[{"data":1,"prerenderedAt":590},["ShallowReactive",2],{"post-\u002Fhow-can-companies-create-value-from-ai":3},{"page":4,"translations":444,"nav":452,"related":572,"random":581},{"id":5,"title":6,"body":7,"categories":414,"category":417,"changeHistory":417,"date":418,"description":419,"disclosures":420,"draft":423,"extension":424,"firstLiveAt":417,"image":425,"imageAlt":426,"kind":427,"lang":428,"meta":429,"navigation":430,"omitGermanLocalizationDisclosure":423,"path":431,"publishedAt":417,"readingTime":432,"rights":417,"seo":433,"seoTitle":6,"slug":434,"sources":417,"stem":434,"tags":435,"translationKey":441,"type":442,"updated":417,"__hash__":443},"posts\u002Fhow-can-companies-create-value-from-ai.md","How Can Companies Create Value From AI?",{"type":8,"value":9,"toc":403},"minimark",[10,43,46,61,64,72,75,78,88,93,96,109,112,115,119,122,189,192,195,203,207,210,218,228,231,239,251,254,257,261,264,267,279,282,297,301,304,307,316,319,322,326,329,332,335,343,346,354,357,361,364,397,400],[11,12,13,21],"blockquote",{},[14,15,16,17],"p",{},"💡 ",[18,19,20],"strong",{},"TL;DR: Key Takeaways",[22,23,24,31,37],"ul",{},[25,26,27,30],"li",{},[18,28,29],{},"Model capability, task results, operating capability, and business value are not the same thing."," Success at one level does not prove that the company has reached the next.",[25,32,33,36],{},[18,34,35],{},"Company-specific value is often created around the model."," Data, process, authority, system integrations, expertise, and measurement turn a widely available model into an operating capability.",[25,38,39,42],{},[18,40,41],{},"AI can create value without reducing headcount."," At the company where I work, it has improved speed, efficiency, quality, and management capability. It made some capabilities economically feasible when relying on human labor alone would not have been.",[14,44,45],{},"A company can easily count how many AI licenses it has bought, how many pilot projects it has launched, and how many AI agents it has built. The harder questions are different. Which systems helped customers complete their work faster? Which allowed the same team to handle more work at the same quality? Which produced a visible change in cost, revenue, capacity, quality, or risk?",[14,47,48,49,60],{},"A ",[50,51,59],"a",{"href":52,"rel":53,"target":56,"className":57},"https:\u002F\u002Fwww.atlantafed.org\u002Fresearch-and-data\u002Fpublications\u002Fworking-papers\u002F2026\u002F03\u002F24\u002F03-firm-data-on-ai",[54,55],"nofollow","noopener","_blank",[58],"dofollow","representative firm survey"," published in 2026 covered roughly 6,000 executives in the United States, the United Kingdom, Germany, and Australia. About 70% of their companies reported using AI. Yet more than 80% said they had seen no effect on productivity or employment over the previous three years. The same executives expected AI to improve productivity over the next three.",[14,62,63],{},"That aggregate picture does not match what I have observed at the company where I work. Every AI project we have run has improved speed, efficiency, quality, and management capability. AI did not reduce headcount. But achieving the same quality and capabilities through human labor alone would have required a larger team and investments we could not have afforded.",[14,65,66,71],{},[50,67,70],{"href":68,"rel":69,"target":56},"\u002Fbeyond-the-bot-lessons-from-building-a-chat-system-for-global-patients",[55],"VaniBot, which I have written about in detail",", is one of the simplest examples we can discuss publicly. It supports pre-consultation communication and structured information gathering, then hands the conversation to the team when needed. It replaced no one. It reduced repetitive communication work and created more room for tasks that require context and judgment. We have other projects that delivered much greater gains in speed, efficiency, and management capability, but I am not sharing details that are not already public.",[14,73,74],{},"One company's experience does not invalidate a representative survey. It does show why reading employment impact only as a reduction in headcount is incomplete. Value can also mean making a level of quality or management capability possible within the current budget when it could not previously be built.",[14,76,77],{},"The survey describes the aggregate picture. This experience describes live systems. Read together, they show that adoption, productivity, headcount, and the capabilities a company gains are not the same measure. The gap between using AI