How AI Is Changing My Work and Life
Before We Begin
I’m not an AI expert.
I’m just an entrepreneur, someone willing to “waste” his energy on new things.
This article is not a tutorial or a guide. It is just an honest record of how I’ve used AI, and what I’ve taken from it.
If you’re new to AI too, I hope my experience gives you something to refer to.
---
Phase One: OpenAI as a Writing Assistant (2023-2024)
How I First Used It
When OpenAI first came out, I treated it as a copy editor.
Every product description at the company now starts as an AI first draft.
Writing a product description used to look like this:
- Hire a copywriter: €50-100 per piece
- 2-3 rounds of revisions
- Time taken: 2-3 days
Now:
- AI first draft: 5 minutes
- Human edits: 10 minutes
- Cost: almost zero
An Unexpected Bonus: Multilingual Translation
The bigger surprise was multilingual translation.
France is a multilingual country, and our customers include:
- Local French people (French)
- Chinese merchants (Chinese)
- Immigrants from other countries (English, Spanish, etc.)
Before:
- We needed employees fluent in several languages
- Translation cost: €0.10-0.20 per word
- Products went online slowly (waiting for translations)
Now:
- AI translates automatically
- Quality: 85-90% usable
- Human proofreading: 10 minutes per product
- Language requirements for employees: no longer a hard requirement
What does that mean?
It means that when we hire, we no longer need someone who is fluent in French, Chinese and English.
It means products go online 3 times faster.
It means our SEO can cover French, Chinese and English versions at the same time.
What I Took Away from This Phase
What AI can do:
- First drafts of text
- Multilingual translation
- Basic polishing
What AI can’t do:
- Make final decisions
- Get the brand tone right
- Understand complex context
My advice:
- Treat AI as an assistant, not a replacement
- Use AI for the first draft, and have a human review the final version
- Translation is fine to use, but proofread anything important
---
Phase Two: AI Photo Retouching (2024-2025)
A Change in Photography
Next, AI took over the work of the photographer.
Shooting product photos used to mean:
- Professional equipment: €5000+
- A photo studio: €1000 per month
- A photographer: €200 per day
- A retoucher: €50 per photo
- Time: 10-20 photos per day
Now:
- Equipment: just a phone
- AI automatic repair: €0.01-0.05 per photo
- Time: 100-200 photos per day
- Quality: e-commerce standard
Workflow
1. Snap it casually - no lighting or backdrop needed 2. AI repair - automatic cutout, color correction, background added 3. E-commerce standard - white background, uniform size, consistent quality
What I Took Away from This Phase
What AI can do:
- Basic retouching
- Batch processing
- Standardized output
What AI can’t do:
- Creative shoots
- High-end commercial photography
- Complex scenes
My advice:
- E-commerce product photos can absolutely be done with AI
- High-end promotional images still need a professional photographer
- Use AI for batch processing, and humans for showcase images
---
Phase Three: Exploring AI Video (2025-2026)
Where Things Stand
AI video is still at the exploration stage.
Problems:
- Length limits (mostly under 1 minute)
- Unstable quality
- High cost
- A steep learning curve
Where it works:
- Short teaser videos
- Simple product demos
- Social media content
Where it doesn’t:
- Brand films
- Complex product introductions
- High-quality content
What I Took Away from This Phase
My assessment:
- 2026 is the year of exploring AI video
- 2027 may bring a breakthrough
- 2028 is when it can really go into production
My advice:
- Keep an eye on it, but there’s no rush
- Try it out on a small budget
- Don’t expect it to replace a professional video team
---
Phase Four: OpenClaw (After the 2026 Spring Festival)
Out of Nowhere
After the Chinese New Year in 2026, OpenClaw burst onto the scene.
Honestly, I didn’t understand it at all at first.
A Month of Torment
I put myself through almost a month of torment before I got reasonably familiar with it and stable.
Why it was hard:
- Too many new concepts
- The documentation is incomplete
- It takes some technical understanding
- Stability isn’t strong enough
But I stuck with it, because:
- I saw the potential
- It can solve real problems
- I didn’t want to be left behind by the times
In Practice
Perhaps because OpenClaw is a free tool, its stability isn’t strong enough for it to be put into real production use.
But here is what it can do:
- Handle the company’s test data
- Run brand-new projects
- Automate repetitive tasks
- Quickly launch new ideas
From Chat Tool to Software That Executes
This is the biggest shift:
AI before:
- Chat
- Give advice
- Write text
- But couldn’t execute
AI with OpenClaw:
- Can execute directly
- Can call tools
- Can manage files
- Can automate processes
What I Took Away from This Phase
How I feel:
- The direction is right
- The product isn’t mature enough yet
- It needs investment from big companies
- I expect a breakthrough by the end of the year
My assessment:
- By the end of 2026, many apps will integrate this kind of AI
- In 2027, supporting hardware will really go into production
- In 2028, it will become standard for businesses
---
Phase Five: Thoughts on How Jobs Will Change
Secretaries and Administrative Staff
I have a vague feeling that many secretarial and administrative workers will be replaced.
