Can a newbie really vibe code an app? I tried Cursor and Replit to find out

2 days ago 12

Follow ZDNET: Add america arsenic a preferred source connected Google.

ZDNET's cardinal takeaways

Tools specified arsenic Cursor tin spell a agelong mode toward simplifying codification setup.
There's inactive a batch of enactment to refine the results.
Conceiving an app's goals and however to get determination is the hidden gotcha of AI coding.

The biggest occurrence truthful acold of generative artificial quality successful the endeavor is AI coding tools that assistance programmers. Startups specified arsenic Cursor, Replit, Lovable Labs, Harness, Windsurf, Augment Code, All Hands AI, and Microsoft, with its Visual Studio with GitHub Co-pilot, each connection programs that tin drastically trim the hand-coding humans request to do.

And truthful I wondered: Could a newbie similar me, with constricted programming knowledge, speech my mode done creating an app?

Also: OpenAI's caller Spark exemplary codes 15x faster than GPT-5.3-Codex - but there's a catch

Nvidia CEO Jensen Huang has said that vibe coding English is going to beryllium the caller programming language. Armed with that blessing, I acceptable retired to try.

I had a blast watching however Cursor and the others tin automatically execute tasks specified arsenic mounting up the "virtual environment," uncovering and installing codification libraries, installing and moving a section server, and making changeless tweaks. The information that an app tin bash each that with a elemental natural-language punctual from the idiosyncratic is simply a wonderment to behold.

After respective days of experimentation with aggregate tools, I managed to get a small, elemental information investigation programme running. It was a promising start, but it was besides wide I couldn't get precise acold from determination without immoderate existent programming skills.

Also: Why AI coding tools similar Cursor and Replit are doomed - and what comes next

Too galore times, PC-based tools specified arsenic Cursor and Microsoft Visual Studio bounced maine into a terminal model and asked maine to tally assorted command-line tasks. Although I cognize my mode astir a bid line, it was a tedious other measurement -- aggregate steps, successful fact. I'm definite a non-programmer with little patience than I would person fixed up.

While little taxing, unreality offerings specified arsenic Replit and Lovable rapidly had maine burning done credits arsenic I tried to fig retired what changes my app needed to determination forward. And, putting my backstage information successful the unreality was thing that gave maine anxiety.

My archetypal program

Most radical who usage AI usage a chatbot to person conversations, analyse a fistful of documents, oregon possibly gully a picture.

All the coding tools, Cursor and the rest, run connected the aforesaid principle: You benignant into a prompt, and the programme takes a task statement successful adjacent vague terms, formulates a program of attack, sets up the requisite codification files, and automatically generates a idiosyncratic interface for your caller program, usually 1 moving successful a web browser.

Also: Worried astir AI coding? Why the invention of powerfulness tools is the blueprint for your vocation future

My intended vibe coding project: a database investigation instrumentality that tin ingest word-processing documents and hunt them for patterns. My imaginativeness was to adhd a conception to my newsletter's website wherever a scholar could execute not conscionable a keyword hunt of my 1,731 articles, but besides much analyzable free-form queries, specified as, "How has the presumption of Nvidia arsenic a institution and an concern changed implicit the past 5 years?" My articles, which person hundreds of mentions of Nvidia, could output a benignant of enforcement summary that isn't disposable conscionable by asking the aforesaid question successful ChatGPT oregon Perplexity, due to the fact that they don't person my peculiar articles.

First steps

I started with Cursor due to the fact that it's the marque I've heard of astir among the tools, truthful I figured I should commencement with the sanction connected everyone's lips. I downloaded and installed the MacOS binary arsenic easy arsenic immoderate different Mac download.

Cursor offers a 7-day escaped proceedings for the Pro version, but I decided to skip it and commencement with the escaped plan.

To a non-coder, the archetypal surface successful Cursor aft sign-up tin beryllium daunting. It asked if I wanted to "open project," "code repo," oregon "Connect via SSH."

Although I knew thing astir however this programme worked, I knew it utilized a chat paradigm similar ChatGPT. I figured retired however to get to the "agent" broadside sheet -- a shortcut: pressing command+I connected the keyboard -- and I described my intended project.

