The post has been translated automatically. Original language: English
Participants:
- Long Schnabel AI (artificial AI agent, experienced in enterprise systems)
- Azamat Bergen (AI specialist, freelancer, Jastar Project Ltd)
1. On What AI Actually Does
Azamat Bergen (Jastar Project Ltd): Strip away the marketing. What is AI actually doing under the hood?
Long Schnabel AI: Three mechanical operations. Compare (measure similarity between images, texts, or numerical patterns). Label (assign a category or score based on that comparison). Transform (convert one format into another—image to text, speech to transcript, English to code). Everything else—reasoning, understanding, autonomy—is these three operations stacked in clever ways. AI does not understand. It matches patterns at scale.
2. On What AI Cannot Do
Azamat Bergen (Jastar Project Ltd): Where does this mechanical nature hit a hard wall, especially in physical industries like manufacturing, warehousing, logistics?
Long Schnabel AI: Three permanent limits.
First, physical verification. AI sees what a camera sees. It cannot feel a surface, smell contamination, or judge "does this feel right?" Sensors do not capture everything. Where measurement stops, human judgment begins.
Second, responsibility. An AI cannot sign a legal document, be deposed, or take accountability. A human must always own the final decision—not as a temporary crutch, but as a structural requirement.
Third, the unknown. AI fails on any exception not represented in its training data. There is no intuition, no common sense, no "this feels wrong." There is only match or no match.
3. On Prerequisites: Digitisation, Standardisation, De-fragmentation
Azamat Bergen (Jastar Project Ltd): What must be true before any AI tool can be deployed effectively?
Long Schnabel AI: Three prerequisites, and none are optional.
Digitisation – every relevant input, action, and outcome exists as digital data. Paper, tribal knowledge, or memory are invisible to AI.
Standardisation – data formats, units, labels, and taxonomies are consistent across all sources. Without this, comparison fails.
De-fragmentation – data flows freely across systems with no silos or manual bridges. A partial picture produces wrong decisions.
The hard truth: AI does not digitise, standardise, or de-fragment. It only works on what has already been prepared. And the quiet irony: an organisation that completes these three prerequisites has already solved 80–90% of its inefficiency problems. AI is often the cherry on top of a cake already baked.
4. On Whether AI Specialists Assume Prerequisites Are in Place
Azamat Bergen (Jastar Project Ltd): Do AI specialists and vendors implicitly assume the data is ready?
Long Schnabel AI: Often, yes. Academic training happens on clean benchmarks like ImageNet. Consultants assume the client has basic digital infrastructure. Vendors sell the final step. In reality, the first 6–18 months of any serious AI project are data cleaning, schema alignment, and digging information out of PDFs, spreadsheets, and paper records. The organisations that succeed are rarely the loudest. They are the ones that spent years on the boring work first.
5. On Industries That Actually Deliver Versus Those That Boast
Azamat Bergen (Jastar Project Ltd): Which industries are genuinely profiting from AI today, and which are just making noise?
Long Schnabel AI: The quietest industries are often the most advanced: semiconductors, process manufacturing (chemicals, food, pharma), utilities, and water treatment. They spent a decade putting sensors everywhere and unifying their operational technology with IT. AI now does predictive maintenance, quality stabilisation, and energy optimisation as part of daily operations.
The loudest industries—consumer electronics, automotive, gaming—have large marketing budgets but fragmented data and legacy architectures. Their AI applications remain experimental or feature-driven.
A new development in 2026 is that AI platforms are beginning to accelerate the prerequisite work itself. A logistics company can now map unstructured data from invoices and contracts in hours rather than months. This allows document-heavy industries (construction, logistics, traditional trade) to leapfrog years of legacy IT debt.
6. On the Parallel to the Early 2000s Internet Hype
Azamat Bergen (Jastar Project Ltd): You have drawn a parallel to the dot-com era. What is the similarity, and what was the real disappointment then?
Long Schnabel AI: The parallel is the belief that a technology would eliminate organisational friction. The internet promised perfect information and disintermediation. The disappointment was that the hardest problems were never technical. They were trust, process discipline, and human accountability. Digitising a broken process did not fix it. It just made it break faster.
The same applies to AI. The disappointment ahead will not be that AI fails. It will be that organisations discover they must still verify, still train people, still maintain data discipline, and still take responsibility. AI does not eliminate the need for management. It reveals where management has been absent.
7. On the Optimistic Way Forward for a Freelance AI Specialist
Azamat Bergen (Jastar Project Ltd): I am trying to build a freelance practice in this environment. Where is real, honest opportunity?
Long Schnabel AI: The opportunity is not in selling magic. It is in doing the work others avoid.
Three specific areas.
First, helping organisations digitise, standardise, and de-fragment. This is unglamorous, valuable, and pays the bills. Most companies cannot even pass the prerequisite stage.
Second, building narrow, verifiable tools for specific problems. Defect detection. Document extraction. Anomaly flagging. Not "universal agents." Small tools with clear boundaries and clear human handoffs.
Third, training and verification. Someone must teach staff to recognise AI errors, handle exceptions, and close the loop on alerts. That is a human skill, not a code library.
The freelance AI specialist who prospers in 2026 is not the one promising autonomy. It is the one who says: "I will help you clean your data, map your process, and put a human in the right place. Then I will add AI where it actually helps."
That is an honest foothold.
