AI Agent & Coding ·
Ten crooked workflows.
English translation of the Chinese original. This version is generated for international readers and may be refined over time.
English translation of the Chinese original. This version is generated for international readers and may be refined over time.
Date: 2026-04-14
Introduction: Where's the time?
Every month, you have about 45 hours to disappear.
There is no big deal to swallow - there is no emergency crisis, there is no major decision-making, there is only piecemeal, repetitive work that could have been done automatically. They're like sand in a sandglass, and a grain leaks, and a month goes by when you come back.
This is the first reaction I've seen when I saw Zodchiii's data. He spent two weeks tracking every moment he used Claude: a time stamp, a job type, a manual processing time for vs AI. The result was straightforward -** a weekly savings of 11. 4 hours per month, 45+ hours**, equivalent to 2 full working days.
And what he did was interesting in itself:** no API, no complex integration, no terminal command.** Just subscribe to the Claude web version Pro with 10 well-designed workflows.
That's what we're talking about today -- not how to use AI, but how to use AI systematically.
Why is the sporadic hint a pit
Let's start with a common error zone. A lot of people use AI to do that: think of one thing, throw a word out; think of one thing, throw another. There are occasional surprises, and most of the time there are regrets - the output is either too broad, or the key messages are missing, or the format is so messy.
It's not AI's problem, it's the way it's used.
There are a few fatal injuries in the sporadic hint:
- ** Unreplicable**: A reminder is written today, and a rewrite is required for a similar mission tomorrow. I can't remember all the details. More versions are messy. -** Unstable results**: Same type of mission, this output A, next output B, all by luck.
- ** always calibrated**: You have to spend time explaining background, format requirements, target readers - these frictions accumulate more than "no AI".
Zodchiii ' s insight is accurate:** workflow is not a hint, it is a system**. Fixed input formats, fixed output structure, predictable results, generic solutions for similar tasks.
In other words, to change "the wheel has to be recreated" to "constructed, used again."
10 high ROI workflows
Here's 10 workflows that are extracted from actual usage. In each case, the amount of time saved was shown and was measured by Zodchiii.
- Content research ** Compressed from 4 hours to 1. 5 hours, saving 2 hours 30 minutes per week**
This process solves the "research hell" before it's written -- it takes half a day to flip through information, find data, figure out what's true.
English:
Here are my sources on [TOPIC]:
[PASTE ALL RAW MATERIAL]
Extract:
1. The 5 facts that matter most for my audience (builders, not consumers)
2. Anything that contradicts the common narrative
3. Specific numbers: stars, users, funding, benchmarks
4. One angle nobody else is covering
If two sources disagree, show me both sides.
Don't summarize fluff. Only signal.
** Chinese version:**
Here are some of the original materials I collected on the subject:
[ pasting all original content]
Please extract:
1. Five facts of the utmost importance to my audience (my audience is a practitioner, not a bystander)
Any information that contradicts the mainstream narrative
3. Specific figures: GitHubstar number, number of users, amount of financing, performance baseline
4. An angle not currently covered
If the two sources contradict each other, the views of both sides are presented.
Don't sum up empty phrases, just extract valuable information.
Key points: ** Just paste the original material and let AI do the extraction instead of making it "just talk."** Fuzzy demands produce vague results and the given structure is structured.
- Rapid research ** Cut from 2 hours to 15 minutes, saving 1 hour 45 minutes per week**
This process is much more efficient than Google searching for and reading 10 articles when you don't know something.
English:
Research [TOPIC]. Structure:
1. Executive summary (3 sentences max)
2. Key findings (top 5, ranked by impact)
3. What's missing: gaps in available info
4. Sources with URLs
If data is insufficient for any claim, say so.
Don't speculate. Don't pad.
** Chinese version:**
Research [theme], structured as follows:
Executive summary (up to 3 sentences)
2. Key findings (Top 5, in order of impact)
3. Missing information: gaps in available information
Source (with URL)
If the data are not sufficient to support a conclusion, it is true.
Don't guess, don't fill in crap.
The value of this template lies in the mandatory simplicity**. 3 Finish the core view and then the details. If you find AI in "fill the crap," it means that the question you ask is space-wide.
3. Reconstruct content
** Compressed from 2 hours to 20 minutes, saving 1 hour 40 minutes per week**
After writing an article, the most painful thing is not writing itself, but rewriting it into different formats and posting different platforms. This workflow allows AI to help you with multiple channels like Twitter/X, LinkedIn, Newsletter.
English:
Here's my article: [PASTE OR UPLOAD]
Create:
1. A 2-sentence hook for X (include a specific number or claim from the article)
2. A 4-paragraph TG post with the key insight
3. A provocative quote-tweet caption (1 sentence)
4. 3 standalone insights that work as separate tweets throughout the week
Each piece must work independently.
