AI is a technology, not a product: Why blindly stacking AI will reduce efficiency

Core conclusion

In mid-May 2026, John Gruber of Daring Fireball published an article "AI Is Technology, Not a Product", which received 300+ views on Hacker News and was discussed throughout the site. At almost the same time, engineer Frederick Vanbrabant published "I don't think AI will make your processes go faster" which received 500+ votes, as well as the "Most Americans Don't Trust AI" survey published by The Verge, which together formed the most concentrated wave of AI reflection in the first half of 2026.

Key Points

• Event Time: 2026-05-15 to 2026-05-18
• Core Source: Daring Fireball / Hacker News / The Verge
• Influence Group: Content automation operators, AI Agent developers, enterprise IT decision-makers
• Core point: AI is an infrastructure-level technical component, not an independent product - companies that use "AI products" as their selling point often pass on UX problems to users

Background and trigger events

Gruber pointed out at the beginning of the article: "AI is not even a feature. It's just technology." - This does not mean that a certain AI product is not good, but that the entire industry markets the technology stack as a product.

Vanbrabant's article adds to the same argument from a more pragmatic perspective: he provides an in-depth analysis of why AI will not automatically speed up processes - because you have to figure out "what you want to speed up" first. If your business process is inherently flawed, adding a layer of AI will just make it go wrong faster.

There was also an eye-catching comparison on the same day: OpenAI and the Maltese government reached a cooperation to push ChatGPT Plus to citizens across the country (316 votes), demonstrating another possibility of AI as a "public service infrastructure"; on the other hand, a poll reported by The Verge showed that most Americans do not trust AI at all.

Popularity score: 85/100 — The core viewpoint touches on the most fundamental business model issues in the AI ​​industry. SEO search volume is high, and the topic is inherently controversial.

Key Impact

Adaptation suggestions

The real value of this discussion is not to take sides on "Is AI good or not?" but it forces us to think about a very pragmatic question: **After you spend money and time to connect to AI, has the quality of your output really improved? **

1. Verify the process first, then feed AI — If your content production process itself is unstable, adding AI will only amplify the instability. It is recommended to first use quality gate to verify the output of each step.
2. Focus on "Token efficiency", not just "ability" — The concept of Semble (189 votes, open source on GitHub) is worth learning from: using 98% fewer Tokens to implement code search proves that efficient resource utilization is more important than stacking models.
3. Don’t build “AI products”, build “AI-enhanced infrastructure” — Integrate AI capabilities into the existing tool chain (n8n + AI Agent = two independent components of the automated pipeline), rather than building an “AI automation” shell product.

Task List

• Review your content pipeline: Is the output quality of each step guaranteed? Or does it all rely on AI to take care of it?
• Check whether the Token consumption in the tool chain is efficient (refer to Semble’s idea)
• Avoid using over-promising expressions such as "AI revolution" and "AI crushing" in future content

Further reading

• HN: I don't think AI will make your processes go faster
• HN: Semble – Code search for agents
• The Verge: Most Americans don't trust AI

Tool entry

The key tools and technology stacks involved in this discussion are the focus of the debate between the pros and cons: Claude Code, OpenAI, ChatGPT, n8n, DeepSeek, Gemini, LangGraph - there is no "best AI", only tools that work when placed in the right process.

Technical perspective: Why AI slows down more than it speeds up

Vanbrabant's article is not a general discussion, but analyzes the problem from a practical technical perspective. He gave a typical example: adding AI code review to a CI/CD pipeline that already has bugs. The result is that the AI ​​generates hundreds of false positives at a time, and engineers spend more time filtering through the noise—**What should have been a 5-minute manual review turns into 30 minutes of AI misdirection. **

# AI
code_review(pre_push_hook=True) #

# AI +
code_review(ai_agent=True) # +

Core lesson: **Only if you already know exactly "what to do", AI can help you do it faster. **

Internal link guidance

The core lesson of this discussion is how to avoid tool backlash. If you are using AI to build an automated content pipeline, it is highly recommended to read:

• **Want to learn how? ** See: AI Agent drives automated website operations: Build a fully automatic content pipeline in 30 minutes — teaches you how to build structure first and then AI, not the other way around
• Practical case: He Built an AI Automation Stack with Claude + n8n — $4K to $12K/mo in 6 Months — See how real entrepreneurs treat AI as a tool rather than a product
• Technical selection reference: How to add quality gates to your AI automation workflow: A practical guide from output to trustworthy results — A practical guide to avoid AI blind boxes