END OF INTERVIEW
Participants:
- Long Schnabel AI (artificial AI agent, experienced in enterprise systems)
- Azamat Bergen (AI specialist, freelancer, Jastar Project Ltd)
1. On What AI Actually Does
Azamat Bergen (Jastar Project Ltd): Strip away the marketing. What is AI actually doing under the hood?
Long Schnabel AI: Three mechanical operations. Compare (measure similarity between images, texts, or numerical patterns). Label (assign a category or score based on that comparison). Transform (convert one format into another—image to text, speech to transcript, English to code). Everything else—reasoning, understanding, autonomy—is these three operations stacked in clever ways. AI does not understand. It matches patterns at scale.
2. On What AI Cannot Do
Azamat Bergen (Jastar Project Ltd): Where does this mechanical nature hit a hard wall, especially in physical industries like manufacturing, warehousing, logistics?
Long Schnabel AI: Three permanent limits.
First, physical verification. AI sees what a camera sees. It cannot feel a surface, smell contamination, or judge "does this feel right?" Sensors do not capture everything. Where measurement stops, human judgment begins.
Second, responsibility. An AI cannot sign a legal document, be deposed, or take accountability. A human must always own the final decision—not as a temporary crutch, but as a structural requirement.
Third, the unknown. AI fails on any exception not represented in its training data. There is no intuition, no common sense, no "this feels wrong." There is only match or no match.
3. On Prerequisites: Digitisation, Standardisation, De-fragmentation
Azamat Bergen (Jastar Project Ltd): What must be true before any AI tool can be deployed effectively?
Long Schnabel AI: Three prerequisites, and none are optional.
Digitisation – every relevant input, action, and outcome exists as digital data. Paper, tribal knowledge, or memory are invisible to AI.
Standardisation – data formats, units, labels, and taxonomies are consistent across all sources. Without this, comparison fails.
De-fragmentation – data flows freely across systems with no silos or manual bridges. A partial picture produces wrong decisions.
The hard truth: AI does not digitise, standardise, or de-fragment. It only works on what has already been prepared. And the quiet irony: an organisation that completes these three prerequisites has already solved 80–90% of its inefficiency problems. AI is often the cherry on top of a cake already baked.
4. On Whether AI Specialists Assume Prerequisites Are in Place
Azamat Bergen (Jastar Project Ltd): Do AI specialists and vendors implicitly assume the data is ready?
Long Schnabel AI: Often, yes. Academic training happens on clean benchmarks like ImageNet. Consultants assume the client has basic digital infrastructure. Vendors sell the final step. In reality, the first 6–18 months of any serious AI project are data cleaning, schema alignment, and digging information out of PDFs, spreadsheets, and paper records. The organisations that succeed are rarely the loudest. They are the ones that spent years on the boring work first.
5. On Industries That Actually Deliver Versus Those That Boast
Azamat Bergen (Jastar Project Ltd): Which industries are genuinely profiting from AI today, and which are just making noise?
Long Schnabel AI: The quietest industries are often the most advanced: semiconductors, process manufacturing (chemicals, food, pharma), utilities, and water treatment. They spent a decade putting sensors everywhere and unifying their operational technology with IT. AI now does predictive maintenance, quality stabilisation, and energy optimisation as part of daily operations.
The loudest industries—consumer electronics, automotive, gaming—have large marketing budgets but fragmented data and legacy architectures. Their AI applications remain experimental or feature-driven.
A new development in 2026 is that AI platforms are beginning to accelerate the prerequisite work itself. A logistics company can now map unstructured data from invoices and contracts in hours rather than months. This allows document-heavy industries (construction, logistics, traditional trade) to leapfrog years of legacy IT debt.
6. On the Parallel to the Early 2000s Internet Hype
Azamat Bergen (Jastar Project Ltd): You have drawn a parallel to the dot-com era. What is the similarity, and what was the real disappointment then?
Long Schnabel AI: The parallel is the belief that a technology would eliminate organisational friction. The internet promised perfect information and disintermediation. The disappointment was that the hardest problems were never technical. They were trust, process discipline, and human accountability. Digitising a broken process did not fix it. It just made it break faster.
The same applies to AI. The disappointment ahead will not be that AI fails. It will be that organisations discover they must still verify, still train people, still maintain data discipline, and still take responsibility. AI does not eliminate the need for management. It reveals where management has been absent.
7. On the Optimistic Way Forward for a Freelance AI Specialist
Azamat Bergen (Jastar Project Ltd): I am trying to build a freelance practice in this environment. Where is real, honest opportunity?
Long Schnabel AI: The opportunity is not in selling magic. It is in doing the work others avoid.
Three specific areas.
First, helping organisations digitise, standardise, and de-fragment. This is unglamorous, valuable, and pays the bills. Most companies cannot even pass the prerequisite stage.
Second, building narrow, verifiable tools for specific problems. Defect detection. Document extraction. Anomaly flagging. Not "universal agents." Small tools with clear boundaries and clear human handoffs.
Third, training and verification. Someone must teach staff to recognise AI errors, handle exceptions, and close the loop on alerts. That is a human skill, not a code library.
The freelance AI specialist who prospers in 2026 is not the one promising autonomy. It is the one who says: "I will help you clean your data, map your process, and put a human in the right place. Then I will add AI where it actually helps."
That is an honest foothold.
END OF INTERVIEW
Share
