Physix FrontierThree Months with WorkBuddy: Deleted All Low-Code Platform Decks
Honestly, my desk is half as cluttered as it was at the start of the year. On the left sits a dusty copy of Low-Code Development in Practice, and on the right is the Nebula PLUS trial account I bought six months ago. Now, the only thing I open is WorkBuddy's web terminal. I'm not trying to argue, but some things really do prove true the more you use them—flashy frameworks are less practical than something that just gets the job done.
The Detour from "Low-Code" to "AI Office"
Starting last Q4, I was responsible for maintaining an internal performance review process. The routine: collect weekly reports, financial data, and project progress from all groups by the 5th of each month, then manually integrate them into a PPT for the boss. Coming from a backend development background, my first instinct was, "Can this be automated?" So, following the ideas in industry news articles, I started researching low-code/AI development platforms.
The articles sounded great: "Nebula PLUS integrates an AI Development Agent, supporting AI-generated code + low-code refinement, covering the full lifecycle." I believed it. I applied for a trial, spent two weeks configuring the environment, building workflows, and writing logic. The result? For a simple workflow—"extract data from the past three weeks in Excel, summarize by department, generate a bar chart"—I wrote 5 visual nodes, 3 conditional judgments, and 1 API call, only to get stuck due to permission issues. Ops said, "The low-code platform needs whitelist access," and IT said, "We only open standard ports." After a week of hassle, I gave up.
Then a colleague sent me a link to WorkBuddy, saying, "Try this, just drag the file in." With a "what do I have to lose" attitude, I dragged a 500-row Excel file in and typed, "Calculate last week's completion rate by department, format as a table, sort by percentage descending." Three seconds. It gave me the result. I didn't even have time to check my phone again.
Low-code platforms solve the problem of "not enough developers," but often our issue is just "office workers don't want to use Excel formulas."
I later posted this sentence in the team wiki and got 12 likes—the whole group except me only has 11 people.
A Full Review of a Real Scenario
Let me give a typical example: the monthly business analysis report. Previously, it required 3 steps:
1. Export 4 Excel sheets from the OA system (revenue, costs, projects, HR)
2. Merge using VLOOKUP and pivot tables, manually adjusting formats
3. Switch to PPT, copy-paste, layout
It took about 4 hours. Sometimes data didn't match, requiring re-checks, taking even longer.
After using WorkBuddy, I changed the process to this:
| Step | Old Tool Time | WorkBuddy Time | Difference |
|---|---|---|---|
| Data Import | 15 mins | 2 mins | Direct drag-and-drop, no manual format conversion |
| Data Cleaning & Merging | 2 hours | 5 mins | Natural language description: "Merge 4 tables, match by project ID, keep all fields" |
| Chart Generation | 1 hour | 3 mins | Auto-generates bar + line charts, supports one-click export |
| Report Generation | 1 hour | 10 mins | Auto-layout into PPT, template selectable |
Total time dropped from 4 hours 15 minutes to 20 minutes, saving roughly 12x the time. And I didn't write a single line of code or configure a single environment variable.
Of course, there were pitfalls. The first time I used WorkBuddy to generate a PPT, it defaulted to a blue theme. My boss saw it and said, "Too plain." I later found out I needed to manually specify "Use 'Report Template v2.pptx' from the company template library." This setting was buried deep in "Output Settings" -> "Template" -> "Select Local File." It took me about 5 minutes to find. I suggest WorkBuddy make "remember frequently used templates" a default feature; otherwise, selecting it every time is annoying.
Why I Abandoned Low-Code Platforms After Comparison
I'm not disparaging tools like Nebula PLUS. They are indeed suitable for MES vendors and logistics system integrators, such as the "Zhongding Smart Warehouse Management Module, cutting custom development cycles by 50%" mentioned in the news. But the problem is, I deal with employee leave requests, expense approvals, weekly report summaries, and PPT beautification—not warehouse management.
Office scenarios and development scenarios are two different things. Characteristics of office scenarios:
- Diverse data sources (email, WeChat, local files, OA exports)
- Fixed output formats (PPT, Word, PDF)
- Simple logic (summing, comparing, filtering, sorting)
- Time-sensitive (needed this afternoon, can't wait two weeks for development)
Low-code platforms require you to first build a "process engine," then write logic, test, and deploy. WorkBuddy simply says, "Give me the file, tell me what you want." This "zero-config" experience might seem "unprofessional" to backend devs, but the results are good.
I even did an extreme test: threw in a 200-page PDF contract and asked, "Find all monetary figures in XX clauses, list them in a table, calculate the total." WorkBuddy took about 8 seconds, returning 23 figures totaling 127 million. I checked against the original text and found it missed one number—hidden in a footnote on page 157. So its recall rate isn't 100%, but for 90% of daily needs, it's completely sufficient.
Feature Suggestions for WorkBuddy
After using it for three months, I've accumulated some complaints, which I'll share here:
1. Multi-tab collaborative editing isn't smooth enough. I often have 3 documents open simultaneously (a spreadsheet, a doc, a PPT). To reference data from one document in another, I currently have to manually copy-paste. If it supported cross-document references like Notion (@ another document), it would be much more comfortable.
2. Local file support has limitations. It prioritizes parsing PDF, Excel, and Word, but occasionally loses formatting when parsing .md and .csv files. For instance, I had a .csv file encoded in GBK; it defaulted to reading it as UTF-8, resulting in garbled text. I had to manually convert the encoding to fix it. Suggestion: Add an "encoding detection" option for parsing.
3. The template library is too "official." Built-in PPT templates are all business-style. Our team prefers minimalist white backgrounds with bold fonts, but those aren't in the defaults. We have to upload our own, which is cumbersome.
4. Output speed is sometimes unstable. When processing PDFs over 100 pages, it occasionally hangs midway, prompting "Task Timeout," requiring resubmission. I guess it's a backend resource allocation issue, but as a user, the experience is poor.
That said, none of these flaws are fatal. The core value is saving time, and it delivers on that. I've even started using it to write code comments: throw in a code file and say, "Add Chinese comments to each function, generate readme.md." The comments are a bit verbose, but with minor edits, they're usable, saving me half an hour.
An Open Question
I've been thinking lately: where exactly is the boundary between AI office tools and low-code development tools? WorkBuddy can already handle many "development" tasks (data cleaning, logic judgment, report generation), but it's not a dev tool. Meanwhile, low-code platforms are adding AI capabilities (like Nebula PLUS's AI Agent), trying to penetrate office scenarios.
Will a tool emerge in the future that can handle temporary tasks with zero configuration like WorkBuddy, while also supporting persistent deployment of complex business logic like low-code platforms? Or conversely, will the two tools eventually merge into a "super productivity platform"? I don't know. But for now, I choose WorkBuddy. Because for someone like me who believes "if I can avoid writing one line of code, I won't write two," it lets me spend time thinking rather than configuring.
What do you think? Should AI in office scenarios be more like "a secretary who knows how to use AI," or more like "a process engine that writes code"?