Not all of them, but many.
Why:
- Writing documents: AI can do it
- Organizing meeting minutes: AI can do it
- Scheduling: AI can do it
- Replying to emails: AI can do it
The work that remains:
- Complex communication
- Personal relationships
- Creative work
- Decision support
Accounting Jobs
Jobs like accounting, which follow rules, are also at risk.
Why:
- Bookkeeping: the rules are clear
- Tax filing: the process is fixed
- Reconciliation: the logic is clear
- Reports: the formats are standard
AI’s advantages:
- Doesn’t get tired
- Doesn’t make mistakes
- Works 24/7
- Costs less
The work that remains:
- Tax planning
- Financial analysis
- Decision support
- Auditing and oversight
My Advice
For people in these jobs: 1. Learn AI tools 2. Build up your creativity 3. Strengthen your relationships 4. Move toward decision support
For business owners: 1. Don’t rush to lay people off 2. Train first, then transform 3. AI + people > AI or people 4. Pay attention to your employees’ development
---
Phase Six: Codex, the Tool I Really Needed
The Encounter
While exploring OpenClaw, I came across Codex.
This is the tool I really need at this stage.
Why Codex
OpenClaw’s problems:
- Not stable enough
- Needs technical understanding
- Hard to debug
- Errors are hard to track down
Codex’s strengths:
- More stable
- Easier to pick up
- Has coding ability
- Can debug
What I Actually Built
At the end of last year, I had planned to move all the company’s various management functions over to Odoo.
But once I had Codex, I realized:
I can build a management platform that fits my company myself!
Why It’s Worth It
Cost comparison:
- Outsourced development: €20,000-50,000
- Buying SaaS: €500-1000 per month
- Building it myself (with Codex): the cost is my time
Advantages:
- Exactly the way I want it
- Fits how the company really works
- Can be adjusted at any time
- No dependence on outsiders
Challenges:
- It takes a lot of energy
- You need to learn some programming basics
- You need to debug and test
But I think it’s worth it, because:
- The process of building it is itself rewarding
- I rather like “wasting” my energy on new things like this
- What I learn belongs to me
- The system I build is one of a kind
---
Codex + OpenClaw: The Best Combination
What I Found
After using them for a while, I found that:
Codex + OpenClaw = the best combination
Why
OpenClaw’s strengths:
- Multi-Agent collaboration
- Task automation
- Tool integration
- Fast prototyping
OpenClaw’s weaknesses:
- Not stable enough
- Errors are hard to track down
- Needs technical understanding
Codex’s strengths:
- Strong at code
- Can debug
- More stable
- Easy to pick up
Codex’s weaknesses:
- Weak multi-Agent collaboration
- Few tool integrations
- Weak automation
Using Them Together
My workflow: 1. Codex - write code, debug, design architecture 2. OpenClaw - run tasks, automate, multi-Agent collaboration
Results:
- Codex solves OpenClaw’s instability
- OpenClaw solves Codex’s automation problem
- For people without a programming background, it saves many detours
What I Took Away from This Phase
My advice:
- Beginners should start with Codex
- Try OpenClaw once you’re comfortable
- Use the two together
- Don’t expect one tool to solve every problem
---
About Claude
I’ve Heard It’s Good
People online say Claude is pretty good.
My Choice
But I haven’t tried it myself.
Why:
- Codex is already enough
- There is still a lot to explore in OpenClaw
- My energy is limited
- I don’t want to learn too many tools
My Advice
For beginners:
- Master one tool first
- Then consider others
- Don’t try to do too much
- Depth > breadth
For advanced users:
- You can try Claude
- Compare different tools
- Find what suits you best
- Build a combination of tools
---
In Summary: My Philosophy of Using AI
1. AI Is a Tool, Not the Goal
I use AI to solve problems, not to show off.
A tool that solves problems is a good tool.
2. “Waste” Your Energy Where It’s Worth It
Some people say I’m “wasting” my energy on these new things.
My answer:
- Learning AI is not a waste
- Building systems is not a waste
- Improving yourself is not a waste
What is truly a waste is spending your time on:
- Meaningless meetings
- Repetitive labor
- A comfort zone where you don’t grow
3. Stay Curious, Keep Learning
AI is moving so fast.
Today’s new tool may be obsolete tomorrow.
The only constant is the ability to learn.
4. Practice > Theory
I’ve looked at plenty of AI tutorials, but I feel that:
Doing it yourself once beats watching ten tutorials.
My articles are not tutorials, just a record of practice.
You don’t need to copy it exactly, but you can borrow my way of thinking.
5. AI + People > AI or People
I don’t believe AI will completely replace people.
But I do believe:
People who can use AI will replace people who can’t.
---
Final Words
This article took a week to write.
I used AI for the first draft, edited it by hand, had AI polish it, then adjusted it by hand again.
This is how I work now:
AI does 80%, humans do 20%.
That 20% is decisions, review, tone, and responsibility.
The 20% AI can’t replace.
I hope my experience gives you some inspiration.
If you’re exploring AI too, I’d love to hear from you.
I’m not an expert, just a fellow traveler.