AI 時代，重新思考軟體工程的價值

Podcast ConversationAI dialogue version of this article

Foreword

Product thinking in the AI era is shifting from point-like improvements, such as "using AI to optimize a single process," to a systemic strategy of "using AI to redefine value creation." Based on our team's experience, simply pursuing development speed or traffic numbers is no longer enough to build a true moat. When development speed increases, quality often fails to keep up; when traffic is acquired, conversion rates stagnate.

This article shares a "dual-engine" framework: internally, we leverage AI-powered DevSecOps to enhance development quality and efficiency; externally, we drive new customer acquisition and conversion through Generative Engine Optimization (GEO) and user experience optimization. The ultimate goal is for these two engines to feed each other—efficient and secure product development supports rapid market iteration, while high-quality external traffic and high conversion rates provide continuous momentum and data insights for the product.

Internal Value Reinvention: The AI-Driven DevSecOps Revolution

The Quality Concerns Behind the Speed Dividend

On the development scene in 2026, AI collaboration is no longer a novelty. According to industry statistics, about 85% of developers use AI tools daily, 41% of code is generated by AI, teams with deep adoption have seen a 21% increase in task completion, and the number of merged PRs has even increased by 98%.

But the other side of the coin is that nearly 48% of AI-generated code contains security vulnerabilities, and PR review times have surprisingly increased by 91%. This figure isn't hard to understand—the volume of AI-generated code has exploded, but every line still needs to be reviewed by a human. Moreover, the reviewer must be extra vigilant about security issues that AI is prone to making (SQL injection, XSS, sensitive data leaks, etc.).

If increased speed comes at the cost of quality, the final price will only be higher. This is a point I've consistently emphasized when I previously discussed AI development trends.

The Core Solution: Shift Left Security + An AI-Driven Automated Defense Line

The core idea of the solution is "Shift Left Security"—pulling security checks from pre-deployment into the development phase, and even right to the moment of coding. The specific practices are:

Real-time scanning within the IDE: As developers write code (or accept AI suggestions), tools can detect vulnerabilities in real-time and provide fix recommendations, rather than waiting for a pile of issues to emerge during the PR stage.
AI-driven automated testing: Let AI automatically generate test cases, especially for edge cases and security scenarios, to cover blind spots that humans easily miss.
Integrating security scans into the CI/CD Pipeline: Automatically run static analysis, dependency vulnerability checks, and container image scanning with every commit.

開發者寫程式碼 / 接受 AI 建議
        │
        ▼
┌──────────────────────┐
│  IDE 即時安全掃描      │ ← 第一道防線（寫的當下就抓）
│  SQL Injection ❌     │
│  XSS 漏洞 ❌         │
│  → 即時修復建議 ✅    │
└──────────────────────┘
        │
        ▼
┌──────────────────────┐
│  AI 自動化測試生成     │ ← 第二道防線（邊界與安全場景）
└──────────────────────┘
        │
        ▼
┌──────────────────────┐
│  CI/CD 安全掃描       │ ← 第三道防線（上線前最終把關）
│  SAST / DAST / SCA   │
└──────────────────────┘
        │
        ▼
   安全地部署到 Production

Effects and Benefits

Through this AI-powered DevSecOps process, teams can enjoy the efficiency dividends of AI while ensuring code quality and security. According to industry data, companies with deep adoption have seen a 20-40% reduction in operational costs. But I believe what's more important is establishing a positive cycle of "high-quality development efficiency"—it's not just about writing code fast, but writing it fast and stable.

This also echoes the thinking in refactoring legacy systems in the AI era: AI is not just an accelerator; it's a quality gatekeeper. Offloading repetitive security checks and test writing to AI allows engineers to focus on architectural design and business logic. This is the true embodiment of the value of software engineering.

Traditional SEO is Being Rewritten

Google AI Overviews, ChatGPT Search, Perplexity—these AI search tools are changing how users get information. Users are increasingly getting answers directly from AI-generated summaries, and traditional click-through rates are declining.

At the same time, the numbers on the e-commerce side aren't optimistic either: the global cart abandonment rate is as high as 70.22%, the mobile conversion rate is only 1.53% (less than half of the desktop's 3.36%), yet mobile accounts for 68% of total traffic. This means a huge amount of potential revenue is being lost.

A Two-Pronged Approach: Transforming Traffic Strategy + Optimizing Conversion Experience

🔍 Traffic Acquisition: From SEO to GEO (Generative Engine Optimization)

The core logic of GEO is to make your content an authoritative source that AI systems are willing to cite.

Structured Content: With clear heading hierarchies, bullet points, and FAQ schema, it's easier for AI to parse and cite your content.
Multimodal Layout: YouTube videos are no longer just a secondary channel but a multimodal data source fed directly into the Google search experience. The podcast audience is growing by 25-35% annually, and its purchase intent metrics outperform all other content formats.
SEO-Oriented Production: The titles, descriptions, and transcripts of videos and podcasts must be optimized so that AI search engines can understand and index them.

💳 Conversion: Reduce Friction, Every Step Counts

To address the 70% cart abandonment rate, here are a few high-ROI optimization directions:

Optimization Strategy	Expected Effect	Description
One-Page Checkout	20% reduction in abandonment rate	Reduces page transitions, lowering user churn.
Diverse Payment Options (Apple Pay, etc.)	22.3% increase in conversion rate	13% of consumers leave because their preferred payment method is unavailable.
Mobile Optimization	Greatest room for optimization	Accounts for 68% of traffic but has only half the conversion rate of desktop.
Social Login	Lowers the registration barrier	Reduces form filling, speeding up the checkout process.

These observations are consistent with the logic discussed in the differences between e-commerce and physical stores—every point of friction in online shopping is a point of churn.

📊 Social Commerce: A High-Conversion Channel That Can't Be Ignored

Social commerce is projected to exceed $1.6 trillion globally by 2026, and the conversion numbers are staggering:

Channel	Average Conversion Rate	Characteristics
Traditional E-commerce	2-3%	Mature but with slowing growth.
TikTok Shop	4.7%	Content-driven, suitable for impulse buying scenarios.
Livestream Commerce	Up to 30%	Real-time interaction + limited-time offers, highest conversion rate.

These numbers illustrate one thing: the design of the user experience and purchasing scenario has a far greater impact on conversion rates than the sheer volume of traffic.

Effects and Benefits

Through a GEO strategy, brands can become authoritative sources in AI search results, capturing high-intent traffic. Combined with a seamless checkout experience and high-conversion social channels, this enables full-funnel growth from traffic acquisition to conversion.

Conclusion

In the AI era, excellent product thinking isn't about perfecting a single point, but about building two mutually reinforcing engines: internal efficiency and external growth. On one hand, we achieve high-quality, rapid development through AI-driven DevSecOps; on the other hand, we achieve efficient conversions in the new traffic ecosystem through GEO and experience optimization strategies.

If your team is also thinking in a similar direction, I suggest starting with a few low-barrier entry points:

Integrate an AI security scanning tool into your IDE to establish a quality baseline.
Add structured data markup and multimodal versions (videos, podcasts) to your existing content.
Review your checkout process and prioritize fixing the biggest friction points.

Only by mastering both technological leverage and market trends can we truly define and capture new value in the AI wave.