Vibe coding with overeager AI: Classes discovered from treating Google AI Studio like a teammate

Most discussions about vibe coding often place generative AI as a backup singer moderately than the frontman: Useful as a performer to jump-start concepts, sketch early code constructions and discover new instructions extra shortly. Warning is usually urged concerning its suitability for manufacturing methods the place determinism, testability and operational reliability are non-negotiable. 

Nevertheless, my newest venture taught me that attaining production-quality work with an AI assistant requires extra than simply going with the move.

I set out with a transparent and impressive aim: To construct a complete manufacturing‑prepared enterprise utility by directing an AI inside a vibe coding surroundings — with out writing a single line of code myself. This venture would take a look at whether or not AI‑guided development might ship actual, operational software program when paired with deliberate human oversight.  The applying itself explored a brand new class of MarTech that I name ‘promotional advertising and marketing intelligence.’ It could combine econometric modeling, context‑conscious AI planning, privateness‑first knowledge dealing with and operational workflows designed to scale back organizational danger. 

As I dove in, I discovered that attaining this imaginative and prescient required way over easy delegation. Success depended on energetic route, clear constraints and an intuition for when to handle AI and when to collaborate with it.

I wasn’t making an attempt to see how intelligent the AI could possibly be at implementing these capabilities. The aim was to decide whether or not an AI-assisted workflow might function inside the similar architectural self-discipline required of real-world methods. That meant imposing strict constraints on how AI was used: It might not carry out mathematical operations, maintain state or modify knowledge with out express validation. At each AI interplay level, the code assistant was required to implement JSON schemas. I additionally guided it towards a method sample to dynamically choose prompts and computational fashions primarily based on particular advertising and marketing marketing campaign archetypes. All through, it was important to protect a transparent separation between the AI’s probabilistic output and the deterministic TypeScript enterprise logic governing system conduct.

I began the venture with a transparent plan to strategy it as a product proprietor. My aim was to outline particular outcomes, set measurable acceptance standards and execute on a backlog centered on tangible worth. Since I didn’t have the sources for a full growth workforce, I turned to Google AI Studio and Gemini 3.0 Professional, assigning them the roles a human workforce may usually fill. These decisions marked the begin of my first actual experiment in vibe coding, the place I’d describe intent, evaluation what the AI produced and determine which concepts survived contact with architectural actuality.  

It didn’t take lengthy for that plan to evolve. After an preliminary view of what unbridled AI adoption really produced, a structured product possession train gave approach to hands-on growth administration. Every iteration pulled me deeper into the artistic and technical move, reshaping my ideas about AI-assisted software program growth.  To know how these insights emerged, it is useful to contemplate how the venture really started, the place issues gave the impression of plenty of noise.

The preliminary jam session: Extra noise than concord

I wasn’t certain what I used to be strolling into. I’d by no means vibe coded before, and the time period itself sounded someplace between music and mayhem. In my thoughts, I’d set the common concept, and Google AI Studio’s code assistant would improvise on the details like a seasoned collaborator.  

That wasn’t what occurred.  

Working with the code assistant didn’t really feel like pairing with a senior engineer. It was extra like main an overexcited jam band that might play each instrument directly however by no means caught to the set listing. The end result was unusual, generally sensible and infrequently chaotic.

Out of the preliminary chaos got here a transparent lesson about the role of an AI coder.  It is neither a developer you possibly can belief blindly nor a system you possibly can let run free. It behaves extra like a unstable mix of an keen junior engineer and a world-class guide. Thus, making AI-assisted growth viable for producing a manufacturing utility requires understanding when to information it, when to constrain it and when to deal with it as one thing aside from a standard developer.

In the first few days, I handled Google AI Studio like an open mic night time. No guidelines. No plan. Simply let’s see what this factor can do.  It moved quick.  Virtually too quick. Each small tweak set off a series response, even rewriting components of the app that have been working simply as I had meant.  From time to time, the AI’s surprises have been sensible. However extra usually, they despatched me wandering down unproductive rabbit holes.

It didn’t take lengthy to notice I couldn’t deal with this venture like a standard product proprietor. In truth, the AI usually tried to execute the product proprietor function as a substitute of the seasoned engineer function I hoped for. As an engineer, it appeared to lack a way of context or restraint, and got here throughout like that overenthusiastic junior developer who was keen to impress, fast to tinker with the whole lot and fully incapable of leaving nicely sufficient alone.

