paying for the strategy are the ones left holding the silence.\nThere is a discipline the AI moment asks of leaders, and it is not fluency. It is the willingness to be the person in that rollout meeting who asks what the word means, and then sits in the silence long enough for everyone else to hear it too. A house of cards stands only as long as no one is allowed to breathe near it.\nThe organizations that will pay the highest price for this AI wave will not be the ones that moved too slowly. They will be the ones that mistook the vocabulary for the capability, staked themselves on a language they could not yet practice, and made sure no one was permitted to notice the difference, until the bill came due, and the words were all that was left.","excerpt":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management.","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 10475 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","summary":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management.","source":"Forbes","date":"2026-07-12T16:29:35+00:00","content":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. In the early nineties, that word was quality. Today it is AI. The vocabulary changes. The dynamic underneath it does not.\nMark Zbaracki noticed the dynamic thirty years ago. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management. What he found was not a story about quality. It was a story about language, about how a set of statistical practices got wrapped in a vocabulary that spread far faster than the practices themselves.\nI spoke with him recently because the gap looks familiar again. When we spoke, he had just reread his own paper to prepare a course on strategy implementation. Rereading your own work from three decades ago is its own kind of test. The dynamic held.\nThe Two Versions of Everything\nTQM existed in two versions at once. There was a technical version: control charts, statistical methods, the group practices that let people understand a production system in more depth. And there was a rhetorical version: the slogans, the presentations, the claim of transformation. The two were not the same, and they did not travel together. The rhetoric moved on its own.\nWhen I asked Zbaracki whether leaders knew they were performing a language they weren't practicing, his answer was no. He believed they wanted to believe. They didn't have the substance behind what they were doing, he told me, and in many cases had no way of accessing it. They were caught in the zeitgeist, using the vocabulary to establish an authority they weren't sure they had earned.\nThe gap between rhetoric and reality is rarely a gap between honest and dishonest people. It is a gap between people and structures. Words are cheap to adopt and practices are expensive to learn, so the language runs ahead, and the leader who has adopted it now has a stake in not being questioned about the substance.\nThe mechanism is plain. Jumping on the bandwagon is easy. Not jumping is hard. The incentives run one direction, cycle after cycle: TQM, then Six Sigma picking up the same themes under a different name, then the next term, and the next.\nWhen the Language Becomes Unquestionable\nA second move happens once the vocabulary is installed, and it is the more dangerous one. The language stops being a description and becomes a boundary. To question it is to mark yourself as someone who isn't keeping up.\nI put this to Zbaracki as a hypothesis: language gets installed partly to hide structural inadequacy, and once installed it becomes unquestionable. If an organization has staked itself on a word, raising your hand to ask what the word actually means reads as resistance, not curiosity. He didn't hesitate. \"Absolutely,\" he said. \"Without a doubt.\"\nThen he told me about the defense contractor. At the end of his study he gave a presentation telling one of the organizations, in effect, that they were not doing TQM. They never spoke to him again. He called it a rookie mistake. But inside that same organization, the true believers were glad to hear it said out loud. They had been frustrated for a long time. The refusal to talk to him wasn't a rebuttal. It was the preservation of a position.\nThat is the tell. When an outside question is met not with an argument but with silence and distance, the silence is the data. It signals that the language is doing load-bearing work the structure cannot do on its own, and that examining it too closely threatens something the organization has decided not to examine.\nWhy AI Is the Harder Case\nHere is where the parallel to AI both holds and breaks, and the break is the part leaders should consider.\nThe parallel holds on hype. Both arrived on a wave, and both waves were pushed. But the pushers are not the same. TQM had its gurus, and even among them a tension ran between the ones who understood the substance, Deming and Juran, and the ones riding the enthusiasm. Those people never held the kind of power now sitting behind AI. A figure like Sam Altman can generate hype at a scale the quality gurus could only have imagined.\nThe parallel breaks somewhere more important. With TQM, Zbaracki understood the tools well enough to walk into a room and quickly tell whether the people talking knew what they were talking about. He could see the substance. With generative AI, he told me, he cannot. He described sitting with capable people at large IT firms, enthusiastically describing how they set AI agents against one another, and having no way to judge the veracity of what they were saying. The tool no longer exposes its own substance to the person using it.