● AI Spend Surge

Enterprise AX 7-Step Formula: After PoC, the real battle is private AI, GPUs, security, and cost

The core point of this article is not simply “let’s adopt AI.”

It explains why enterprise AI transformation stalls even after passing a PoC, why organizational productivity stays flat even though individuals have become faster with AI, why cloud AI costs balloon like a snowball, and why companies eventually start considering their own GPUs and private AI servers.

In particular, based on the experience of Jiransoft CAIO Park Jong-cheon, enterprise AX has been reconstructed from a practical perspective from stage 0 to stage 6.

If you connect development-team vibe coding, VDLC, hiring methods in the AI era, AI full-stack developers, organizational knowledge libraries, AI security, and GPU investment all at once, the future AI industry and digital transformation trends become much clearer.

1. What does a CAIO do: a role that puts a hand on every AI decision in the company

A CAIO is not simply an executive who knows AI technology well.

It is closer to a role that changes how a company works around AI, adds AI features to existing products, and designs AI transformation for client companies as well.

The role described by CAIO Park Jong-cheon can be summarized into three broad areas.

• Internal AX transformation: changing the way employees work so an organization of 200 people can operate with AI.
• AI productization: adding features such as an AI security manager inside existing security solutions to improve product competitiveness.
• Supporting client AX: providing consulting, solutions, and implementation support so external companies can undergo AI transformation.

The important point here is that a CAIO is not just someone who looks at technology.

They must also look at how people work, organizational culture, security policies, cost structures, data flows, and development processes all together.

That is why a company’s AI transformation is closer to a management project that redesigns enterprise productivity than a simple IT project.

2. Development teams are already moving from SDLC to VDLC

The traditional software development method was SDLC.

SDLC stands for Software Development Life Cycle, the traditional development process that proceeds in sequence through requirements definition, design, development, testing, and deployment.

But in the era of vibe coding, this structure is changing.

CAIO Park Jong-cheon refers to this as VDLC.

VDLC is a method in which development proceeds not around code but around intent and documents.

• Humans spend less time writing code directly.
• Humans organize their intent, requirements, and design documents well.
• AI creates and tests code based on those documents.
• Developers become not code writers but designers and reviewers who instruct AI what to do.

This change is bigger than it may seem.

In the past, code was the core asset, but now clear intent and specification documents become the core asset.

Code is the output AI produces, and what really matters is the ability to define work so AI does not misunderstand it.

This trend is already moving quickly at global big tech companies as well.

For example, in organizations like Amazon, tickets posted to issue systems are handed over to AI tools like Claude Code, and multiple sessions are run at the same time to conduct problem analysis, fix planning, and testing.

When one person handles multiple AI sessions simultaneously, work that used to require several people can be completed much faster.

However, this approach inevitably increases token usage fees and AI tool costs.

From the perspective of the global economic outlook, AI usage fees and GPU infrastructure costs are likely to emerge as important items in corporate cost structures, on par with labor costs.

3. The downside of vibe coding: more frightening than cognitive debt is communication debt

Vibe coding greatly increases development speed.

But it also has side effects.

One commonly discussed issue is cognitive debt.

Because AI creates code, developers end up not fully understanding the internal implementation.

But what CAIO Park Jong-cheon viewed as even more important was communication debt.

As the time spent working with AI increases, the time spent talking with people decreases.

AI listens well, responds quickly, and never gets tired.

By contrast, collaboration with people is slower, involves conflicting opinions, and requires coordination.

So there is a risk that developers will increasingly talk only to AI and stop communicating with teammates.

But organizational work is not completed alone.

What I produce must connect to other people’s work and align with the direction of the whole team.

Therefore, as vibe coding becomes stronger, human communication mechanisms such as daily standups, team retrospectives, and communication workshops become even more important.

The more AI raises individual productivity, the more intentionally organizations must design communication structures.

4. Developer hiring in the AI era becomes ‘building a product with AI,’ not a coding test

Developer hiring methods are also changing in the AI era.

At Jiransoft, they reportedly do not use a simple algorithmic coding test, but instead conduct a test in which candidates build an actual product using vibe coding.

For example, they ask candidates to build a difficult feature such as an office agent on their own within a week.

Applicants must clone existing source code and use AI tools to create a working result.

The even more interesting point is that the evaluation is not done by humans first.