and creating value from it begins there.",[14,79,80,81,87],{},"The same distinction appears in George Westerman's ",[50,82,86],{"href":83,"rel":84,"target":56,"className":85},"https:\u002F\u002Fmitsloan.mit.edu\u002Fideas-made-to-matter\u002F6-questions-to-guide-your-ai-strategy",[54,55],[58],"six-question AI strategy framework",", published by MIT Sloan on August 3, 2026. Westerman treats shared ambition, governance, scaling, the technology foundation, culture, and employee skills as parts of the same strategy. The point is not only to choose a capable model, but to define how the company will work differently and sustain that change.",[89,90,92],"h2",{"id":91},"ai-does-improve-productivity-in-some-tasks","AI Does Improve Productivity in Some Tasks",[14,94,95],{},"The gains I have seen in practice also have measured counterparts in other workflows.",[14,97,98,99,108],{},"In ",[50,100,104],{"href":101,"rel":102,"target":56,"className":103},"https:\u002F\u002Fwww.nber.org\u002Fpapers\u002Fw31161",[54,55],[58],[105,106,107],"em",{},"Generative AI at Work",", researchers studied more than 5,000 customer support workers using an AI assistant. The number of issues resolved per hour increased by roughly 14% on average. The gains were larger for less experienced workers and far smaller for experienced workers.",[14,110,111],{},"That is a measured productivity gain among real workers in a defined workflow. It is also one customer support operation at one company. We cannot treat 14% as a universal return on AI. The study does not establish how much of the faster task performance became profit or a durable competitive advantage.",[14,113,114],{},"Completing a task faster is still valuable. Turning that gain into capacity, lower cost, or a better customer outcome requires several more transitions.",[89,116,118],{"id":117},"four-transitions-from-model-capability-to-business-value","Four Transitions From Model Capability to Business Value",[14,120,121],{},"It is more useful to evaluate an AI investment at four separate levels than to compress it into one success rate:",[123,124,125,141],"table",{},[126,127,128],"thead",{},[129,130,131,135,138],"tr",{},[132,133,134],"th",{},"Level",[132,136,137],{},"Question",[132,139,140],{},"Appropriate evidence",[142,143,144,156,167,178],"tbody",{},[129,145,146,150,153],{},[147,148,149],"td",{},"Model capability",[147,151,152],{},"Can the model perform the task in controlled examples?",[147,154,155],{},"Test cases, expert review, and error categories",[129,157,158,161,164],{},[147,159,160],{},"Task result",[147,162,163],{},"Does the user or system complete the real task faster or better?",[147,165,166],{},"Time, quality, completion, and error measures",[129,168,169,172,175],{},[147,170,171],{},"Operating capability",[147,173,174],{},"Can the company produce that result repeatedly with the right data, authority, and process?",[147,176,177],{},"System-of-record data, production performance, exceptions, and recovery",[129,179,180,183,186],{},[147,181,182],{},"Business value",[147,184,185],{},"Does that capability change cost, revenue, capacity, quality, risk, or the customer outcome?",[147,187,188],{},"A baseline, a suitable comparison, and enough observation time",[14,190,191],{},"Success at one level may be necessary for the next, but it is not sufficient. A model performing well in a test does not prove that the company can produce the same result reliably in a live system.",[14,193,194],{},"A model may correctly identify a return request in a customer message. An AI-assisted representative may process that request faster. The company's operating capability must also find the correct order, apply the current policy, check authority, execute the refund safely, and recover when something fails. Business value appears only when that arrangement changes customer retention, processing cost, resolution time, or another intended result.",[14,196,197,202],{},[50,198,201],{"href":199,"rel":200,"target":56},"\u002Fis-your-ai-system-delivering-business-results",[55],"In my article on evaluating whether an AI system actually works",", I argued that model output, the path through the system, completed work, and business impact require separate evidence. The same distinction applies here. The easiest level to measure should not become a substitute for the company's actual objective.",[89,204,206],{"id":205},"access-to-the-same-model-is-not-the-same-capability","Access to the Same Model Is Not the Same Capability",[14,208,209],{},"Many companies can now buy enterprise access