New: outcomized services. Pay per result, with a monthly cap you set. See pricing →
Engineering

Lovable, Cursor, Claude, and ChatGPT: How We Use Each

Updated 2026-10 · 10 min read · By the Founder of Outcomizer, former CTO and co-founder

We use four AI tools every day, and each one does a different job. Using the wrong one for a task is the most common way we see AI-assisted projects go sideways: a prototype tool asked to run production, or a chat window asked to edit a codebase it cannot see.

This is how a three-person studio that has built 50+ products splits work across Lovable, Cursor, Claude, and ChatGPT, how we move a clickable prototype into a real codebase, and the guardrails that keep the output reviewable.

Want this built, not just scoped?A live AI MVP on your domain in 15 days for $4,500. Then $1,500 a week sprints.

Lovable: day-one clickable prototypes

On the first day of a build, the most useful thing we can put in front of a client is something they can click. Lovable is our tool for that. In a working session we can go from "here is the idea" to a clickable set of screens, change the layout while the client watches, and try two or three versions of a flow before anyone argues about it in the abstract.

That changes the scope conversation. A client who has clicked through the onboarding flow says "we do not need this step" far more readily than one reading a spec. Several of our 15-day MVPs lost a whole screen on day one because of this.

Where it is the wrong tool: anything you plan to run in production for real users with real data, without an engineer reviewing what was generated. A prototype is built to answer questions. It is not built to handle permissions, migrations, background jobs, and the failure cases a production app has to survive.

Cursor: working inside the codebase

When we are editing code by hand, we work in Cursor. It is an editor first, which is the point. Tab completion, inline edits, and asking about a function while looking at it are faster than switching to another window for small, focused changes.

We use it most for the work an engineer wants their hands on: reading unfamiliar code, tightening a component, fixing a test, or reviewing a diff an agent produced with the file open next to it.

Where it is the wrong tool: long, multi-step tasks that touch many files and need a plan, a test loop, and a review. Doing those through editor prompts one at a time is slow. That is agent work.

Claude and Claude Code: agentic implementation and review

Most of our implementation runs through Claude, mainly Claude Code in the terminal. It reads the repository, writes a plan with checks, makes the change, runs the tests, and hands back a diff with proof. We are a member of Anthropic's Claude Partner Network, and this is the tool our studio is built around.

The difference for us is our skills. We run 30+ custom skills that encode how we work: the coding rules every agent loads, framework rules for Tauri and Shopify, a code review checklist, deploy steps, and SEO and content procedures. A new task does not start from a blank prompt. It starts from the same playbook a senior engineer on our team would follow.

We also use Claude for reviews. A second agent reviews the first agent's diff against the checklist before a human looks. The human review still happens on every change. It just starts from a list of flagged issues, not from zero.

Want this built, not just scoped?

A live AI MVP on your domain in 15 days for $4,500. Then $1,500 a week sprints.

See the 15-day AI MVP

ChatGPT: research, brainstorming, second opinions

ChatGPT is where we think out loud. Researching how a payment provider handles refunds, brainstorming names for a feature, comparing three ways to model a schema, or asking for a second opinion on an approach we have already picked. A different model looking at the same problem catches different things.

Where it is the wrong tool: editing a codebase it cannot see. Pasting files into a chat, copying code back, and hoping nothing drifted is how bugs arrive with no history. If the change belongs in the repository, it goes through an agent or an editor that works in the repository.

From Lovable prototype to production codebase

The handoff is where most AI-built products get into trouble, so we treat it as its own step. First we freeze the prototype and record a walkthrough of every screen and flow with the client's notes. That recording becomes the spec.

Then we decide, screen by screen, what to keep. Layouts and component styles often carry over. Data access, auth, and business logic almost never do. We set up the production repository with our stack, schema, auth, and tests first, then have Claude Code rebuild each flow inside it, using the prototype as the visual reference and the recording as the behavior reference.

The result looks like what the client approved and is built like something we are willing to run. If you already shipped a prototype to real users and it is breaking, that is a different job, and it usually starts with a diagnostic, not new features.

Guardrails that keep the output reviewable

Rules files. Every repository has an instructions file the agent loads on every session: our four coding rules (think first, simplicity, surgical changes, verified goals), the stack, the commands to run tests, and the things the agent must never do, like touching migrations without asking.

A review checklist. Every diff gets checked for unrelated changes, missing tests, secrets, permission changes, and new dependencies. The checklist is a skill, so the reviewing agent and the human use the same list.

Tests as the definition of done. A feature is not finished when the agent says it is. It is finished when a test that failed before now passes and a human has read the diff. Without that, speed from AI tools turns into rework a week later.

Key takeaways

  • Lovable answers product questions on day one. It is not where production code comes from.
  • Cursor is for hands-on edits inside the codebase. Long multi-file tasks belong to an agent.
  • Claude Code does most of our implementation and first-pass review, driven by 30+ custom skills.
  • ChatGPT is for research and second opinions, not for editing code it cannot see.
  • Treat the prototype-to-production handoff as its own step, guarded by rules files, a review checklist, and tests.

Frequently asked questions

No. They are tools we pay for and use daily. Our only formal program membership is Anthropic's Claude Partner Network.

Yes. We review what exists, keep what holds up (often the screens and styles), and rebuild data, auth, and logic in a production codebase. If the app is already live and failing, we start with a diagnostic.

No. Every change gets a human review and passing tests before it merges. Agents write and check. People approve.

No. You see clickable prototypes, recorded walkthroughs, and the live product. The tooling is our job.

Free playbook

The 15-Day AI MVP Playbook

Day-by-day plan, a starter CLAUDE.md, the Lovable-to-production checklist, eval and cost worksheets, and 10 prompts. Pure value, no pitch.

Get the playbook →
Next step

Book 20 minutes, or send a brief

A former CTO replies. You leave with a written plan and a fixed estimate.

No spam. One reply within one business day. Or skip the form and book a call.

Thank you. Your brief is on its way. Grab a time now so we can talk it through.

Prefer LinkedIn? Connect with Harish.

HG
Harish Ganapathi, Founder of Outcomizer, former CTO and co-founder

I started Outcomizer because hourly billing rewards the wrong thing. As a CTO I led an 18-engineer team and shipped the products in these case studies, including a Fortune 100 shipping system and a SaaS we later sold. Now I put my fee on the result. I still do the work. You work with me, not a sales layer on top of a bench.

Connect on LinkedIn →
Book a callFree lead test