An AI evaluation system reviews the applicant’s code and creates a report analyzing architecture, implementation ability, quality, and productivity.

Multiple AI tools such as Claude Code and Codex are used for cross-evaluation, and only after a certain score is exceeded do they look at the resume.

This approach shows the direction of hiring in the AI era.

• People who can produce results with AI become more important than people who simply know coding syntax.
• AI full-stack capability becomes more important than the distinction between front-end and back-end.
• How well someone directs AI, verifies results, and turns them into a product becomes the core competency.

Now even junior developers can quickly produce results at a practical level if they use AI well.

Conversely, experienced developers who cannot use AI well may lose in the productivity competition.

5. The enterprise AX 7-step formula: if there is no stage 0, you cannot even start

CAIO Park Jong-cheon organizes enterprise AX transformation into seven stages.

The interesting point is that it starts not from stage 1 but from stage 0.

Stage 0 is so important that without it, you cannot even begin.

Stage 0: AX Readiness, checking preparedness

The start of enterprise AX is not technology but readiness.

Readiness requires three things.

• Leadership commitment: top management must genuinely push the AI transition.
• Dedicated AX team: an internal TF is needed to coordinate between the business side and external partners.
• Sufficient time: AI transformation is not a technology issue but a cultural change, so time is needed.

This is why many companies succeed in PoC but fail in actual expansion.

The CEO lacks interest, there is no dedicated internal team, the business side is busy, and employees are wary that AI might replace them.

In this state, even a good solution will not run properly.

Stage 1: AX Kick-off, an enterprise-wide launch declaration

A kick-off may look like a simple event, but it is important.

It is the stage where the company officially conveys to all employees, “From now on, the company will work in an AI-based way.”

The goal is to build a shared understanding that AI transformation is not an experiment by a few teams but the direction of the whole company.

Stage 2: AX Discovery, work analysis and idea discovery

At this stage, actual work is analyzed.

The goal is to find what tasks are repetitive, what tasks have data, and what tasks can be automated or improved by AI.

Business understanding is extremely important.

Even if it looks easy to an AI expert, if there is no business data or if the responsible person does not accept it, it fails.

On the other hand, among the tasks employees struggle with every day, there are many areas where AI can deliver major results.

Stage 3: AX PoC, actually building it

Ideas alone are not enough.

You must actually build something small and verify whether it works.

This is the PoC we commonly talk about.

But the important point is that passing the PoC is not the end.

A PoC is only a “possibility check,” and the real difficulty of enterprise AX begins after that.

Stage 4: AX Adoption, applying it to actual work

Once the PoC succeeds, it must be inserted into real work.

At this point, employees’ attitudes begin to change.

Even employees who were initially lukewarm start to think differently when they see AI outputs being faster and better than their own work.

They begin to feel not “I need to compete with AI,” but “I need to ride with AI.”

It is similar to not trying to run faster than a car, but instead riding in the car.

Stage 5: AX Scale-up, enterprise-wide expansion

When results appear in one or two departments, other departments begin raising their hands.

Requests such as “Please do this for our department too” and “Please change this task too” increase.

From this point on, AX becomes an enterprise-wide transformation, not a project.

But at the same time, cost, security, and data management problems grow.

Stage 6: Private Transformation, private AI transformation

Once scale-up becomes serious, two problems eventually slow things down.

Security and cost.

The more employees use AI, the more token costs increase.

The more business data accumulates in AI systems, the greater the security risk.

So companies eventually begin to consider private AI, their own GPUs, and internal AI servers.

This trend is not just a technological fad but a structural change combining cloud cost, data sovereignty, and enterprise security.

6. After PoC, the real cost begins: rent or buy

Many companies initially approach this by renting cloud GPUs.

At first, this seems reasonable.

But as usage increases, the calculation changes.

As mentioned in the video, when they tried to rent GPU infrastructure in the cloud for about 200 employees, the annual estimate came to around 500 million won.

By contrast, directly purchasing a high-performance GPU server could be reviewed at roughly the 200 million won range.

Of course, the actual cost varies depending on hardware configuration, maintenance, operating personnel, and model optimization level.

But the core point is clear.

Once AI usage exceeds a certain level, building in-house may be more economical than cloud rental.

This means corporate AI infrastructure investment can become not just an expense but a mid- to long-term CAPEX strategy.

In the future, companies are likely to think about GPU investment the way they think about hiring employees.

