Skip to content

AI news

Vibe Coding Restrictions: What Non-Developers Should Know

Goldman Sachs is restricting vibe coding. Here's what the report means for founders and freelancers who build apps with AI, and what to check first.

On this page
  1. What happened at Goldman Sachs
  2. Why developers are frustrated
  3. Why this matters if you build without code
  4. Data, security and legal basics to check first
  5. How to vibe code responsibly inside a company
  6. Build it with FromScratch
  7. FAQ
Laptop displaying code beside a padlock icon on a clean desk

Vibe coding restrictions are now part of the conversation in large companies, and that affects anyone who builds software by describing it to an AI tool. eFinancialCareers reports that Goldman Sachs is restricting vibe coding, and that developers are openly frustrated by it. This article explains what the report says, why the reaction matters if you build without code, and what to check before you make an app that touches real data.

What happened at Goldman Sachs

According to eFinancialCareers (efinancialcareers.com), Goldman Sachs is restricting vibe coding. The headline also reports that developers find the restriction annoying. The source summary is short, so we are not going to fill in details it doesn't give. For the scope of the restriction, who it covers, and any exact rules, read the original eFinancialCareers article.

Vibe coding is building software by describing what you want in plain language to an AI tool, then testing and refining the result by chatting rather than writing code by hand. The term is informal, and companies don't always use it the same way. Some firms treat any AI-generated code as a different category from code written by staff. Others only restrict specific tools or specific uses. That's why the wording of a policy matters more than the label on it.

The key point for readers is simple: a restriction at a major bank is a decision about how AI-written software may be used inside a regulated organization. It is not a verdict on whether AI app building works.

Why developers are frustrated

The headline says developers find the restriction annoying. We don't have quotes from the article in this summary, so we won't attribute specific complaints to anyone. But the general tension behind this kind of reaction is familiar across the industry.

Teams adopt AI tools because they want speed. A feature that once took a sprint can sometimes be prototyped in an afternoon. Then a policy arrives that slows that down, and the people doing the work see the friction first.

Firms, on the other hand, care about things that individual developers don't always weigh on a daily basis:

  • Code quality. Who checked the output, and does it match the team's standards?
  • Security. Does generated code introduce vulnerabilities, exposed keys or unsafe data handling?
  • Accountability. When something breaks or leaks, who answers for it?

Developers often feel that rules written with one tool or one risk in mind end up blocking work that was never risky. A small internal script that formats a report is not the same as a payment system. A good policy separates them. A blunt one doesn't, and that's where the annoyance tends to come from.

Why this matters if you build without code

If you are a founder, freelancer or operator building apps with an AI tool, you are often the person most exposed to rules like this. Professional developers usually work inside a process: pull requests, code review, staging environments, security scans. Many non-developers have none of that built in. The person who describes the app is also the person who ships it.

That makes the review step your responsibility. A few implications follow:

  • A restriction at a large bank is a signal. Reviewers, auditors, clients and regulators are likely to ask similar questions of any business app that handles customer or financial data: what data it uses, who can access it, where it lives, and who approved it.
  • Your company may have rules you haven't read. Check whether your employer, your client or your platform has an AI or software policy before you build anything that touches internal data. Ask before you start, not after a demo.
  • Internal does not mean exempt. A tool used only by your team can still expose personal data, contracts or pricing if it is built carelessly.

If you want a broader view of how these builders work and where the limits are, the FromScratch guide to AI app builders is a useful starting point. For the common ways non-developers end up building internal tools, see how to build an internal tool with AI.

Before you build anything, answer four questions in writing:

  1. What data goes in? List every field, file and source. Mark anything that is personal, financial, health-related, confidential or covered by a client contract.
  2. Who can see it? Name the people who will log in, and what each one should be able to view or change. "Everyone in the company" is an answer, but it should be a deliberate one.
  3. Where is it stored? Know which database and which hosting region are involved, and whether a third party processes the data.
  4. Who owns the output? Check your agreements with the AI tool, your employer and any client whose work the app will use.