\nHe drew a line most of the AI conversation skips. Predictive AI, to him, is not that different from TQM. There is a data scientist behind the model, the predictions can be tested against reality, and a domain expert can come to understand what the tool is doing. Netflix, dynamic pricing, much of medicine: predictive AI is legible to someone willing to do the hard work. Generative AI is the stranger case. The data used to build the model shapes the output profoundly, and we do not control that data, do not have access to it, and often cannot say what the tool is doing at all.\nSo the danger does not sit in the tool. It lies with the choices we make around it. That distinction, between the technology and the choices surrounding it, is exactly the distinction the rhetoric is built to blur.\nThe Gap, Now Measured\nFor most of the last two years this was a pattern you could feel in a room before you could prove it. Now the proof has arrived, and it describes Zbaracki's gap almost to the decimal.\nWhen MIT's NANDA initiative studied enterprise AI in 2025, it found that despite roughly $40 billion invested, about 95% of generative AI pilots delivered no measurable impact on the bottom line. The researchers named it directly: the divide between high adoption and low transformation. One manufacturing executive gave them a sentence that could have come from the TQM study: the hype says everything has changed, but in their operations, nothing fundamental had shifted. That is rhetoric and reality, thirty years later, in a single line.\nMy employer Gallup’s data shows the same split from the other side of the desk. Half of U.S. employees now use AI at work at least occasionally. But asked whether it has changed how the organization actually works, only 12% strongly agree that AI has transformed how work gets done where they are. The practice is real where a person can hold it in their hands and rhetorical everywhere above that. Meanwhile only 22% of employees say their organization has communicated a clear AI plan. \"We are an AI company\" gets asserted from the top of organizations where most people cannot confirm it from where they sit. The word arrived. The structure that would make it true did not.\nPull the Rhetoric Toward the Experience\nNone of this is an argument against AI. Zbaracki was careful about that. Things necessarily come in as rhetoric, he told me, because you do not yet have the experience; the concept arrives before the practice can. The question is not how to keep the language out. It is how to use it to pull people toward the experience, in ways that let them confront the inadequacy of their own words rather than defend them.\nThe first discipline is to make the tool expose its own substance. Before any initiative names what it will deliver, it should name what it actually does. For a predictive system that is answerable. For a generative system the honest answer is often \"we cannot fully say,\" and that admission belongs on the table rather than buried under use-case language. MIT found the pilots that survived were integrated into a real workflow and able to admit what they did not know, while the ones that died were the demos and use cases that looked flawless in the boardroom and collapsed in the field. The discipline is the plain question asked out loud: walk me through what this is doing and where the data came from.\nThe second is to check whether anything but the vocabulary has changed. The transformations that failed ran on single-loop learning: people changing what they said while leaving what they did untouched. A team can adopt the language of AI fluency while the approval layers, the incentives, and the question of who is allowed to dissent stay exactly where they were. When that happens, the rhetoric is running ahead of the reality, and it will keep running ahead until the structure moves.\nHow to Call It Out\nNotice what happens when someone asks a plain question. When \"what does this actually do?\" is met not with an answer but with a certain distance, a sense that the questioner is lagging behind, not future-forward, the distance is the answer. Substance survives an honest question. Rhetoric cannot, which is why it defends itself with the implication that asking is the problem.\nI have watched what that costs when no one names it. A company rolls out its AI transformation across the front line. On paper the case is airtight: faster decisions, leaner operations, better insight. Six months in, adoption has stalled. The managers cannot translate the change into language their teams can use, because no one translated it into language they could use first. At a rollout meeting, someone finally asks the plain question, \"what does this actually do?\", and the silence that follows drains the energy from the room. That silence is the whole story. It is the defense contractor declining to talk, relocated to a conference room where the people paying for the strategy are the ones left holding the silence.\nThere is a discipline the AI moment asks of leaders, and it is not fluency. It is the willingness to be the person in that rollout meeting who asks what the word means, and then sits in the silence long enough for everyone else to hear it too. A house of cards stands only as long as no one is allowed to breathe near it.\nThe organizations that will pay the highest price for this AI wave will not be the ones that moved too slowly. They will be the ones that mistook the vocabulary for the capability, staked themselves on a language they could not yet practice, and made sure no one was permitted to notice the difference, until the bill came due, and the words were all that was left.