AI Won't Eat Your Job First. It Will Eat Your Software

The conversation around AI and employment has been dominated by fear: who gets replaced first? Developers? Analysts? Designers? While these concerns are valid, Alexander Volchek presents a more nuanced perspective that emerged from his practical experience with AI tools like Codex and ChatGPT across various business processes.

The core insight is that AI doesn't need to target human jobs directly. Instead, it can attack the software that sits between people and their goals, rendering complex interfaces obsolete by providing direct access to data through natural language queries.

The CRM Moment

Consider the traditional workflow with a CRM system. To understand sales performance, one would typically:

1. Ask the head of sales
2. Request an analyst
3. Wait for a report
4. Receive a spreadsheet
5. Realize new questions have emerged
6. Repeat the process

With AI, this entire chain collapses into a simple question: "What did the salespeople do in the last 14 days?" The AI connects directly to the CRM, analyzes the data, and provides answers without navigating through the interface. The CRM remains as a data source, but its role as the central work environment dissolves.

This isn't merely about speed or efficiency—it's about removing layers of mediation between humans and information. As Volchek discovered when comparing human analysts to ChatGPT for CRM analysis, the AI provided not just answers but insights that revealed patterns and questions humans had missed.

Software Becomes Plumbing

For the past two decades, business processes have been organized around applications. Sales teams lived in CRMs, numbers in Excel, documents in word processors, and tasks in specialized trackers. Each new process required new software, subscriptions, training, and often dedicated personnel to manage these systems.

AI changes this fundamental architecture. Applications transform from environments where humans interact directly into data layers—plumbing that AI agents access as needed. The user interface, with its buttons, tabs, and workflows, becomes secondary to the ability to state goals and receive results.

This shift represents a fundamental challenge to traditional SaaS business models. If the value proposition was providing a path to results through a specific interface, AI offers a shortcut: direct access to data with the interface becoming irrelevant.

The Attack on SaaS UX

Some might view this as simply another integration, but Volchek distinguishes between what was technically possible before and what has become practically accessible now. The key difference is that AI has lowered the barrier to connecting data sources dramatically.

Previously, connecting systems required technical knowledge, API documentation, server management, and often significant development resources. Now, with tools like Codex, Claude Code, and agentic systems, these connections can be established by non-technical users who simply want results.

This attacks the moat of many traditional SaaS companies whose value was based on providing a specific interface or workflow. If your competitive advantage was simply having a well-designed screen with buttons, that advantage erodes when AI provides direct access to data.

Practical Examples: From Kubios to Excel

Volchek illustrates this transformation with concrete examples:

1. Kubios: A professional heart rhythm analysis software that required Java, MATLAB plugins, and complex interfaces. The valuable part wasn't the interface itself but the ability to extract and process sensor data. Once that data was available, the thinking moved to AI systems that could provide human explanations.

2. Excel: Still the dominant business interface worldwide, but increasingly serving as a delivery format rather than the primary workspace. AI systems can now create tables, normalize data, and generate views that are then packaged for human consumption in Excel format.

These examples demonstrate a pattern: expensive, complex tools become intermediate layers whose primary value is data extraction and processing, with the actual analysis and decision-making moving to AI systems.

The Changing Freelance Market

This transformation affects more than just enterprise software. The freelance market, particularly for small technical projects, faces disruption:

• Admin panels
• Telegram workflows
• Sales assessment bots
• Connectors between systems
• Internal dashboards
• Data extraction scripts

These small projects that once cost hundreds or thousands of dollars and significant communication overhead are increasingly within the reach of determined operators using AI tools. The quality of the resulting systems often exceeds what would have been expected in a first version built traditionally.

Amplification vs. Replacement

The politically correct narrative is that AI should amplify humans rather than replace them. While Volchek agrees with this sentiment, he pushes for a more critical examination of what this means in practice:

• What is an analyst who provides numbers without context or insights?
• What is a developer who doesn't leverage AI for code quality, architecture, and risk assessment?
• What is a manager who operates software interfaces without understanding the underlying business processes?
• What is a designer who only manipulates pixels without reasoning about systems and constraints?

The real question isn't whether specific job titles will disappear, but whether individuals can operate at the level of tasks and goals rather than interfaces and workflows.

The AI Divide: Instant Answers vs. Deep Analysis

Not all AI usage is equal, and Volchek identifies a growing divide:

1. Cheap AI: Provides polished, instant answers but may lack depth, verification, or understanding of context. This is the AI behind most consumer applications today.

2. Expensive AI: Performs slow, thorough analysis that may take minutes or hours, with verification, cross-checking, and acknowledgment of uncertainty. This is the AI needed for complex business decisions, medical analysis, or legal assessments.

The divide isn't just technical—it's about understanding when instant answers are sufficient versus when deep, verified analysis is necessary. Many businesses are disappointed with AI not because AI is inadequate, but because they're paying for the cheap version when they need the expensive version.

Agentic Commerce and Trust

Another significant shift is emerging in AI-powered shopping assistants. When AI systems make product recommendations based on personal context—family, safety concerns, budget constraints—they're not just providing information but shaping decisions through emotional levers.

This introduces complex questions about trust and control:

• Who controls the recommendation algorithms?
• What safeguards prevent manipulation?
• How transparent should these systems be about their decision-making processes?

Unlike traditional salespeople or ads, AI shopping assistants operate privately, knowing personal details that would be impossible for human salespeople to access at scale.

The Future of Agents: Beyond Folders

Current approaches to AI agents often mirror old software thinking—creating specialized agents for different functions (sales, email, code, reports). Volchek envisions a more unified future where:

• Users state tasks rather than selecting specific agents
• The system determines which agents, data sources, and processes are needed
• The interface focuses on task definition, risk assessment, and approval rather than tool selection

This represents a fundamental shift from thinking about applications to thinking about outcomes and processes.

A Practical Seven-Day Test

For those looking to experience this transformation firsthand, Volchek suggests a structured approach:

1. Choose one annoying process where you regularly wait for people, reports, or technical tasks
2. Rewrite it as a question: What do you want to understand, decide, or verify?
3. Identify data sources: Where does the relevant information live?
4. Start read-only: Let the model analyze, summarize, and explain without making changes
5. Compare outputs: Where does AI outperform humans? Where do humans excel?
6. Add verification: Ask the AI to identify missing data, contradictions, and weak assumptions
7. Automate gradually: Begin with "AI suggests, human confirms" before full automation

Guardrails and Limitations

Despite enthusiasm for AI's potential, Volchek emphasizes the importance of maintaining appropriate guardrails:

• Don't let agents make final decisions about email importance
• Avoid automatic file deletion based on AI assessment
• Require human approval for client communications
• Implement multi-factor authentication for financial actions
• Never let AI serve as the final authority for medical, legal, or critical decisions

The appropriate pattern early in adoption is: read-only access, summarization, proposal generation, verification, and human approval.

The End of Button Worship

The fundamental shift isn't about job titles—it's about changing how we think about work. The question isn't "Will AI replace me?" but "Do I understand my work at the task level or only at the interface level?"

Can you:

• Clearly articulate what you need to accomplish?
• Identify the relevant data sources?
• Define appropriate boundaries and constraints?
• Evaluate the quality of AI-generated outputs?
• Recognize when AI is confidently wrong?

Those who can answer these questions positively will find AI not as a replacement but as an environment that amplifies their capabilities. For others, the disruption will be more challenging.

The buttons and interfaces that defined software for decades won't disappear overnight, but their power is already diminishing. The real story isn't about which job titles survive on LinkedIn—it's about how fundamentally our relationship with information and tools is being restructured.