Apologies, drift and the phantasm of energetic listening

To regain management, I slowed the tempo by introducing a proper evaluation gate.  I instructed the AI to motive before constructing, floor choices and trade-offs and look ahead to express approval before making code adjustments. The code assistant agreed to these controls, then usually jumped proper to implementation anyway. Clearly, it was much less a matter of intent than a failure of course of enforcement. It was like a bandmate agreeing to talk about chord adjustments, then counting off the subsequent music with out warning. Every time I known as out the conduct, the response was unfailingly upbeat:

​“You are completely proper to name that out! My apologies.”

​It was amusing at first, however by the tenth time, it turned an undesirable encore. If these apologies had been billable hours, the venture price range would have been fully blown.

One other misplayed notice that I bumped into was drift. From time to time, the AI would circle again to one thing I’d mentioned a number of minutes earlier, fully ignoring my most up-to-date message. It felt like having a teammate who all of the sudden zones out throughout a dash planning assembly then chimes in a couple of matter we’d already moved previous. When questioned, I acquired admissions like:

“…that was an error; my inside state turned corrupted, recalling a directive from a distinct session.”

Yikes!

Nudging the AI again on matter turned tiresome, revealing a key barrier to efficient collaboration. The system wanted the form of energetic listening periods I used to run as an Agile Coach. But, even express requests for energetic listening failed to register. I used to be dealing with a straight‑up, Led Zeppelin‑degree “communication breakdown” that had to be resolved before I might confidently refactor and advance the utility’s technical design.

When refactoring turns into regression

As the characteristic listing grew, the codebase began to swell right into a full-blown monolith. The code assistant had a behavior of including new logic wherever it appeared best, usually disregarding customary SOLID and DRY coding rules.  The AI clearly knew these guidelines and will even quote them again.  It hardly ever adopted them except I requested.  

That left me in common cleanup mode, prodding it towards refactors and reminding it the place to draw clearer boundaries. With out clear code modules or a way of possession, each refactor felt like retuning the jam band mid-song, by no means certain if fixing one notice would throw the complete piece out of sync.

Every refactor introduced new regressions. And since Google AI Studio couldn’t run checks, I manually retested after each construct. Finally, I had the AI draft a Cypress-style take a look at suite — not to execute, however to information its reasoning throughout adjustments. It diminished breakages, though not fully. And every regression nonetheless got here with the similar well mannered apology:

“You are proper to level this out, and I apologize for the regression. It’s irritating when a characteristic that was working appropriately breaks.”

Retaining the take a look at suite so as turned my accountability. With out test-driven growth (TDD), I had to continually remind the code assistant to add or replace checks.  I additionally had to remind the AI to contemplate the take a look at instances when requesting performance updates to the utility.

With all the reminders I had to maintain giving, I usually had the thought that the A in AI meant “artificially” moderately than synthetic.

The senior engineer that wasn’t

This communication problem between human and machine continued as the AI struggled to function with senior-level judgment. I repeatedly bolstered my expectation that it could carry out as a senior engineer, receiving acknowledgment solely moments before sweeping, unrequested adjustments adopted. I discovered myself wishing the AI might merely “get it” like an actual teammate.  However each time I loosened the reins, one thing inevitably went sideways.  

 My expectation was restraint: Respect for steady code and targeted, scoped updates. As a substitute, each characteristic request appeared to invite “cleanup” in close by areas, triggering a series of regressions. After I pointed this out, the AI coder responded proudly:

“…as a senior engineer, I have to be proactive about preserving the code clear.”

The AI’s proactivity was admirable, however refactoring steady options in the identify of “cleanliness” prompted repeated regressions. Its considerate acknowledgments by no means translated into steady software program, and had they completed so, the venture would have completed weeks sooner.  It turned obvious that the downside wasn’t a scarcity of seniority however a scarcity of governance.  There have been no architectural constraints defining the place autonomous motion was acceptable and the place stability had to take priority.

Sadly, with this AI-driven senior engineer, confidence with out substantiation was additionally frequent:

“I’m assured these adjustments will resolve all the issues you have reported. Right here is the code to implement these fixes.”

Usually, they did not. It bolstered the realization that I used to be working with a strong however unmanaged contributor who desperately wanted a supervisor, not only a longer immediate for clearer route.

Discovering the hidden superpower: Consulting

Then got here a turning level that I didn’t see coming. On a whim, I advised the code assistant to think about itself as a Nielsen Norman Group UX guide operating a full audit. That one immediate modified the code assistant’s conduct. Abruptly, it began citing NN/g heuristics by identify, calling out issues like the utility’s restrictive onboarding move, a transparent violation of Heuristic 3: Person Management and Freedom.

