What impact will AI have on the software industry?

"What's past is prologue." — William Shakespeare

1. What Is SWE (Software Engineering)?

1.1 The Lifecycle Perspective

From the lens of the software development lifecycle, Software Engineering (SWE) encompasses the entire process of software production—whether traditional applications, internet platforms, or next-generation AI systems.It spans requirement analysis, technical design, implementation and code generation, code review, testing, deployment, and monitoring and operations.SWE is a highly complex, multi-stage system. Writing code is only a small fragment of this lifecycle.

1.2 The Architectural Perspective

From the perspective of software architecture, no production-grade online service operates as a single module.

Every real system depends on underlying computing resources—networks, CPUs/GPUs, storage—as well as numerous interconnected services.

Today's AI systems focus narrowly on generating code within isolated modules. That is only the first step of software creation. The rest of SWE remains unsolved:

• How should a frontend connect securely to backend APIs?
• How should a database be defined—with schemas, permissions, and migration scripts?
• Once development is complete, how should it be deployed, monitored, and alerted?
• ...and countless similar questions.

In practice, these modules, configurations, and environments must fit together seamlessly. Otherwise, nothing runs.

1.3 The Challenge: Complexity

In one sentence: SWE is too complex—and therefore too expensive.

Despite decades of progress, engineers still face immense friction in integrating the right tools and workflows. Scripts can orchestrate certain tasks, but flexibility remains low, and human intervention is still indispensable.

More fundamentally, we should ask:

• Are today's architecture and language design principles truly first principles?
• Why do we still rely on notions like objects, types, functions, layers, cohesion, coupling, and APIs?

"Programs are meant to be read by humans and only incidentally for computers to execute." — Structure and Interpretation of Computer Programs (SICP)2. Where Is AI Headed?The rise of agents is no longer debated—it's accepted fact. But how these agents will evolve is the real question.We see four defining trends:

1. Agents will gain far stronger tool-use capabilities. Today, they invoke shell commands, Python interpreters, or web browsers. Tomorrow, they'll seamlessly integrate with the entire ecosystem of existing SWE software, frameworks, and services.
2. Traditional tools will evolve toward being AI-friendly. MCP (Model Context Protocol) is only the beginning—a network-layer component. The next wave will focus on AI-friendly APIs, data formats, and interaction patterns. This is a massive infrastructure and ecosystem opportunity for the coming decade.
3. Beyond LLMs: End-to-End Agentic Learning. Cutting-edge frontier models are already embedding tool-use within training (e.g., Deep Research). This trend will accelerate toward autonomously reasoning and acting systems with deep integration of external tools.
4. Continuous training and feedback loops will outperform general-purpose LLMs. In specific product contexts, user behavior creates closed feedback loops, enabling data flywheels that drive continuous improvement. Pure API-based models—which see only prompts but not outcomes—cannot replicate this. As a result, domain-specific models trained on in-product data will likely surpass general LLMs in both performance and cost.

3. AI + SWE = ?3.1 First-Order Impact: Transformation of the WorkflowThe entire digital world rests on software. If AI only helps write code, it addresses just a fraction of the problem.The true transformation lies in AI-SWE Agents that act as the planners and controllers of the full software lifecycle—from design and testing to deployment and maintenance. They orchestrate and optimize the entire software development workflow.AI coding is merely the prologue of a far larger industrial revolution.When these AI-SWE Agents mature, they could become the operating system and traffic layer of the software industry—creating opportunities on the scale of Microsoft, Oracle, Salesforce, or GitHub.And as software production costs plummet, new demand will explode—from SMEs, creators, and individual developers. This will expand the total addressable market and fundamentally reshape the supply‒demand dynamics of software creation.3.2 Second-Order Impact: The Rise of AI-Native InfrastructureIn an AI-native world, we must ask:

• Do abstractions like structured programming, object orientation, data types, layered architecture or high cohesion and low coupling still matter?
• What replaces them when AI, not humans, becomes the primary software engineer?

When machines become the main builders, the entire foundation must evolve—just as high-speed rail required an overhaul of traditional railway systems.All existing infrastructure was designed around human cognition—to be readable, maintainable, and debuggable by people.An AGI-level AI-SWE Agent, however, can operate via direct, ultra-efficient command channels. This calls for a new foundation: AI-Native Infrastructure—not "translating human tools for AI use," but building from zero for AI's native execution speed and concurrency.This is not theoretical. Consider Redis: today, humans use its APIs because modifying its core code is costly. An AI, however, could directly reconfigure low-level primitives for its needs—a paradigm where infrastructure itself becomes malleable.This opens revolutionary questions:

• What will AI-native programming languages look like?
• How will AI-generated code evolve?
• What will software engineering principles mean when the "engineer" is no longer human?

When the underlying principles shift, entire new industries will emerge.4. What Is Verdent Building?We believe four pillars are essential for building a defensible, high-impact AI-SWE ecosystem—not only for Verdent, but for the future of software as a whole.1. SWE-Agent System & ProductThis is the user-facing layer. As Agents become more autonomous, human interaction shifts from manual operations to task delegation. Users will issue requests, step away, and return for results. AI becomes not a tool to operate, but a task delegation and management system.Technically, this will be a controllable, interactive, asynchronous multi-agent system.2. Integrated Runtime Environment (IRE)A standardized, sandboxed environment for Agent execution—built using container and sandbox technologies. The IRE packages and optimizes tools, manages resource allocation, enforces isolation, and unifies runtime interfaces.Benefits include:

1. Agents can call tools safely without risking user environments.
2. Container-level protection eliminates constant user confirmations.
3. Standardized environments simplify compatibility across systems and hardware.
4. Agent Tool Discovery Service.

A curated, high-availability registry of open-source and commercial tools/services. Since LLMs often lack awareness of available tools, this service enables agents to discover and invoke the right ones securely and reliably. It will become a cornerstone of the agentic ecosystem.The IRE and Discovery Service together form a developer platform, where third-party tool providers can contribute their components—creating a thriving ecosystem.Today, most products on the market cover only a small portion of this vision—primarily AI coding. We believe that integrating all three layers is the key to unlocking the full potential of AI-SWE.