The phrase “AI employees start with GPUs” did not come out of nowhere.

7. Why is a GPU an ‘AI employee’?

People go home.

But GPUs can work all night long.

When a company owns its own GPUs, AI can continue handling document analysis, code generation, log analysis, customer request organization, and first-draft proposal writing even after work hours end.

From this perspective, a GPU is not just equipment but the foundation of an AI employee that works 24 hours a day.

Of course, buying GPUs does not automatically increase productivity.

To use GPUs properly, you need models, agent frameworks, tool calling, internal data integration, and security systems all together.

But once this structure is in place, a company can move away from the model of continuously paying external AI usage fees and instead build productivity as an internal asset.

This can have a major impact not only on GPU companies like Nvidia but also on server infrastructure, open-source models, AI security, and the data platform market.

8. Why individual productivity does not lead to organizational productivity

These days, individuals use AI fairly well.

They create report drafts, write emails, code, and do market research.

But if you ask whether the productivity of the whole company increased that much, the answer is unclear.

The reason is simple.

AI used by individuals often operates only within the person’s own knowledge and files.

If it is not connected to the organization’s data, documents, history, customer information, and work processes, AI remains just a personal tool.

For enterprise AX to succeed, it must move from personal AI to organizational AI.

The key here is the organizational knowledge library.

Company documents, emails, meeting notes, customer requests, proposals, product manuals, and policy documents must be connected.

Only then can AI retrieve the necessary knowledge and work with it, instead of wandering around looking for information.

If you consider that employees spend a large portion of their work time searching for information and then redoing work because of incorrect information, the organizational knowledge library becomes a core infrastructure for enterprise productivity.

9. Four stages of enterprise AI maturity: from personal AI to organizational AI employees

If we reorganize enterprise AI maturity based on CAIO Park Jong-cheon’s perspective, it can be seen in four stages.

Stage 1: Personal AI

This is the stage where an individual asks tools like ChatGPT or Claude Chat questions and gets answers.

It is a level that most company employees have already experienced.

It helps the company somewhat, but it is hard to call it a change at the organizational level.

Stage 2: Work AI

This is the stage where an individual puts personal documents, proposals, and materials into a tool like Claude Code and creates actual work results.

At this stage, productivity in document writing, report writing, proposal writing, and code writing rises significantly.

CAIO Park Jong-cheon believes that humans must now move beyond the era of directly creating PowerPoint or Word documents.

Humans should define intent and specifications, and AI should produce the results.

Stage 3: Organizational knowledge-based AI

This is the stage where AI works based not on personal documents but on the company’s entire knowledge library.

Once you enter this stage, the time employees spend searching for information decreases.

AI can create outputs based on company documents, customer history, emails, and internal materials.

Stage 4: System-connected AI employee

This is the stage where AI goes beyond a simple response tool and becomes connected to company systems.

Email, CRM, approval systems, customer support systems, document repositories, development tickets, and dashboards become connected.

For example, AI can handle tasks such as “Analyze the customer requests received yesterday, create a draft related proposal, and share it with the person in charge.”

At this stage, AI begins to work like an employee rather than just a tool.

10. AI security is not optional; it is essential

The more company data AI handles, the greater the security risk becomes.

In the past, data was scattered, so there was limited material for attackers to steal.

But as AX progresses, important company data gathers around AI systems.

That makes the company a much more attractive target from an attacker’s perspective.

So AI productivity and AI security cannot go in separate directions.

As much as AI is used to raise productivity, a security framework that responds to AI hacking must also develop alongside it.

Existing rule-based security alone has limits.

That is because AI-based attacks keep changing shape and create new risks such as prompt injection, data leakage, unauthorized privilege use, and exposure of internal documents.

In the future, enterprise AI security is likely to evolve beyond simple access control toward tracking what data AI read, what decisions it made, and what actions it carried out.

11. AI full-stack structure: you must understand everything from GPUs to tool calling

To build enterprise private AI, you need to understand multiple layers.

The core structure is as follows.

• GPU: the physical computing infrastructure that runs AI models.
• AI model: an open-source LLM or enterprise-specific model is loaded here.
• Agent framework: a framework like LangGraph uses the model to organize task flow.
• Tool calling: the function by which AI calls external tools such as email, documents, search, approval, and databases.
• Service integration: connects with internal company documents, mail, chat, customer data, and business systems.

The parts companies usually need to handle well themselves are tool calling and internal system integration.

