AI Made Me 10x Faster—Here's What I Had to Change

I've been working as an IT engineer in multiple industries for over 25 years, from .Net developer, BI developer, data architect, data scientist, and finally to AI solutions architect. Recently, the team in my organization is developing and using AI programming tools and has achieved 8x to 20x code output efficiency compared to ordinary high-performing teams. See this LinkedIn post for more details.

Reading this, you might think: Wow, are programmers going to lose their jobs? Is AI going to replace humans?

But my view is exactly the opposite. As a frontline professional programmer, I'm telling you responsibly: When your speed increases 10x, the risks and bottlenecks you face may also be magnified 10x.

I'll be honest - even I was skeptical at first. Could we really sustain this pace without everything falling apart?

What this means is: AI has fundamentally changed how "costs" and "benefits" are calculated in software engineering. But to truly enjoy this improvement, the entire software development system needs to be upgraded simultaneously. This insight applies not only to programming but has profound implications for everyone who uses AI tools.

1. AI Hasn't Made Programmers Unemployed, But Has Fundamentally Changed How People Work

Here's how my team works:

In the code our team submits, 80% - 90% is written by AI. But this is definitely not that casual "Vibe Coding (i.e., coding without thinking)", which is not a good coding practice. This workflow is called "Agentic Coding" (i.e., coding with AI agents).

In this model, AI plays the role of a "highly capable but irresponsible junior programmer."

And the human engineers? They are "experienced tech leads or architects."

Specifically, the engineer's workflow has become:

1. Design, plan and break down tasks - Figure it out yourself first, or brainstorm with AI
2. Give AI instructions - Clearly tell AI what to do
3. Review and refine AI's output - This is the most critical step
4. Iterate repeatedly - Until completely satisfied with the quality
5. Submit and take full responsibility - Ultimately, humans are still responsible for the code

See that? The role of humans hasn't diminished; it's become more important. The focus of work has just shifted from "writing code by hand" to "defining requirements" and "code review."

Think about your own work: What would change if 80% of your output came from AI?

An analogy: Previously you were a worker carrying bricks on a construction site. Now you're an operator commanding an excavator. Although you no longer carry bricks with your own hands, your judgment, operational skills, and responsibilities have actually increased.

2. Speed Increases 10x, Accident Rate May Also Increase 10x

When you're speeding down a track at 200 km/h, you need massive "downforce" to keep the car firmly planted on the ground. Otherwise, you'll fly off at the first curve.

In software engineering, "flying off" means bugs and system crashes.

Some alarming data:

• Before: A team might only encounter one or two severe production bugs per year
• Now: When you're submitting code at 10x speed, even if the probability of bugs stays the same, the absolute number of bugs you encounter will also increase 10x

What does this mean?

Incidents that used to happen once a year might now happen every week. Imagine explaining to your boss why production went down every Monday.

This kind of "accident rate" is unbearable for any team. Yet many people promoting "AI omnipotence" have intentionally or unintentionally ignored this problem.

To enjoy the 10x coding speed boost from AI, you must also find ways to reduce the "probability of problems" by 10x, or even more.

Having a good engine isn't enough; you also need a better braking system.

3. The True Value of AI: Making "Good but Expensive Methods" Affordable

So how do you reduce risk while increasing speed?

AI isn't just about letting you write faster; it's about making those "good but too expensive" best practices in software engineering become affordable and feasible.

3.1 Change #1: Build a "Wind Tunnel Testing" Environment

What is wind tunnel testing?

Just like building an airplane - before it actually takes flight, the model is put in a wind tunnel to test various extreme conditions.

In software development, this means building a "high-fidelity simulation environment" locally. For example, if your system depends on 10 external services (databases, authentication, payments, etc.), you run or simulate all 10 services locally.

This way, on your computer you can run complete end-to-end tests, and even simulate various extreme failure scenarios.

This kind of testing can catch a lot of bugs hidden in the "cracks between components."

Why didn't we do this before? Too expensive!

Simulating and maintaining these services was too much work, so most teams gave up.

Why can we do it now? AI excels at this!

AI is very good at writing these simulation services with clear logic and well-defined behavior. Especially by using AI agents with Model Context Protocol (MCP) and Agent2Agent Protocol (A2A), we can easily build a complete local "wind tunnel" for our fairly complex system.

Work that used to take weeks or even months can now be done in days.

3.2 Change #2: Upgrade CI/CD (Continuous Integration/Deployment)

In the early days of waterfall development, everyone developed separately and then integrated after development. The result was a pile of problems during integration, taking a long time to stabilize.

Later, the concept of "continuous integration" became popular:

The earlier you integrate, the earlier you get feedback. The more frequently you integrate, the more you can reduce problem complexity.

Now, CI/CD is recognized as the best practice in software engineering. But not many teams actually do it well, because building and maintaining it is still not cheap.

What's worse is that many teams, despite having CI/CD, have extremely time-consuming processes. One code commit, waiting for all tests and deployment to run through - at minimum ten minutes, sometimes several hours.

Before AI, these problems weren't obvious. Now that AI is more capable, they've become obstacles.

So CI/CD also needs to be upgraded along with it, compressing the feedback loop from "hours" to "minutes." You need infrastructure that's fast to an absurd degree, able to discover, isolate, and roll back problematic changes within minutes.

3.3 Change #3: Decision-Making and Communication Systems Also Need Upgrading

10x code output means you also need 10x or more communication and decision-making efficiency.