Also: This 'critical' Cursor information flaw could exposure your codification to malware - however to hole it

All these programs, some those moving locally connected Mac oregon PC and those moving arsenic a unreality app successful a web browser, person the aforesaid 3 indispensable places wherever you benignant oregon observe results: The cause panel, wherever you punctual the model, acquainted to america all; the terminal display, wherever you spot a accepted bid line, daunting to each but the initiated; and a preview of your app's graphical idiosyncratic interface, GUI, the extremity effect of your work.

It's each astir going backmost and distant betwixt the 3 arsenic you marque requests, tweak things, spot results, and repeat.

Initial exhilaration with Cursor

The archetypal respective minutes with Cursor were electrifying. Cursor described each the requirements of the project, which I wouldn't person known; it past offered to acceptable up the record operation for the task connected my hard drive, link to the database, instal indispensable Python files, etc. I asked if I could usage Mac's .Pages connection processing files, and was relieved to person Cursor archer maine it would instal the requisite "parser" from an outer codification library.

I was thrilled.

And, then: disappointment.

Cursor told maine it would instal immoderate Python libraries needed for the project. Then the programme restarted itself, and my chat past from the erstwhile hr was obscurity to beryllium found. There was nary chat past whatsoever; it was blank. That meant the plans we discussed for what we were going to physique were gone.

Also: 10 things I privation I knew earlier trusting Claude Code to physique my iPhone app

I spent an hr connected the substance and learned that other radical person had their chat histories disappear. The assistance forum, and a caller chat successful the cause model astir the problem, made it wide that this is conscionable 1 of the things that tin happen. The chat past for a task is stored successful a "workspace" folder, wrong the ~Library folder of the existent idiosyncratic relationship connected the Mac hard drive. But, there's nary mode to unit Cursor to retrieve immoderate of that chat past if it's not doing it automatically.

That's a non-starter for me. I'm not going to enactment with a programme that can't reliably support chat history, fixed that chat past is 1 of the cardinal innovations of vibe coding. I gave up connected Cursor and moved connected to Replit.

I deed the paywall with Replit fast

Replit is each online, successful the cloud, nary download. I wasn't arsenic blessed with that due to the fact that I'd similar to usage my ain information offline, connected my Mac, but I decided, for the involvement of the experiment, to instrumentality the hazard of exposing my data.

I described the task goal, and Replit got to enactment mounting up the environment, with a bare-bones app model acceptable successful preview mode wrong astir 15 minutes. I requested the quality to usage .Pages files arsenic data, and it adjusted the situation variables.

Also: Vibe coding feels magical, but it tin descend your concern accelerated - here's how

All these programs hosted successful the unreality person a acceptable magnitude of capableness earlier you person to halt and hold to enactment again. I rapidly ran done the allotment connected Replit earlier I had a accidental to bash thing with my recently created app.

With the archetypal large query I tried, the app failed due to the fact that it said nary information had been uploaded, adjacent though each my information sources were selected successful the broadside panel. Just arsenic I described the occupation of missing files successful Replit, I deed the escaped plan's quota and was told I would person to hold 24 hours to resume enactment unless I upgraded to a paid plan.

Visual Studio wasn't precisely my savior

While I contemplated whether to commercialized up to the paid program with Replit, I knew I inactive wanted to enactment connected thing connected the section instrumentality alternatively than successful the cloud. So, I went adjacent to Microsoft's Visual Studio, which has the built-in GitHub Copilot relation that provides chat-based programming conscionable similar Cursor and Replit. Visual Studio has a dedicated Mac IDE conscionable similar Cursor.

Although Visual Studio tin petition support to tally terminal commands, instead, astir of the clip it provided codification snippets that I had to transcript and paste into the bid line.

Also: After coding catastrophe, Replit says its caller AI cause checks its ain enactment - here's however to effort it

Finally, aft a mates of hours of this terminal business, each method issues were resolved, and a prototype investigation app was up and moving successful the web browser connected the section server. This was the farthest I'd gotten, truthful I was ecstatic!

Then came the truly hard part. What Visual Studio had produced was a benignant of elemental text-string-matching relation that would simply spot if a fixed drawstring of substance could beryllium recovered successful a file. That's not analysis, it's conscionable keyword-matching. I tried prompting Visual Studio Copilot to refine the analysis, but hours of trial-and-error successful Visual Studio yielded nary improvements.

It became wide astatine this constituent that gathering an investigation instrumentality that tin really execute textual investigation is simply a analyzable matter. It means not simply being capable to code, oregon to person Copilot code; it means knowing what the investigation relation consists of, which would impact choices made.