to powerful models or connect those models to their own systems. That access matters. A model that competitors can buy as well, however, does not create a company-specific advantage on its own.",[14,211,212,217],{},[50,213,216],{"href":214,"rel":215,"target":56},"\u002Fstart-with-the-business-problem-not-the-ai-model",[55],"I have previously explained why an AI project should not begin with model selection",". The difference between a model and an operating capability is created by the company's trusted data, business rules, existing systems, user permissions, expert judgment, and feedback mechanisms.",[14,219,220,221,227],{},"An ",[50,222,226],{"href":223,"rel":224,"target":56,"className":225},"https:\u002F\u002Fwww.oecd.org\u002Fen\u002Fpublications\u002Fa-portrait-of-ai-adopters-across-countries_0fb79bb9-en.html",[54,55],[58],"OECD study of AI adopters"," applied a harmonized method to official firm surveys in 11 countries. It found that AI adopters were also more likely to have supporting assets such as digital infrastructure, other digital technologies, and information and communication skills.",[14,229,230],{},"That association does not prove that AI caused higher productivity. Firms that were already more productive may also have been better able to invest in both AI and those complementary capabilities. Still, it points to the problem with treating the technology separately from the company that must operate it.",[14,232,233,238],{},[50,234,237],{"href":235,"rel":236,"target":56},"\u002Fwhat-makes-an-ai-system-specific-to-your-business",[55],"In my article on what makes an AI system specific to a business",", I explain that not every company-specific requirement belongs inside the model. Current information may live in a system of record, explicit rules in software, and transaction authority in an access layer. Those parts should remain with the company even when the model changes.",[14,240,241,242,250],{},"The ",[50,243,246,249],{"href":244,"rel":245,"target":56},"https:\u002F\u002Fdigitaleconomy.stanford.edu\u002Fpublication\u002Fenterprise-ai-playbook\u002F",[54,55],[105,247,248],{},"Enterprise AI Playbook"," report from Stanford Digital Economy Lab"," examines 51 production deployments across 41 organizations. It found that, in 77% of those cases, the hardest challenges were invisible costs such as change management, data quality, and process redesign rather than the model itself.",[14,252,253],{},"That supports the distinction in this article between model capability and operating capability. But the sample contains only deployments that reached measurable value, so it should not be read as a general success rate.",[14,255,256],{},"Two objectives need to be separated. A company may use AI simply to maintain cost or service parity with competitors. That is economic value too. Not every investment needs to produce a unique competitive advantage. But if the objective is differentiation, access to a common model is not enough. The difference lies in how the model works with company-specific data and processes, and in how well the company operates the resulting system.",[89,258,260],{"id":259},"prototypes-became-cheaper-production-did-not-become-free","Prototypes Became Cheaper. Production Did Not Become Free",[14,262,263],{},"It is now easy to give a model a few documents and assemble a convincing prototype. That ease can create the impression that a production system will be equally fast and inexpensive to build.",[14,265,266],{},"Production brings different questions. Did the model use the current source? Was the user authorized to see the record? Did the transaction actually happen? Will a retry create it twice? When something fails, will the system stop, reverse the action, or hand the work to a specialist? How will quality be compared when the model or provider changes?",[14,268,269,270,278],{},"MIT CISR's August 2026 research briefing, ",[50,271,275],{"href":272,"rel":273,"target":56,"className":274},"https:\u002F\u002Fcisr.mit.edu\u002Fpublication\u002F2026_0801_AIValuePropositions_Woerner",[54,55],[58],[105,276,277],{},"AI Value Creation: Five Provocative Propositions",", argues that trust must be built into infrastructure. It also proposes that the number of AI agents a company has built will eventually be unimportant. Durable value, in its view, will depend on the ability to put these systems into production and scale them.",[14,280,281],{},"These propositions do not come from a representative experiment or an industry average. MIT CISR presents them as discussion