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 10475 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/34782","export_markdown":"/api/items/34782/export?format=markdown","export_json":"/api/items/34782/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/"},"formats":{"full":{"id":34782,"title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","source":"Forbes","author":null,"published_at":"2026-07-12T16:29:35+00:00","locale":"en","topic":"ai","tags":[],"excerpt":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management.","full_text":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. In the early nineties, that word was quality. Today it is AI. The vocabulary changes. The dynamic underneath it does not.\nMark Zbaracki noticed the dynamic thirty years ago. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management. What he found was not a story about quality. It was a story about language, about how a set of statistical practices got wrapped in a vocabulary that spread far faster than the practices themselves.\nI spoke with him recently because the gap looks familiar again. When we spoke, he had just reread his own paper to prepare a course on strategy implementation. Rereading your own work from three decades ago is its own kind of test. The dynamic held.\nThe Two Versions of Everything\nTQM existed in two versions at once. There was a technical version: control charts, statistical methods, the group practices that let people understand a production system in more depth. And there was a rhetorical version: the slogans, the presentations, the claim of transformation. The two were not the same, and they did not travel together. The rhetoric moved on its own.\nWhen I asked Zbaracki whether leaders knew they were performing a language they weren't practicing, his answer was no. He believed they wanted to believe. They didn't have the substance behind what they were doing, he told me, and in many cases had no way of accessing it. They were caught in the zeitgeist, using the vocabulary to establish an authority they weren't sure they had earned.\nThe gap between rhetoric and reality is rarely a gap between honest and dishonest people. It is a gap between people and structures. Words are cheap to adopt and practices are expensive to learn, so the language runs ahead, and the leader who has adopted it now has a stake in not being questioned about the substance.\nThe mechanism is plain. Jumping on the bandwagon is easy. Not jumping is hard. The incentives run one direction, cycle after cycle: TQM, then Six Sigma picking up the same themes under a different name, then the next term, and the next.\nWhen the Language Becomes Unquestionable\nA second move happens once the vocabulary is installed, and it is the more dangerous one. The language stops being a description and becomes a boundary. To question it is to mark yourself as someone who isn't keeping up.\nI put this to Zbaracki as a hypothesis: language gets installed partly to hide structural inadequacy, and once installed it becomes unquestionable. If an organization has staked itself on a word, raising your hand to ask what the word actually means reads as resistance, not curiosity. He didn't hesitate. \"Absolutely,\" he said. \"Without a doubt.\"\nThen he told me about the defense contractor. At the end of his study he gave a presentation telling one of the organizations, in effect, that they were not doing TQM. They never spoke to him again. He called it a rookie mistake. But inside that same organization, the true believers were glad to hear it said out loud. They had been frustrated for a long time. The refusal to talk to him wasn't a rebuttal. It was the preservation of a position.\nThat is the tell. When an outside question is met not with an argument but with silence and distance, the silence is the data. It signals that the language is doing load-bearing work the structure cannot do on its own, and that examining it too closely threatens something the organization has decided not to examine.\nWhy AI Is the Harder Case\nHere is where the parallel to AI both holds and breaks, and the break is the part leaders should consider.\nThe parallel holds on hype. Both arrived on a wave, and both waves were pushed. But the pushers are not the same. TQM had its gurus, and even among them a tension ran between the ones who understood the substance, Deming and Juran, and the ones riding the enthusiasm. Those people never held the kind of power now sitting behind AI. A figure like Sam Altman can generate hype at a scale the quality gurus could only have imagined.\nThe parallel breaks somewhere more important. With TQM, Zbaracki understood the tools well enough to walk into a room and quickly tell whether the people talking knew what they were talking about. He could see the substance. With generative AI, he told me, he cannot. He described sitting with capable people at large IT firms, enthusiastically describing how they set AI agents against one another, and having no way to judge the veracity of what they were saying. The tool no longer exposes its own substance to the person using it.\nHe drew a line most of the AI conversation skips. Predictive AI, to him, is not that different from TQM. There is a data scientist behind the model, the predictions can be tested against reality, and a domain expert can come to understand what the tool is doing. Netflix, dynamic pricing, much of medicine: predictive AI is legible to someone willing to do the hard work. Generative AI is the stranger case. The data used to build the model shapes the output profoundly, and we do not control that data, do not have access to it, and often cannot say what the tool is doing at all.\nSo the danger does not sit in the tool. It lies with the choices we make around it. That distinction, between the technology and the choices surrounding it, is exactly the distinction the rhetoric is built to blur.