Before, developing a system required various meetings, lengthy discussions, and only then could work begin. After all, you had to depend on other people's modules, so you had to define agreements first, otherwise you couldn't integrate later.

Various technical decisions also required repeated discussion for a long time, because development costs were high back then, and if decisions were wrong, the cost of rework was too great.

But now, if we still have the same communication efficiency as before, it will greatly drag down overall efficiency.

Perhaps the most efficient approach is to minimize communication as much as possible, letting everyone do their tasks as independently as possible from others.

For example, microservices architecture might be a good choice in the AI era.

For technical decisions, now you can actually have more opportunities to experiment. You don't need to be as rigorous as before in repeatedly verifying technical decisions. Because development costs have decreased, the cost of experimentation has also decreased.

4. Insights for Everyone

AI isn't a "stimulant" that makes you run faster; it's giving you a "supercar."

But the question is: Are you ready to drive it?

4.1 Tool Upgrade Doesn't Equal System Upgrade

Using AI is like upgrading your car with a brand new engine. If you just install it on your old "vintage car," what you get isn't 10x speed, but 10x problems.

This principle applies to each of us.

When you learn to use AI tools (ChatGPT, Gemini, Claude, Midjourney, various AI assistants), don't assume your work efficiency will automatically improve.

Your workflows, quality inspection mechanisms, and collaboration methods all need to be adjusted accordingly.

For example:

Using AI to write code is fast, but if you don't have a rigorous review process, you might produce a lot of low-quality, buggy code.

Using AI to quickly generate investment advice - but has your risk assessment ability kept up?

Using AI to make quick decisions - but have you established a review and error-correction mechanism?

4.2 Speed Increase Must Come with Risk Management Improvement

The "accident rate paradox" doesn't only exist in programming.

• The faster food delivery, the higher the traffic accident risk
• The faster product iteration, the more quality issues there might be
• The faster decision-making, the greater the probability of making mistakes

So don't blindly pursue "fast." Ask yourself: Has my "braking system" been upgraded?

Build your own "wind tunnel testing": Try on a small scale first, rather than pushing forward comprehensively right away.

4.3 Re-examine Those "Good but Expensive" Methods

Here's what finally clicked for me after months of using AI tools:

The true value of AI isn't just about "writing faster"; it's about making those "good but too expensive" best practices become affordable and feasible.

This realization made me re-examine many good habits I had abandoned because they were "too troublesome":

• Personal finance management: Keeping track of expenses used to be too troublesome. Now AI can help you automatically categorize and analyze
• Learning new skills: Hiring a private tutor used to be too expensive. Now AI can provide personalized tutoring
• Health management: Nutritionists used to be too expensive. Now AI can customize meal plans
• Content creation: Making videos used to require a team. Now individuals can also produce high-quality content

The key is: Don't just use AI for "fast production." Use it to achieve "things you wanted to do before but couldn't."

4.4 Your Role is Changing

In programming, my role has shifted from "executor" to "decision-maker + quality inspector." But this pattern applies everywhere:

• Writers are becoming editors who review and refine AI drafts
• Designers are becoming creative directors who guide AI-generated concepts
• Analysts are becoming strategists who interpret AI-processed data
• Managers are becoming orchestrators who coordinate AI-assisted workflows

The common thread? Responsibilities haven't decreased; they've actually increased.

In the AI era, your core competitiveness is:

1. Judgment - Being able to distinguish good from bad AI output
2. Questioning ability - Being able to give AI clear instructions
3. Sense of responsibility - Being willing to take responsibility for the final results
4. Systems thinking - Understanding the entire process, not just one part

Many people can use AI to write reports, but those who can review AI's logical flaws are valuable.

Many people can use AI to design solutions, but those who can judge a solution's feasibility are irreplaceable.

4.5 Build Your AI Work System

For us ordinary people, don't just focus on "individual AI tools." Build your own "AI work system":

1. Input system - How to quickly and accurately provide AI with information and instructions
2. Quality inspection system - How to efficiently review AI's output
3. Feedback system - How to iterate and improve quickly
4. Knowledge management - How to accumulate and reuse experience from AI collaboration

For example:

It's not just about using ChatGPT; you also need to build your own prompt library, review checklist, and iteration workflow.

It's not just about using AI for image generation; you also need to establish a style library, quality standards, and version management.

This is the true way of working in the AI era.

5. Summary

The AI era requires "systems thinking," not "tool thinking."

Many people treat AI as a "fast production tool," hoping to use it to accelerate existing work.

But those who truly understand how to leverage AI treat it as an "opportunity for system upgrade," rethinking the entire workflow.

AI isn't just about upgrading the car's engine; it's also about upgrading the roads the car frequently drives on.

The goal for veteran drivers isn't to be replaced by AI, but to help them adapt to the new high-speed engine, giving them a comfortable and safe driving environment.

So when AI increases your speed 10x or 20x, don't rush to celebrate.

First ask yourself:

• Has my quality inspection mechanism been upgraded?
• Has my risk management ability improved?
• Has my workflow been restructured?
• Am I ready to take full responsibility for the results?

Remember: You are the driver who is responsible for the final results.

AI just gave you a supercar, but whether you can arrive at your destination safely and efficiently depends on your driving skills and road conditions.

May we all become good drivers in the AI era - ones who can step on the gas, and also know when to brake.