Two practical rules follow from these questions. First, do not paste real customer records, login credentials, API keys or confidential documents into a prompt until you know the rules for that data. Use dummy data to design the app, then connect real data only after review. Second, read the legal checklist in our guide on vibe coding legal risks and how to prompt safely. It covers the questions that come up most often for non-developers.

How to vibe code responsibly inside a company

Good habits make it easier to get approval, and they reduce the chance that a restriction lands on you.

Start with a prototype and dummy data. Build the screens, the workflow and the rules using made-up records. Show the prototype to the person who owns the process. This catches the wrong requirements early, when changes are cheap.

Keep a simple build record. Note what you built, when, which tool you used, what prompts or specs drove it, and who reviewed it. This is not bureaucracy for its own sake. When someone asks how an app was made, you will have an answer.

Ask for a security or IT review before real users arrive. Bring the four data questions above. Ask specifically about access control, data storage and any third-party services. If your company has a formal review process, use it. If it doesn't, propose a short checklist and ask for sign-off on that.

Use version history. If a change causes a problem, you want to roll back to a known-good state rather than rebuild from memory. Version restore is one of the simplest safety tools available, and it is worth using from the first day.

Separate the test version from the live one. Publishing should be a deliberate step, taken after review and approval. Treat it that way even when the tool makes it one click.

Build it with FromScratch

The steps below show how a non-developer can build a review tool for AI-built apps, using FromScratch to test everything with dummy data before anything goes live. The app itself helps you run the checklist from the previous section.

  1. Open FromScratch and start a new project. Sign in on the free plan, or use your existing account.
  2. Describe the app in the chat. Paste the prompt below. The AI agent builds a React front-end, a Python (FastAPI) back-end and a database, then shows a live preview within minutes.
  3. Test with dummy data. Create two or three sample projects with invented answers. Check that the risk summary flags what you expect, such as personal data with no login requirement.
  4. Refine by chatting. Ask for changes in plain language, for example "add a field for data retention period" or "make the flagged risks appear at the top of the summary."
  5. Check the printable report. Open a project's report and confirm it includes every answer and the summary.
  6. Save and review versions. Every change is saved, so you can restore an earlier version if a refinement breaks something.
  7. Publish only after approval. When your reviewer signs off, publish to a public URL or your own custom domain. Until then, keep it private.

Use this prompt in the chat:

Build a simple web app called Vibe Check that helps a non-developer review a planned AI-built app before it goes live. Each project has a form with questions about the data it uses, who can log in, where data is stored, and who approves the work, and the app shows a summary with flagged risks. Include a login, a saved list of projects with their checklist answers, and a printable report for each project.

Once it's running, you can use the same review flow for your own projects. If you are comparing builders before you commit, our FromScratch vs Lovable vs Bolt comparison lays out the differences. For a related project type, see how to build an internal tool with AI.

FAQ

Is vibe coding banned at Goldman Sachs?

The eFinancialCareers headline reports that Goldman Sachs is restricting vibe coding. It does not give us enough detail to say whether the restriction is a full ban or limited to certain uses. Read the original article for the specifics before drawing conclusions.

Can I vibe code at work if my employer has restrictions?

Check your company's policy first, and ask IT or security before you build anything that uses real data. If the policy is unclear, ask in writing what is allowed. A short prototype with dummy data is usually the safest way to start a conversation.

Does using an AI app builder mean I can skip security?

No. Data handling, access control, storage and review still apply, whether a developer or an AI tool writes the code. The builder makes the work faster, but the responsibility for what goes live stays with you and your organization.

Where do I start if I am not a developer?

Read how to build a web app without coding, then prototype a small app using dummy data. Once the workflow works, bring the data and security questions to your reviewer before you connect anything real.

Written by

FromScratch TeamEditorial team

The team building FromScratch, the AI app builder that turns a plain-language description into a real, published full-stack app.