Someone who never read the article should still get value.
** Chinese version:**
Here's my article:
Please create:
1. A two-word hook of x tweets (must contain a specific number or view in the article)
2. A Telegram post highlighting core insight
A cross-reference to the discussion (one sentence)
4. 3 Independent insights, which can serve as separate tweets for the current week
Each content must be read independently.
Even if the other party had not read the original text, it would be worth it.
** A fish eats too much.** Good content deserves to be distributed on multiple platforms, but manual adaptation is a black hole of time. This process ensures that each version is independently readable, not simply "replicate paste".
This process can also** reverse operations** - integration of scattered posts into growth articles:
English:
Here are 5-6 short TG posts from this week: [PASTE ALL]
Find the connecting thread and draft a long-form article outline.
** Chinese version:**
The following are 5-6 short posts for this week:
Find a link between them and draft a long outline.
4. Data Insight
** Cut from 2 hours to 20 minutes, saving 1 hour 40 minutes per week**
In the face of a data set, most people either don't know where to start, or just remember the numbers. This process allows AI to make a preliminary analysis before you decide where to dig.
English:
Analyze this data. I need:
1. Top 3 trends over time
2. Anything unusual or unexpected
3. Correlations between [COLUMN A] and [COLUMN B]
Table first, then a 2-paragraph summary explaining
what this means in plain English.
If the dataset is too small for a conclusion, say so.
** Chinese version:**
To analyze this data, I need:
Top 3 trends in time dimension
2. Any unusual or unexpected discovery
3. Relevance between [column A] and [column B]
The table, followed by two paragraphs summing up, explains what this means in common language.
If the data set is too small to draw conclusions, it is true.
** The tables are available for reading ** - the order is important. The tables are original information and read as cognitive conclusions. It's easy to mix it up into "seems but unverifiable" crap.
Code review ** Compressed from 2 hours to 15 minutes, saving 1 hour and 45 minutes per week**
The code review is a technical exercise, but most of the review time is actually spent on finding conventional problems - security loopholes, code styles, performance hazards. AI's good at this and not tired.
English:
Review this code for:
- Security issues (exposed keys, injection, XSS)
- Logic errors and edge cases I might have missed
- Performance problems
- Anything that would make a senior dev uncomfortable
For each issue: severity (Critical/High/Medium/Low),
exact location, why it matters, and the corrected code.
Be harsh. "Looks good overall" is not helpful.
[PASTE CODE]
** Chinese version:**
Review this code, focusing on:
- Security issues (discovered keys, leaks, XSS, etc.)
- Logic error and the border I might have missed.
- Performance issues.
- Anything that makes senior engineers uncomfortable.
Please specify the severity of each question (Critical/High/Medium/Low),
Specific location, why important, and amended code.
Be tough. "The whole looks good." That feedback doesn't help.
[ pasting code]
There's a mindset to adjust: code review isn't "prove I'm careful," it's "discovery real problems." AI helps you filter out 80% of the routine problems, and you can focus on judgment at the structural level.
- GitHub repository analysis ** Compressed from 1. 5 hours to 20 minutes, saving 1 hour 10 minutes per week**
In order to write an article on the AI tool, you need to select items from over 1, 000 warehouses that really deserve to be mentioned. The manual check takes 50+ hours, and AI helps you downsize the process.
English:
Here's a list of GitHub repos with descriptions:
[PASTE BATCH]
For each repo, evaluate:
1. What it actually does (1 sentence, no marketing speak)
2. Traction signals: stars, recent commit activity, contributor count
3. Category: agent framework / dev tool / MCP / infrastructure / other
4. Worth featuring? Yes/No with one reason
Skip anything that's just a wrapper, a tutorial repo,
or has no commits in 30+ days.
Sort the "Yes" picks by most interesting first.
** Chinese version:**
Below is the GitHub repository list and description:
Assessment of each warehouse:
1. What does it actually do (1 sentence, no marketing language)
2. Growth signals: number ofstars, recent submission of activity, number of contributors
Classification: Agent Framework / Development Tool / MCP / Infrastructure / Other
4. Are there recommendations? Yes/No, with one reason
Skips items that are only packaging layers, tutorial stores, or unsubmitted items for more than 30 days.
Sort the "recommended" option from high to low value.
Process 50-100 warehouses in batches, automatically filtering abandoned items (30 days without submission). This process reduced the number of manual inspections required from 1000 to 80.
7. Competing analysis
** Compressed from 1 hour to 15 minutes, saving 45 minutes per week**
It usually takes an hour to analyze the content, location and audience of the competition, and AI helps you quickly to build structured awareness.