\nThe Gap, Now Measured\nFor most of the last two years this was a pattern you could feel in a room before you could prove it. Now the proof has arrived, and it describes Zbaracki's gap almost to the decimal.\nWhen MIT's NANDA initiative studied enterprise AI in 2025, it found that despite roughly $40 billion invested, about 95% of generative AI pilots delivered no measurable impact on the bottom line. The researchers named it directly: the divide between high adoption and low transformation. One manufacturing executive gave them a sentence that could have come from the TQM study: the hype says everything has changed, but in their operations, nothing fundamental had shifted. That is rhetoric and reality, thirty years later, in a single line.\nMy employer Gallup’s data shows the same split from the other side of the desk. Half of U.S. employees now use AI at work at least occasionally. But asked whether it has changed how the organization actually works, only 12% strongly agree that AI has transformed how work gets done where they are. The practice is real where a person can hold it in their hands and rhetorical everywhere above that. Meanwhile only 22% of employees say their organization has communicated a clear AI plan. \"We are an AI company\" gets asserted from the top of organizations where most people cannot confirm it from where they sit. The word arrived. The structure that would make it true did not.\nPull the Rhetoric Toward the Experience\nNone of this is an argument against AI. Zbaracki was careful about that. Things necessarily come in as rhetoric, he told me, because you do not yet have the experience; the concept arrives before the practice can. The question is not how to keep the language out. It is how to use it to pull people toward the experience, in ways that let them confront the inadequacy of their own words rather than defend them.\nThe first discipline is to make the tool expose its own substance. Before any initiative names what it will deliver, it should name what it actually does. For a predictive system that is answerable. For a generative system the honest answer is often \"we cannot fully say,\" and that admission belongs on the table rather than buried under use-case language. MIT found the pilots that survived were integrated into a real workflow and able to admit what they did not know, while the ones that died were the demos and use cases that looked flawless in the boardroom and collapsed in the field. The discipline is the plain question asked out loud: walk me through what this is doing and where the data came from.\nThe second is to check whether anything but the vocabulary has changed. The transformations that failed ran on single-loop learning: people changing what they said while leaving what they did untouched. A team can adopt the language of AI fluency while the approval layers, the incentives, and the question of who is allowed to dissent stay exactly where they were. When that happens, the rhetoric is running ahead of the reality, and it will keep running ahead until the structure moves.\nHow to Call It Out\nNotice what happens when someone asks a plain question. When \"what does this actually do?\" is met not with an answer but with a certain distance, a sense that the questioner is lagging behind, not future-forward, the distance is the answer. Substance survives an honest question. Rhetoric cannot, which is why it defends itself with the implication that asking is the problem.\nI have watched what that costs when no one names it. A company rolls out its AI transformation across the front line. On paper the case is airtight: faster decisions, leaner operations, better insight. Six months in, adoption has stalled. The managers cannot translate the change into language their teams can use, because no one translated it into language they could use first. At a rollout meeting, someone finally asks the plain question, \"what does this actually do?\", and the silence that follows drains the energy from the room. That silence is the whole story. It is the defense contractor declining to talk, relocated to a conference room where the people paying for the strategy are the ones left holding the silence.\nThere is a discipline the AI moment asks of leaders, and it is not fluency. It is the willingness to be the person in that rollout meeting who asks what the word means, and then sits in the silence long enough for everyone else to hear it too. A house of cards stands only as long as no one is allowed to breathe near it.\nThe organizations that will pay the highest price for this AI wave will not be the ones that moved too slowly. They will be the ones that mistook the vocabulary for the capability, staked themselves on a language they could not yet practice, and made sure no one was permitted to notice the difference, until the bill came due, and the words were all that was left.","reading_time_min":9,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 10475 characters.","diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}},"actions":{"read":"/item/34782","export_markdown":"/api/items/34782/export?format=markdown","export_json":"/api/items/34782/export?format=json","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/"}},"digest":{"id":34782,"title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","source":"Forbes","topic":"ai","published_at":"2026-07-12T16:29:35+00:00","excerpt":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 10475 characters.","reading_time_min":9,"cluster_id":null},"card":{"display_title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","subtitle":"Forbes · 2026-07-12","summary":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and…","badges":["quality:high"],"links":{"read":"/item/34782","original":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","diagnose":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/"},"quality_warning":null},"export":{"title":"Why AI Pilots Fail: A 1998 Paper Might Have Seen It Coming - Forbes","url":"https://www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","summary":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management.","source":"Forbes","date":"2026-07-12T16:29:35+00:00","content":"Every few years a word arrives that everyone agrees to use before anyone agrees on what it means. In the early nineties, that word was quality. Today it is AI. The vocabulary changes. The dynamic underneath it does not.