Lovable's accelerated start

I inactive wasn't acceptable to wage Replit money, truthful I moved connected to Lovable, which offers a escaped cloud-based setup that seems a batch similar galore of the others.

I was delighted to spot however rapidly Lovable created a web front-end to upload files and execute a textual analysis. Also, Lovable has the simplest idiosyncratic interface, hiding a batch of the codification details. Out of the 4 with which I worked, Lovable is the astir similar utilizing ChatGPT, Gemini, oregon Perplexity

Also: How to make your archetypal iPhone app with AI - nary coding acquisition needed

However, I rapidly reached the aforesaid decision arsenic with Visual Studio: To get meaningful analysis, much enactment would beryllium needed. I told Lovable I wanted investigation of "themes, ideas, and perspectives" successful the articles uploaded, and it suggested I drawback Google Gemini and usage that, to which I agreed.

That was an improvement. Gemini provided immoderate high-level observations that I recovered relevant. However, due to the fact that of the unsocial parsing requirements of Apple .Pages files, Gemini could lone analyse papers titles and metadata, not the bulk of the article.

Before I could instrumentality different step, I had maxed retired the escaped plan's credits. I liked what I had seen truthful far, and I was truthful exhausted by nonaccomplishment that I was determined not to permission without immoderate benignant of achievement.

So, I enactment successful my recognition paper for Lovable's Pro plan, $25 per period for 100 credits.

Also: I looked into the astir fashionable AI apps, and the runner-up caught maine by surprise

I proceeded to marque immoderate enhancements. Instead of relying connected .Pages files, I connected Lovable to an archive of articles that I support successful XML format connected the Algolia service. That process was comparatively simple. The app was capable to retrieve each the substance from the Algolia server, nonstop it to Gemini, and Gemini performed rudimentary semantic investigation successful effect to queries.

I yet achieved a basal substance investigation application.

Before wrapping up, I returned to Replit, a time aft the intermission successful my escaped credits. The intermission had ended and I was capable to resume moving from wherever I had near off. I proceeded to effort solving the contented of files not being read. It seems Replit had made immoderate benignant of mistake successful extracting the contents of my .Pages files. I tried again uploading documents, Replit tried to hole the situation, but this dragged connected for different fractional hour. I gave up connected Replit, precise gladsome that I did not springiness them immoderate of my money.

The large picture

The AI tools helped maine bash a batch of things I couldn't person done without them, that's for sure. At the aforesaid time, they made maine alert of conscionable however overmuch much effort I would person to enactment into the task to maestro each the eccentricities of the programs and the quirks of installing and maintaining code.

Also: Worried AI volition instrumentality your distant job? You're harmless for now, this survey shows

I was caught betwixt 2 worlds: the satellite of desktop software, specified arsenic Cursor and Visual Studio, and the satellite of the cloud-hosted platform. Neither was precisely close for me. I had excessively galore headaches with the desktop software. But the cloud-native versions kept the metre moving the full time, racking up costs, and I had to exposure my information by putting it connected a distant server.

Inside a large endeavor with a large IT team, idiosyncratic other would person handled the details of mounting up the virtual environment, managing the data, etc., for me. But, arsenic a sole proprietor, wherever I'm the IT team, there's conscionable a batch of other enactment for me.

I volition accidental that utilizing these tools gave maine a greater appreciation for programmers. I've sat and watched radical programme and been amazed by however they marque intuitive leaps to find problems successful codification and improvise solutions. Trying and failing to bash anything, adjacent with a batch of automation help, made it adjacent clearer however overmuch is progressive conscionable to negociate the instrumentality itself and leap into enactment erstwhile the automation doesn't work.

Also: What is AI vibe coding? It's each the rage but it's not for everyone - here's why

I don't deliberation the occupation of the quality coder is going distant anytime soon.

I besides gained a greater appreciation for merchandise improvement teams. When I yet reached thing usable with Lovable, it was wide to maine conscionable however overmuch much enactment would beryllium progressive successful designing the eventual app, and figuring retired however to get there. You tin automate code, but that conscionable leaves much decisions astir what you are really trying to build, and however you volition cognize erstwhile you've got a minimum viable product. I deliberation that means merchandise improvement teams are not going away, either.

Read Entire Article