The last moat of humanity that can never be trained, even if smart AI models swallow it all up: a value with history
Date
Jun 28, 2026
Classification
- AI/IT/Productivity
[Advertisement]
Things that cannot be trained
#AIDespair #ThinRapper #UntrainableRealm #TrustAsset #AuthorityAndResponsibility
🫑 3-Line Summary
•
While despair is spreading in the AI industry by mid-2026 that AI will do everything better and absorb all services built upon it, this applies only to measurable areas.
•
While the coding agent Devin has made writing code incredibly cheap by raising benchmark task completion rates to the high 80 percent range in just a year and a half, the amount of code deployed in the field has increased by only 30 percent, demonstrating that human review remains a key bottleneck.
•
Tasks that can be scored and measured on public leaderboards eventually become subjects for AI learning and are reduced to ordinary public goods, but AI can never encroach upon untrainable domains such as private answers, trust, authority, and responsibility within a company.
🥦 Insight
•
The Weight of Authority and Responsibility Becomes More Precious as Intelligence Becomes Cheaper
With the advancement of technology, the cost of renting smart intelligence is approaching near zero. However, no matter how advanced AI models become, they cannot own a company's internal files, bear legal liability, or become the party facing a lawsuit when results are incorrect. The real bottleneck in business lies not in the intelligence of technology, but in the realm of authority and responsibility held by humans.
•
The pitfalls facing services that settle for quantified scores
Achieving first place in publicly released benchmark scores is not something to boast about; rather, it serves as a warning signal that the domain will soon be taken over by artificial intelligence. Planners and partners must break away from thin wrapper structures that build services relying solely on the performance of general-purpose models. We can only survive by holding the standards to define what constitutes the best answer within our company or domain.
•
The power of values with a history that cannot be replicated through training
Even the best AI researchers cannot instantly train doctors' long-standing clinical habits or the complex operational flows experienced by lawyers. True value lies not in the flashy model itself, but in the tedious translation work of engaging with clients and aligning the AI with the realities of the field so that it can handle them. Value built over a long period through relationships and trust is not easily replicated.
🥄 A spoonful of execution
•
Among the tasks or projects we are working on today, what are our team's unique, private answers or customer trust assets that AI can never discover through data learning alone? Rather than getting bogged down in the competition for quantifiable efficiency, try to solidify that mere 3 percent of genuine substance that can only be filled with human judgment and responsibility.
—— View Original ——
•
By mid-2026, a sense of "AI-psychotic" despair is spreading among investors, with the core logic being that once models become better at everything, companies built upon them are nothing more than thin wrappers to be absorbed.
•
Devin, the first AI SWE, solved only 13% of standard benchmark tasks in 2024, but in just a year and a half, the top agent reached the high 80% range, with models rapidly encroaching on measurable areas .
•
Measurable things become training subjects and degenerate into commodities, but private and cost-effective correct answers cannot be read by leaderboards.
•
True value lies in areas beyond the reach of models, such as license, liability, and accountability, and the bottleneck is not intelligence but permission and accountability.
Want to read this article to the end? Click!
Please click (touch) the link block below.
Go to the platform where the article was written.
(* Separate registration may be required depending on the platform.)
Collaboration Proposals / Business Inquiries: hello@startlounge.kr
IR/Pitch deck manufacturing inquiries: hello@pitchmound.kr
IR/Pitch deck manufacturing inquiries: hello@pitchmound.kr
Everything about startups . Start Lounge - Launch Pad 「Lunch Pad」
ⓒ 2026. Start lounge. All rights reserved.