propositions drawn from company interviews, cases, and ongoing research. The operational needs behind them are still concrete: evaluation, monitoring, verification in the system of record, authorization, appeals, reversibility, and recovery after failure.",[14,283,284,285,290,291,296],{},"That is why I have treated ",[50,286,289],{"href":287,"rel":288,"target":56},"\u002Fhow-much-authority-should-ai-have",[55],"the authority given to AI systems"," and ",[50,292,295],{"href":293,"rel":294,"target":56},"\u002Fwhen-do-you-actually-need-an-ai-agent",[55],"the choice between a workflow and an AI agent"," as business design questions rather than technical details. The number of agents built may be an activity measure. It is not a value measure if the system cannot operate reliably or prove its results.",[89,298,300],{"id":299},"distributing-knowledge-without-losing-expertise","Distributing Knowledge Without Losing Expertise",[14,302,303],{},"The customer support study found the largest gains among less experienced workers. AI can distribute the methods used by experienced people across a wider team, making existing organizational knowledge easier to access.",[14,305,306],{},"But expertise is not merely access to an existing answer. It also means recognizing a wrong answer, identifying an exception, developing a rule when a new situation appears, and producing the knowledge that others will use.",[14,308,48,309,315],{},[50,310,314],{"href":311,"rel":312,"target":56,"className":313},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2601.20245",[54,55],[58],"narrow programming experiment"," published in 2026 involved 52 participants. The group using AI scored 17% lower on a conceptual assessment after the task. The experiment found no statistically significant reduction in average completion time.",[14,317,318],{},"The paper has not yet completed peer review. It covers one programming task and a small sample. It does not show that AI reduces expertise across all occupations. It offers a limited warning that how AI is used may affect learning as well as immediate output.",[14,320,321],{},"This makes MIT CISR's proposition about the loss of existing expertise worth taking seriously. If companies teach people only how to use AI tools without preserving the domain knowledge needed to challenge their output, they may trade future verification capacity for short-term speed.",[89,323,325],{"id":324},"can-the-company-capture-the-time-it-saves","Can the Company Capture the Time It Saves?",[14,327,328],{},"If a task takes ten minutes instead of twenty, there is a ten-minute benefit. Its value to the business depends on what happens to that time.",[14,330,331],{},"Did the same team resolve more requests at the same quality? Did overtime fall? Did customers wait less? Could employees use the freed capacity for more valuable work? Or did a few minutes disappear across different parts of the day without changing the team's workload or objectives?",[14,333,334],{},"The real question is whether the benefit reached a company outcome. Not every improvement that makes work easier has to appear directly as profit. Lower error risk, more consistent service, and less employee strain can all be meaningful outcomes. But if the intended value is not defined at the start, any positive signal found after the project can be presented as proof of success.",[14,336,337,342],{},[50,338,341],{"href":339,"rel":340,"target":56},"\u002Fdo-you-know-how-dependent-your-company-is-on-ai",[55],"I have separately examined how these new capabilities can create a dependency in the company's operating model",".",[14,344,345],{},"The effect of an AI investment may also appear with a delay. Redesigning processes, correcting data, and adapting to a new way of working take time. That delay does not make measurement unnecessary. The company should state which intermediate measure is expected to change, when it should change, and under what conditions the investment decision will be reconsidered.",[14,347,241,348,353],{},[50,349,352],{"href":350,"rel":351,"target":56},"\u002Fdesigning-ai-systems-for-business",[55],"Enterprise Intelligence Architecture working model"," examines an AI system through the business outcome, decomposition of the work, allocation of responsibilities, controlled access and action, and evidence.",[14,355,356],{},"The five questions below are not a separate evaluation framework. They are that working model applied to an investment decision.",[89,358,360],{"id":359},"five-questions-for-evaluating-an-ai-investment","Five