\nMark Zbaracki noticed the dynamic thirty years ago. His 1998 paper, \"The Rhetoric and Reality of Total Quality Management,\" published in Administrative Science Quarterly and later named the best paper of its year in 1999, studied five organizations that had all adopted Total Quality Management. What he found was not a story about quality. It was a story about language, about how a set of statistical practices got wrapped in a vocabulary that spread far faster than the practices themselves.\nI spoke with him recently because the gap looks familiar again. When we spoke, he had just reread his own paper to prepare a course on strategy implementation. Rereading your own work from three decades ago is its own kind of test. The dynamic held.\nThe Two Versions of Everything\nTQM existed in two versions at once. There was a technical version: control charts, statistical methods, the group practices that let people understand a production system in more depth. And there was a rhetorical version: the slogans, the presentations, the claim of transformation. The two were not the same, and they did not travel together. The rhetoric moved on its own.\nWhen I asked Zbaracki whether leaders knew they were performing a language they weren't practicing, his answer was no. He believed they wanted to believe. They didn't have the substance behind what they were doing, he told me, and in many cases had no way of accessing it. They were caught in the zeitgeist, using the vocabulary to establish an authority they weren't sure they had earned.\nThe gap between rhetoric and reality is rarely a gap between honest and dishonest people. It is a gap between people and structures. Words are cheap to adopt and practices are expensive to learn, so the language runs ahead, and the leader who has adopted it now has a stake in not being questioned about the substance.\nThe mechanism is plain. Jumping on the bandwagon is easy. Not jumping is hard. The incentives run one direction, cycle after cycle: TQM, then Six Sigma picking up the same themes under a different name, then the next term, and the next.\nWhen the Language Becomes Unquestionable\nA second move happens once the vocabulary is installed, and it is the more dangerous one. The language stops being a description and becomes a boundary. To question it is to mark yourself as someone who isn't keeping up.\nI put this to Zbaracki as a hypothesis: language gets installed partly to hide structural inadequacy, and once installed it becomes unquestionable. If an organization has staked itself on a word, raising your hand to ask what the word actually means reads as resistance, not curiosity. He didn't hesitate. \"Absolutely,\" he said. \"Without a doubt.\"\nThen he told me about the defense contractor. At the end of his study he gave a presentation telling one of the organizations, in effect, that they were not doing TQM. They never spoke to him again. He called it a rookie mistake. But inside that same organization, the true believers were glad to hear it said out loud. They had been frustrated for a long time. The refusal to talk to him wasn't a rebuttal. It was the preservation of a position.\nThat is the tell. When an outside question is met not with an argument but with silence and distance, the silence is the data. It signals that the language is doing load-bearing work the structure cannot do on its own, and that examining it too closely threatens something the organization has decided not to examine.\nWhy AI Is the Harder Case\nHere is where the parallel to AI both holds and breaks, and the break is the part leaders should consider.\nThe parallel holds on hype. Both arrived on a wave, and both waves were pushed. But the pushers are not the same. TQM had its gurus, and even among them a tension ran between the ones who understood the substance, Deming and Juran, and the ones riding the enthusiasm. Those people never held the kind of power now sitting behind AI. A figure like Sam Altman can generate hype at a scale the quality gurus could only have imagined.\nThe parallel breaks somewhere more important. With TQM, Zbaracki understood the tools well enough to walk into a room and quickly tell whether the people talking knew what they were talking about. He could see the substance. With generative AI, he told me, he cannot. He described sitting with capable people at large IT firms, enthusiastically describing how they set AI agents against one another, and having no way to judge the veracity of what they were saying. The tool no longer exposes its own substance to the person using it.\nHe drew a line most of the AI conversation skips. Predictive AI, to him, is not that different from TQM. There is a data scientist behind the model, the predictions can be tested against reality, and a domain expert can come to understand what the tool is doing. Netflix, dynamic pricing, much of medicine: predictive AI is legible to someone willing to do the hard work. Generative AI is the stranger case. The data used to build the model shapes the output profoundly, and we do not control that data, do not have access to it, and often cannot say what the tool is doing at all.