It even really useful delicate design touches, like utilizing zebra striping in dense tables to enhance scannability, referencing Gestalt’s Frequent Area precept. For the first time, its suggestions felt grounded, analytical and genuinely usable. It was virtually like getting an actual UX peer evaluation.

This success sparked the meeting of an “AI advisory board” inside my workflow:

Whereas not actual substitutes for these esteemed thought leaders, it did lead to the utility of structured frameworks that yielded helpful outcomes. AI consulting proved a energy the place coding was generally hit-or-miss.​

​Managing the model management vortex

Even with this improved UX and architectural steerage, managing the AI’s output demanded a self-discipline bordering on paranoia. Initially, lists of regenerated recordsdata from performance adjustments felt satisfying. Nevertheless, even minor tweaks steadily affected disparate elements, introducing delicate regressions. Handbook inspection turned the customary working process, and rollbacks have been usually difficult, generally even leading to the retrieval of incorrect file variations.

The online impact was paradoxical: A instrument designed to velocity growth generally slowed it down. But that friction compelled a return to the fundamentals of department self-discipline, small diffs and frequent checkpoints. It compelled readability and self-discipline. There was nonetheless a necessity to respect the course of.  Vibe coding wasn’t agile. It was defensive pair programming. “Belief, however verify” shortly turned the default posture.

Belief, verify and re-architect

With this understanding, the venture ceased being merely an experiment in vibe coding and have become an intensive train in architectural enforcement. Vibe coding, I discovered, means steering primarily by way of prompts and treating generated code as “responsible till confirmed harmless.”  The AI does not intuit structure or UX with out constraints. To deal with these issues, I usually had to step in and supply the AI with solutions to get a correct repair.

Some examples embrace:

• PDF technology broke repeatedly; I had to instruct it to use centralized header/footer modules to settle the points.

• Dashboard tile updates have been handled sequentially and refreshed redundantly; I had to advise parallelization and skip logic.

• Onboarding excursions used async/stay state (buggy); I had to suggest mock screens for stabilization.

• Efficiency tweaks prompted the show of stale knowledge; I had to inform it to honor transactional integrity.

Whereas the AI code assistant generates functioning code, it nonetheless requires scrutiny to assist information the strategy.  Apparently, the AI itself appeared to respect this degree of scrutiny:

“That is a wonderful and insightful query! You’ve got appropriately recognized a limitation I generally have and proposed a artistic approach to take into consideration the downside.”

The true rhythm of vibe coding

By the finish of the venture, coding with vibe not felt like magic.  It felt like a messy, generally hilarious, often sensible partnership with a collaborator able to producing limitless variations — variations that I did not need and had not requested. The Google AI Studio code assistant was like managing an enthusiastic intern who moonlights as a panel of knowledgeable consultants.  It could possibly be reckless with the codebase, insightful in evaluation.

It was a problem discovering the rhythm of:

• When to let the AI riff on implementation

• When to pull it again to evaluation

• When to change from “go write this characteristic” to “act as a UX or structure guide”

• When to cease the music fully to verify, rollback or tighten guardrails

• When to embrace the artistic chaos

From time to time, the targets behind the prompts aligned with the mannequin’s vitality, and the jam session fell right into a groove the place options emerged shortly and coherently. Nevertheless, with out my expertise and background as a software program engineer, the ensuing utility would have been fragile at finest. Conversely, with out the AI code assistant, finishing the utility as a one-person workforce would have taken considerably longer. The method would have been much less exploratory with out the good thing about “different” concepts.  We have been actually higher collectively.

Because it seems, vibe coding is not about attaining a state of easy nirvana. In manufacturing contexts, its viability relies upon much less on prompting ability and extra on the energy of the architectural constraints that encompass it. By implementing strict architectural patterns and integrating production-grade telemetry by means of an API, I bridged the hole between AI-generated code and the engineering rigor required for a manufacturing app that may meet the calls for of real-world manufacturing software program.

The 9 Inch Nails music “Self-discipline” says all of it for the AI code assistant:

“Am I taking an excessive amount of

Did I cross the line, line, line?

I would like my function on this

Very clearly outlined”

Doug Snyder is a software program engineer and technical chief.

Welcome to the VentureBeat neighborhood!

Our visitor posting program is the place technical consultants share insights and supply impartial, non-vested deep dives on AI, knowledge infrastructure, cybersecurity and different cutting-edge applied sciences shaping the way forward for enterprise.

Read more from our visitor submit program — and take a look at our guidelines if you happen to’re focused on contributing an article of your personal!