English:
I'm analyzing [COMPETITOR/ACCOUNT].
Based on what you know + the data I'm providing:
1. Top 3 things they're doing well (be specific)
2. Gaps or weaknesses in their approach
3. What I can learn from them
4. How my positioning is different
About me: I write about AI tools, vibe coding, and crypto
for builders. TG + X.
Don't say "they have a strong brand."
Tell me WHY and what specifically makes it work.
** Chinese version:**
I'm analyzing [competing/account].
Based on what I know + what I've provided:
1. Three things they do best.
2. Gaps or weaknesses in their approach
What can I learn from them?
What's the difference in my position?
About me: My audience is AI Tool Practice, Vibe Coder,
Crypto Builder. The distribution channel is Telegram+X.
Don't say "they have very strong brand" crap.
Tell me why.** What specific factors make their approach work.
Must provide the context of who I am. Without this, you get a general SWOT analysis, not targeted insight.
- Morning briefing ** Compressed from 45 minutes to 5 minutes, saving 40 minutes per week**
It takes 45 minutes every morning to brush X "Keep the update" and be disturbed by all kinds of noise. AI helped you filter out 95% of the noise.
English:
3-minute briefing:
1. Top 3 AI news from last 24 hours (one sentence each)
2. Crypto: major moves, liquidations, new narratives
3. Anything I should know before posting content today
Be specific: names, numbers, links.
Skip anything that isn't genuinely important.
3 real updates > 10 filler items.
** Chinese version:**
Give me a three-minute briefing:
Top 3 AI News (each sentence)
Crypto dynamics: main situations, liquidations, emerging narratives
3. What needs to be known before today ' s release
Specifically: Name, number, link. Select only information that really matters.
3 Real update > 10 piece of crap.
Why is it better than just painting social media: eliminating noise and distraction, providing only really important information, presenting it in a structured way, saving 40 minutes a day?
9. Mail drafting
** Compressed from 30 minutes to 5 minutes, saving 25 minutes per week**
It takes 15 minutes to think about the tone -- too formal, too free to be a robot. AI helps you with the language.
English:
Draft an email.
To: [NAME + how I know them]
Goal: [WHAT I WANT THEM TO DO]
Tone: professional but sounds like a real person
Max: 5 sentences
Context: [THE SITUATION]
Does not sound like: a cold pitch template, corporate speak,
or something ChatGPT would write.
No "I hope this email finds you well."
** Chinese version:**
Draft an email.
Recipient: [name] + How do I know them?
Target: [What I want them to do]
Language: Professional but sounds like real people.
Length: Up to 5 words
Background: [specific]
Don't write it in: cold template packages, business blacks,
Or ChatGPT would write that kind.
Don't start with "I hope this e-mail is all right".
The key to success is to say what you want, not let AI guess. "Write an e-mail" is a vague order, and it is an enforceable order to write an e-mail that is firmly worded, but not disrespectful, and requires confirmation of the programme by Friday.
10. Weekly review
** Compressed from 1 hour to 30 minutes, saving 30 minutes per week**
It takes an hour to organize a week's notes and ideas, and it's easy to miss out on important models. AI helps you look at yourself from God's perspective.
English:
Here are my notes and ideas from this week: [PASTE EVERYTHING]
Help me:
1. Find patterns: what topics am I gravitating toward?
2. Which 3 ideas have the most content potential?
3. What am I ignoring that I shouldn't be?
4. Content plan for next week: 3 TG posts + 1 article topic
Be honest. If an idea is weak, say so.
Don't tell me everything is great.
** Chinese version:**
Here are my notes and thoughts this week:
Help me:
1. Finding a pattern: What topics am I attracted to?
2. Which 3 ideas have the greatest potential for content?
3. What am I neglecting?
4. Next week 's content plan: 3 Telegram posts + 1 selected articles
Be honest. If something's weak, just say it.
Don't tell me everything's great.
AI's value as a think partner: identifying models and connections that you might ignore, helping you to see a bigger picture, providing honest feedback rather than meaningless praises, planning in advance for next week's content, avoiding anxiety about writing.
Time saving data: spending every minute on blades Let's go.
| Workflow | Time required for manual operations | AI processing time | Weekly savings |
|---|---|---|---|
| Content research | 4 hours | 1. 5 Hours | 2 hours 30 minutes |
| Rapid research | 2 hours | 15 min | 1 hour 45 minutes |
| Data Insight | 2 hours | 20 min | 1 hour 40 minutes |
| Code Review | 2 hours | 15 min | 1 hour 45 minutes |
| GitHub Warehouse Analysis | 1. 5 Hours | 20 min | 1 hour 10 minutes |
| Content Reconstruct | 2 hours | 20 min | 1 hour 40 minutes |
| Competing analysis | 1 hour | 15 min | 45 min |
| Morning briefing | 45 min | 5 min | 40 min |
| Mail Drafting | 30 min | 5 min | 25 min |
| Weekly review | 1 hour | 30 min | 30 min |
| Total** | ** ~ 16 hours** | ** ~3 hours** | ** About 13 hours/month** |
Note: The above figures are Zodchiii personal measurements. Your results depend on the nature of the work and the degree of friction of the current process. But even with 3-4 workflows, it is realistic to save 5+ hours per month.