Questions for Evaluating an AI Investment",[14,362,363],{},"Before expanding an AI system, write down five things:",[365,366,367,373,379,385,391],"ol",{},[25,368,369,372],{},[18,370,371],{},"What is the baseline?"," How are time, quality, cost, capacity, risk, or customer outcomes measured today?",[25,374,375,378],{},[18,376,377],{},"What will prove that the work was completed?"," Beyond the model's response, which record will change in which system of record?",[25,380,381,384],{},[18,382,383],{},"Where will the return appear?"," How will saved time become capacity, lower cost, better quality, revenue, or lower risk?",[25,386,387,390],{},[18,388,389],{},"What is missing before production?"," How will the company provide data, integrations, authorization, expertise, evaluation, and recovery after failure?",[25,392,393,396],{},[18,394,395],{},"What result will make us continue, change, or stop?"," What is the success threshold, observation period, and unacceptable cost of error?",[14,398,399],{},"These questions do not make the model irrelevant. The model directly affects what the system can do, its cost, and its quality limits. Value appears when that capability is connected to the company's actual work.",[14,401,402],{},"AI adoption will probably continue to spread. What separates companies will not be only the number of licenses they bought or AI agents they built. A more useful question is this: What result did they prove, in which system, over what period, and what did they do with the value they created?",{"title":404,"searchDepth":405,"depth":405,"links":406},"",2,[407,408,409,410,411,412,413],{"id":91,"depth":405,"text":92},{"id":117,"depth":405,"text":118},{"id":205,"depth":405,"text":206},{"id":259,"depth":405,"text":260},{"id":299,"depth":405,"text":300},{"id":324,"depth":405,"text":325},{"id":359,"depth":405,"text":360},[415,416],"ai","business",null,"2026-09-04","When does using AI become business value? I examine the links between model capability, completed work, operating capability, and measurable results.",{"aiUse":421,"aiNote":422},"ai-assisted","This article is based on Evren Bal’s views and experience. AI-assisted tools were used during the research and editorial development process.",false,"md","\u002Fimages\u002Fhero\u002Fai-business-value-capability-threshold.avif","An intact team crosses a lowered threshold toward better quality, capacity, and management visibility.","Analysis","en",{},true,"\u002Fhow-can-companies-create-value-from-ai",11,{"title":6,"description":419},"how-can-companies-create-value-from-ai",[436,437,438,439,440],"ai-strategy","ai-roi","business-capability","digital-transformation","enterprise-ai","from-ai-use-to-business-value","post","ccciOGxY4UpEcQXlQ9qy6yzu0tofm1tYrcmn15kEEuY",{"en":445,"tr":446,"de":449},{"path":431,"title":6},{"path":447,"title":448},"\u002Ftr\u002Fai-kullanmak-ile-ai-dan-deger-uretmek-arasinda-ne-var","Şirketler Yapay Zekâdan Nasıl Değer Üretebilir?",{"path":450,"title":451},"\u002Fde\u002Fwie-koennen-unternehmen-mit-ki-wert-schaffen","Wie können Unternehmen mit KI Wert schaffen?",{"prev":453,"next":417,"others":455,"lucky":571,"readingTime":432},{"path":235,"title":454},"What Makes an AI System Specific to Your Business?",[456,458,461,464,467,470,473,476,479,482,485,488,491,493,496,499,502,505,508,511,514,517,520,523,526,528,531,534,537,540,543,544,547,550,553,556,559,562,565,568],{"path":339,"title":457},"Do You Know How Dependent Your Company Is on AI?",{"path":459,"title":460},"\u002Fhow-to-route-requests-across-multiple-ai-models","How Should You Route Requests Across Multiple AI Models?",{"path":462,"title":463},"\u002Fwhat-does-the-eu-ai-act-actually-regulate","What Is the EU AI Act, and What Does It Regulate?",{"path":465,"title":466},"\u002Fdo-ai-visibility-tools-really-work","Do AI Visibility Tools Really Work? What They Actually Measure",{"path":468,"title":469},"\u002Fwordpress-to-nuxt-ai-powered-content-pipeline","From WordPress to Nuxt: Building an AI-Powered Content Pipeline",{"path":471,"title":472},"\u002Fhow-i-built-a-modern-infrastructure-using-open-source-tools-and-the-power-of-cloudflare","Why I Run camiler.org on a Single VPS",{"path":474,"title":475},"\u002Fraising-children-in-the-age-of-artificial-intelligence","Raising Children in the Age of Artificial Intelligence",{"path":477,"title":478},"\u002Fai-assisted-rest-api-development","Preserving API Quality in AI-Assisted Development",{"path":480,"title":481},"\u002Fthe-threshold-collapsed-to-zero","The Threshold Collapsed: What