\nSo the danger does not sit in the tool. It lies with the choices we make around it. That distinction, between the technology and the choices surrounding it, is exactly the distinction the rhetoric is built to blur.\nThe Gap, Now Measured\nFor most of the last two years this was a pattern you could feel in a room before you could prove it. Now the proof has arrived, and it describes Zbaracki's gap almost to the decimal.\nWhen MIT's NANDA initiative studied enterprise AI in 2025, it found that despite roughly $40 billion invested, about 95% of generative AI pilots delivered no measurable impact on the bottom line. The researchers named it directly: the divide between high adoption and low transformation. One manufacturing executive gave them a sentence that could have come from the TQM study: the hype says everything has changed, but in their operations, nothing fundamental had shifted. That is rhetoric and reality, thirty years later, in a single line.\nMy employer Gallup’s data shows the same split from the other side of the desk. Half of U.S. employees now use AI at work at least occasionally. But asked whether it has changed how the organization actually works, only 12% strongly agree that AI has transformed how work gets done where they are. The practice is real where a person can hold it in their hands and rhetorical everywhere above that. Meanwhile only 22% of employees say their organization has communicated a clear AI plan. \"We are an AI company\" gets asserted from the top of organizations where most people cannot confirm it from where they sit. The word arrived. The structure that would make it true did not.\nPull the Rhetoric Toward the Experience\nNone of this is an argument against AI. Zbaracki was careful about that. Things necessarily come in as rhetoric, he told me, because you do not yet have the experience; the concept arrives before the practice can. The question is not how to keep the language out. It is how to use it to pull people toward the experience, in ways that let them confront the inadequacy of their own words rather than defend them.\nThe first discipline is to make the tool expose its own substance. Before any initiative names what it will deliver, it should name what it actually does. For a predictive system that is answerable. For a generative system the honest answer is often \"we cannot fully say,\" and that admission belongs on the table rather than buried under use-case language. MIT found the pilots that survived were integrated into a real workflow and able to admit what they did not know, while the ones that died were the demos and use cases that looked flawless in the boardroom and collapsed in the field. The discipline is the plain question asked out loud: walk me through what this is doing and where the data came from.\nThe second is to check whether anything but the vocabulary has changed. The transformations that failed ran on single-loop learning: people changing what they said while leaving what they did untouched. A team can adopt the language of AI fluency while the approval layers, the incentives, and the question of who is allowed to dissent stay exactly where they were. When that happens, the rhetoric is running ahead of the reality, and it will keep running ahead until the structure moves.\nHow to Call It Out\nNotice what happens when someone asks a plain question. When \"what does this actually do?\" is met not with an answer but with a certain distance, a sense that the questioner is lagging behind, not future-forward, the distance is the answer. Substance survives an honest question. Rhetoric cannot, which is why it defends itself with the implication that asking is the problem.\nI have watched what that costs when no one names it. A company rolls out its AI transformation across the front line. On paper the case is airtight: faster decisions, leaner operations, better insight. Six months in, adoption has stalled. The managers cannot translate the change into language their teams can use, because no one translated it into language they could use first. At a rollout meeting, someone finally asks the plain question, \"what does this actually do?\", and the silence that follows drains the energy from the room. That silence is the whole story. It is the defense contractor declining to talk, relocated to a conference room where the people paying for the strategy are the ones left holding the silence.\nThere is a discipline the AI moment asks of leaders, and it is not fluency. It is the willingness to be the person in that rollout meeting who asks what the word means, and then sits in the silence long enough for everyone else to hear it too. A house of cards stands only as long as no one is allowed to breathe near it.\nThe organizations that will pay the highest price for this AI wave will not be the ones that moved too slowly. They will be the ones that mistook the vocabulary for the capability, staked themselves on a language they could not yet practice, and made sure no one was permitted to notice the difference, until the bill came due, and the words were all that was left.","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//www.forbes.com/sites/vibhasratanjee/2026/07/12/why-ai-pilots-fail-a-1998-paper-might-have-seen-it-coming/","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 10475 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 10475 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":10475,"summary_length":341,"usable_text_length":10475,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":10475,"summary_length":341}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}