Deep insight: the truth of time
From this track, Zodchiii came to a deeper conclusion than saving time:** Most jobs are friction.**
Where did our time go?
-** Context Switch**: Open tab, reread, reposition - 15-20 minutes per switch
- ** Duplicate **: Same information sorting over and over - Every time it's "reinventing the wheel."
- ** Noise filter**: find the valuable one from a lot of irrelevant information ** A needle.
- Formatization of labour: change the same content to a different format - low technical content but extremely time-consuming
AI changed what?
AI did not replace human work, it replaced friction around work. The parts of humanity that are truly needed - reflection, decision-making, editing - remain, but the efficiency of those parts has increased dramatically.
** Core correction**: A lot of people say "AI can't do my job", which is usually right. But AI can do 60 percent of the frictional tasks around you, and this 60 percent is where time is wasted most.
Establish a practice guide on workflow
Step 1: Identification of high friction tasks
Look at your work day and ask yourself:
- What missions do I do at least two times a week?
- Which mandates need to repeat the same steps?
- Which tasks bore me or irritate me?
These are ideal candidates for creating workflows.
Step 2: Analyse current process
Record the steps you are now taking to complete this task:
- What kind of input?
- What do you need?
- What are the predictable components?
- Where's the most vulnerable?
Step 3: Create structured hints
Based on the analysis, create a hint containing the following elements:
- Clear mission description
- Enter format requirements
- Output Structure Template
- Quality control standards.
- Key elements of success
Step 4: Test and optimization
Use this hint to perform several tasks and ask yourself:
- Is the result satisfactory?
- Any adjustments?
- How to make it more precise?
Step 5: Recording and standardization
Once the workflow stabilizes, records:
- Type of task
- Input requirements
- A reminder template
- Expected output format
- Common problems and solutions
Common error zones and avoidance methods
** Error 1: Seek perfection**: Work processes need not be perfect, just faster than manual. The quest for perfection will keep you from starting.
** Error 2: Too many created at once**: from 1-2 high-value workflows, validated and extended. Too much to chew.
** Error 3: Neglected quality control**: Add quality requirements (e. g. "tight" only) to the hint, verify results regularly, not allow AI output to become a new source of noise.
** Error area 4: no further updates after establishment**: regular review of workflows, continuous and iterative based on use of feedback. One-time workflows are less than systems that are continuously optimized.
** Error Area 5: Neglect of human and aircraft boundaries**: AI is good at implementation and collation, not judgement and decision-making. The workflow should leave "what needs to be judged" to humans and "repeat execution" to AI.
Future trends: what happens next
** Short-term (6-12 months)**
- ** Workflow Intelligence**: AI will better understand personal work styles and proactively recommend optimising space, not just passive noise Response
- ** Cross-platform connection**: workflows will be able to integrate directly with more tools (Notion, Linear, Slack) to reduce manual copy glue Paste
- Structural hint market: a certified high-quality tip template will be available to lower the threshold of "from zero"
** Mid-term (1 - 2)**
- ** Personalized workflow engine**: automatic generation and optimization of workflows based on your historical behavioral data, from "builder" to "auditer"
- More Agent collaborations: complex tasks will be performed by multiple AI Agent divisions, each Agent focusing on a workflow to form a streaming line
** More important trends**
AI Universalisation of workflows will redefine the meaning of "professional competence":
- ** Questions are more important than implementation**: The most scarce capabilities in the future are clearly defined issues, not skilled operational tools
- judgment becomes core competitiveness: AI can do it, but judging what to do, judging the outcome is still human.
- ** Systemic thinking becomes an essential skill**: People who can abstract repetitive tasks into processes, and only one person who can do it, the efficiency gap will widen.
Write at the end
Forty-five hours sound like a lot of hours, spread to the point of 1. 5 hours per month.
But this is 1. 5 hours that don't fall from the sky -- it's the result of shifting the energy that should have been spent on "operational" to "thinking."
AI won't make a decision for you, but I can do it for you. That's the correct position to use AI.
The rest is the selection of a workflow, which begins today.
Note: This paper is based on Zodchiii sharing on Twitter/X. More daily notes and AI practical ideas can be found on her Telegram channel (t. me/zodchixquant).