Skip to main content

This site is not ready to launch — leads are not being delivered

  • Critical: No contact email is configured, so the site has no working way to reach you. Set up the ImprovMX alias, then set email in src/config/site.ts.
  • Incomplete: Team facts (location, stacks) are unset. The About page is showing an explicit placeholder instead of real claims. Get the real details from Benson and fill in team.* in src/config/site.ts. Never invent these.

This banner is generated from src/config/site.ts and disappears by itself once every item is resolved. Run npm run check:launch to verify from the terminal. The full launch sequence is in the repository README.

Vibe Code Engineers

by Anysphere

Built it with Cursor?

An AI-native code editor that works on your own repository, in whatever language and framework you already use. It is a development tool rather than a hosted app builder.

What it generates

  • Whatever you are already using — no imposed stack
  • Your own repository, your own infrastructure
  • Agent modes that can plan and apply changes across many files
  • Project rules files that steer the model's conventions

What it’s good at

  • You own everything from the start. There is no export, no lock-in, and no platform boundary to migrate across.
  • Works on existing codebases of any size, not only new projects.
  • In experienced hands it is a genuine multiplier, and many professional engineering teams use it daily. Including ours.

What Cursor leaves you to build

Mostly scope rather than shortcoming — these are the boundaries of what the product sets out to do.

Review

This is the whole difference. Cursor will write whatever is asked for, and if nobody reads the result with a critical eye, unreviewed code accumulates at a rate no team previously had to handle. Analysis by Apiiro found AI-assisted developers producing three to four times more code, with architectural flaws up 153% and privilege-escalation paths up 322%, while syntax errors fell sharply. The shallow problems went away and the deep ones multiplied.

Architectural consistency

Across many sessions the model will solve the same problem several different ways, because it has no memory of the earlier decision. The result is a codebase with several competing conventions and no clear reason for any of them.

Someone who understands the whole system

A codebase where every part was generated and none of it was studied has what one widely-shared post called a bus factor of zero. Nobody can reason about the system as a whole, including the person who prompted it into existence.

Dependency scrutiny

Models sometimes reference packages that do not exist. Peer-reviewed work presented at USENIX Security in 2025 found 19.7% of generated code samples referenced a non-existent package, and that 43% of those fabricated names recur across runs — which is what makes registering them as malware a viable attack.

Signs you’ve outgrown it

  • You cannot explain how a significant part of your own codebase works.
  • The same concept is implemented several different ways in different files.
  • Pull requests are too large for anyone to review honestly, so nobody does.
  • You are avoiding certain files because changes to them have unpredictable effects.

What a handoff involves

Handoff is technically trivial — it is a normal repository. What takes time is building an accurate picture of a system nobody has ever read end to end.
Establishing conventions and writing them into project rules pays for itself immediately, because it makes the tool's future output consistent rather than novel each time.
The goal is not to use Cursor less. It is to have someone reviewing what it produces, which is the step that turns generation into engineering.

Cursor — questions

Is code from Cursor worse than code written by hand?

Not inherently, and the framing misleads. Reviewed AI-assisted code from a competent engineer is fine, and is what most professional teams now ship. Unreviewed code is the risk, and it would be a risk regardless of what produced it. The difference is that AI makes it possible to generate unreviewed code far faster than before.

We use Cursor as a real engineering team. Do we need help?

Often not, and you would know better than us. The teams that get value from an outside review are usually ones that scaled output faster than review capacity, which is a specific and recognizable situation rather than a general condition.

Last reviewed 2026-07-20. Cursor ships changes frequently. If something here is out of date, tell us and we will correct it — we would rather be accurate than flattering.

Not sure what your Cursor app needs?

A fixed-price review tells you what's actually wrong, what it would cost to fix, and whether you need us at all.

Request a Codebase Review