You can buy GPUs, load open-source models, and use proven agent frameworks.

But connecting a company’s document system, customer data, legacy systems, and security policies to AI is different for every company.

That is why the value of AI full-stack developers will increase in the future.

An AI full-stack developer is not just someone who knows front-end and back-end.

They understand the whole structure, from GPUs, models, RAG, agents, MCP, and tool calling to internal system integration, and can assemble it all.

However, this is not the era of building everything from the ground up by yourself.

The ability to bring in good open-source tools and frameworks, assemble them quickly, and connect them to company workflows is becoming more important.

12. The SaaS market may also be shaken: wait, or build it yourself?

In the AI era, existing SaaS products also come under pressure.

In the past, it was natural to buy and use tools like Jira, Notion, CRM, and document management systems.

But if simple business systems can be built quickly with AI, companies may not wait for existing SaaS products to adapt to the AI era.

Instead, it may be faster to directly build simple tools tailored exactly to internal work using vibe coding.

This trend connects to what some call the SaaS apocalypse discussion.

Of course, not all SaaS will disappear.

But SaaS that cannot deeply integrate with AI may lose competitiveness.

On the other hand, SaaS that connects well with company data, workflows, and AI agents may become stronger.

13. The most important point that other news or YouTube rarely explains well

The real core point is not “whether you use AI tools,” but “whether AI can access the organization’s knowledge and systems and work on its own.”

Many pieces of content talk about how to use ChatGPT, how to write prompts, and tips for workflow automation.

But the essence of enterprise AX is not how to use personal productivity tools.

For a company to truly change, the following three things must exist at the same time.

• Organizational knowledge library: AI must understand the company’s context.
• Private AI infrastructure: costs and security must be controllable.
• Tool calling and system integration: AI must be able to act in real business systems, not just talk.

Without these three, AI remains a smart search box.

Conversely, if these three are in place, AI becomes a digital employee that works through the night.

This is likely where the difference in corporate competitiveness will widen.

In future global economic outlooks, an important variable may be not only interest rates, exchange rates, and supply chains, but also how quickly companies internalize AI infrastructure.

The productivity gap between companies that use AI only as an external tool and companies that absorb it into their internal operating system is likely to widen over time.

14. Changes to watch from an investment and industry perspective

This story is not just one company’s AX experience.

It is also a signal showing where money in the AI industry may flow.

• GPU demand is likely to remain strong. As companies consider their own AI servers, the Nvidia-centered GPU ecosystem becomes even more important.
• The cloud cost optimization market may grow. As AI usage increases, companies must calculate whether to rent, buy, or go hybrid.
• The AI security market is likely to become more active. The more data accumulates, the more valuable it becomes to attack.
• The importance of open-source models and agent frameworks increases. Companies will move away from a structure that only uses GPT APIs and consider operating their own models.
• Demand for AI full-stack talent is likely to grow. Rather than distinguishing only between front-end and back-end, the ability to assemble the entire AI system becomes more important.

In particular, from a company’s perspective, AI investment decisions may become more complex.

That is because they must look beyond simply paying a monthly subscription fee for an AI service and also consider GPU server purchases, data security, internal system integration, employee training, and redesigning work processes.

From now on, AX is becoming not just an IT budget issue but a matter of management strategy and capital allocation.

< Summary >

Enterprise AX does not end when only the PoC succeeds.

The real battle is decided in actual work adoption, enterprise-wide expansion, security, cost, and private AI construction.

The AX seven-stage framework proposed by CAIO Park Jong-cheon can be organized into stage 0 readiness, kick-off, work discovery, PoC, actual work adoption, scale-up, and private transformation.

Development teams are moving from SDLC to VDLC, and people are shifting from writing code to managing intent and documents.

For personal AI productivity to lead to organizational productivity, an organizational knowledge library and system integration are essential.

As AI usage increases, cloud costs and security issues grow, and eventually in-house GPUs and private AI servers become important options.

Going forward, corporate competitiveness will likely be determined not by how many AI tools a company uses, but by how deeply it internalizes AI within its organization’s knowledge and business systems.

[Related Articles…]

• Analysis of AI transformation and changes in enterprise productivity
• Outlook for GPU investment and private AI infrastructure

*Source: [ 티타임즈TV ]

– 경험으로 만든 AX 7단계 공식 (박종천 지란지교소프트 CAIO)

---