ProductLog Taught Me About Building in Public",{"path":483,"title":484},"\u002Fai-made-code-cheap-verification-is-still-expensive","AI Made Code Cheap. Verification Is Still Expensive.",{"path":486,"title":487},"\u002Fbank-account-api-integration","Integrating One Bank Is Easy. Keeping Forty Running Is Not.",{"path":489,"title":490},"\u002Faccessing-know-how-is-not-the-same-as-creating-it","Accessing Know-How Is Not the Same as Creating It",{"path":293,"title":492},"When Do You Actually Need an AI Agent?",{"path":494,"title":495},"\u002Fthe-end-of-coding-or-a-new-renaissance-the-invisible-crisis-of-ai","The End of Coding or a New Renaissance? The Invisible Crisis of AI",{"path":497,"title":498},"\u002Fai-visibility-illusion-bing-citation-share","The AI Visibility Illusion: What Bing's Citation Share Data Actually Reveals",{"path":500,"title":501},"\u002Fgoogle-generative-ai-data-ai-citation-timing","AI Visibility Dropped Before Search: The Google Data That Changed My Theory",{"path":503,"title":504},"\u002Fthe-job-ai-wont-take-and-the-five-it-prevents","AI Is Reducing Hiring Without Layoffs",{"path":506,"title":507},"\u002Fredar-ai-powered-summaries-for-kap-disclosures-and-open-sources","Redar: AI-Powered Summaries for KAP Disclosures and Open Sources",{"path":509,"title":510},"\u002Fan-seo-experiment-in-a-low-competition-serp-with-google-maps-and-openai","Building camiler.org: A Programmatic SEO Experiment with Google Maps and OpenAI",{"path":512,"title":513},"\u002Fai-turned-a-refactor-i-wouldnt-do-into-a-one-hour-job","With AI, a Refactor I Wouldn't Bother With Took One Hour",{"path":515,"title":516},"\u002Fdoes-enterprise-ai-really-need-fine-tuning","Does Enterprise AI Really Need Fine-Tuning?",{"path":518,"title":519},"\u002Fkeeping-customers-happy-isnt-enough","Keeping Customers Happy Isn’t Enough. You Have to Follow Up.",{"path":521,"title":522},"\u002Ftesting-a-button-treating-an-entire-website-redesign-as-a-sure-thing","Testing a Button, Treating an Entire Website Redesign as a Sure Thing",{"path":524,"title":525},"\u002Fbuild-in-public-2-0","Build in Public in the AI Era: What to Share and What to Keep Private",{"path":214,"title":527},"Start With the Business Problem, Not the AI Model",{"path":529,"title":530},"\u002Fthe-ai-productivity-baseline-is-moving-faster-than-we-remember","AI Wasn’t Always This Good. We Just Got Used to It.",{"path":532,"title":533},"\u002Fturkeys-first-real-time-mystery-shopping-reporting","Turkey's First Real-Time Mystery Shopping Reporting Platform",{"path":535,"title":536},"\u002Fis-your-data-ready-for-ai","Is Your Data Ready for AI? Start With the Decision It Must Support",{"path":538,"title":539},"\u002Fwhen-does-enterprise-ai-need-rag","When Does Enterprise AI Actually Need RAG?",{"path":541,"title":542},"\u002Fthe-era-of-the-previous-vibe-coder-begins","The Era of the \"Previous Vibe Coder\" Begins: The Invisibility of Clean Code and the Technical Debt Bill of AI",{"path":235,"title":454},{"path":545,"title":546},"\u002Fllms-txt-was-never-the-point","llms.txt Was Never the Point",{"path":548,"title":549},"\u002Fhow-to-do-content-pruning-a-real-world-case-study","How to Do Content Pruning: A Real-World Case Study",{"path":551,"title":552},"\u002Fcomprehension-debt-the-bill-comes-due-alone","Comprehension Debt: The Bill Comes Due Alone",{"path":554,"title":555},"\u002Fwhen-process-automation-actually-needs-ai","When Does Process Automation Actually Need AI?",{"path":557,"title":558},"\u002Fwho-captures-ai-productivity-gains","Who Captures the Productivity Gains from AI?",{"path":560,"title":561},"\u002Ffrom-asking-questions-to-delegating-work","From Asking Questions to Delegating Work: What the Codex Study Shows",{"path":563,"title":564},"\u002Fone-victory-several-defeats","One Victory, Several Defeats",{"path":566,"title":567},"\u002Fwhy-im-building-rankextension-making-google-search-console-actually-make-sense","What Happened to RankExtension?",{"path":569,"title":570},"\u002Fwhy-sales-and-other-departments-keep-clashing","Why Sales and Other Departments Keep Clashing",{"path":518,"title":519},[573,575,577,579],{"path":235,"title":454,"date":574},"2026-09-03",{"path":535,"title":536,"date":576},"2026-08-31",{"path":214,"title":527,"date":578},"2026-08-27",{"path":459,"title":460,"date":580},"2026-09-02",[582,586,588],{"path":583,"title":584,"date":585},"\u002Fwhen-is-one-ai-model-enough","When Is One AI Model Enough?","2026-09-01",{"path":489,"title":490,"date":587},"2026-08-15",{"path":566,"title":567,"date":589},"2025-06-05",1788481883488]