FOXNUT

Comparison

Should you use n8n or custom code?

Use n8n when the task is repetitive and deterministic. Write custom code when a step needs judgment. For one fixed task, a plain script can beat both.

By the Foxnut team · Updated

When should you use n8n and when custom code?

In n8n vs custom code, use n8n when the task is repetitive and deterministic, and write code when it is not. That is the whole verdict. It is a condition and not a number: no items-per-day threshold or budget line survives contact with real workflows, but the shape of the task does.

We are Foxnut Studios, a senior-only practice in Bengaluru and Paris, and custom AI systems are what we build for a living. There’s a link to talk to us at the end of this article. Let’s first find out which side of the line your task sits on.

For repetitive, deterministic work, meaning the same steps, in the same order, on inputs whose form you can predict, n8n or Zapier are perfectly good. We say so against our own interest, and we still routinely work with both tools ourselves. Connector tools fail when the problem or the workflow is non-deterministic, meaning a step needs judgment about what arrived and not plumbing to pass it along.

There is a second, slower condition that catches up with organisations that automate seriously. As an organisation becomes more AI-native, the cost of paying for connector tools overtakes the cost of building in-house. This page walks through the three legitimate options and the line between them, using the same judgment as the rest of this studio’s AI writing.

How do n8n, plain scripts and custom code compare?

They differ on the shape of task each one suits, and you are choosing between three things, not two: a connector tool, a plain script in software you already own, or a real system built in code. Most published comparisons leave out the middle option, which is a mistake because it wins a whole category of tasks outright.

AxisConnector tools (n8n, Zapier)Plain scripting (Sheets + GScript)Custom code, a real system
Right task shapeRepetitive, deterministic workflows across several toolsRepetitive, fully deterministic tasks living in one placeNon-deterministic problems that need judgment, reconciliation, or reasoning
What a step assumesEach step’s output is predictable enough to wire to the nextThe data is already structured and the logic is fixedOutputs vary, so the system needs its own tests and its own owner
Where it breaksThe first step that requires judgment about the contentScale, and logic that outgrows a scriptBeing pointed at plumbing a connector tool already does
Running cost shapeA subscription that grows with usage and with the number of workflowsFree inside tools already paid forBuilt once, owned in-house
Who runs itWhoever owns the tool accountWhoever owns the sheetThe trained team the system is handed to

The last row is not an afterthought. A real system changes hands with documentation, training and tests. How its output gets evaluated, and what happens when its success recruits more scope, are each deep enough subjects to have their own pages in this territory.

When should you choose each one?

Choose by the shape of the task, because each option wins a real category of work and fails in a characteristic way. The two worked cases below are our own examples, from the same operational territory, on purpose: the line between them is the line this page is about.

When does a connector tool win?

n8n or Zapier wins when the workflow is repetitive and deterministic and spans several tools: a form submission that becomes a CRM row, a payment that triggers an invoice, a file drop that fans out into notifications. The steps are fixed, the inputs are predictable, and the tool’s whole value is that nobody writes or maintains integration code. We reach for these tools ourselves when the task has that shape, because custom development there would be buying expertise the problem does not need.

Failure mode: the first non-deterministic step. That is a node that must judge what a document means, decide which of several unclear cases applies, or handle inputs whose form keeps changing. The canvas assumes predictable outputs wired to predictable inputs. Stretch it over judgment work and you get the familiar dead workflow, branch piled on branch, still wrong on the cases that matter.

There is also a slower failure, and it is financial, not technical. Connector pricing grows with usage, so an organisation that keeps getting more AI-native keeps paying more for plumbing it could by then own. Once that cost overtakes the cost of building in-house, the subscription has become the wrong purchase.

When does a plain script win?

The most under-recommended option in this comparison costs nothing at all. A well-maintained daily PnL sheet that needs a daily summary sent to leadership is Google Sheets plus GScript, for free, and nothing else is needed. There is no connector subscription, no build engagement and no new tool to learn, because the data is already structured, the logic is fully deterministic, and the script lives inside software you already run.

A surprising share of automation requests have exactly this shape, and the honest answer to them is an afternoon of scripting, not a purchase.

Failure mode: scope. Scripts are written by whoever needed them and maintained by nobody, so a script that quietly grows into something a workflow depends on becomes an unowned dependency. And the moment the task stops being deterministic, a script fails the same way a connector tool does, only with less visibility when it breaks.

When does custom code, a real system, win?

Code wins when the problem is non-deterministic: when the work is judgment, not plumbing. Reconciling invoices, purchase orders and receivables against cash-in-bank and working capital is our own example. The inputs arrive in inconsistent forms, the matching requires judgment, the exceptions are the actual work, and what you need is what our founder describes as a ten-times-more-powerful micro-CFO on call.

No connector canvas and no sheet script produces that, because the value is in reasoning over the data, not in moving it between tools. A system with that job has to be built, tested against the cases it must not get wrong, documented and handed to a trained team, which is why it is an engagement and not a subscription.

Failure mode: this one runs in the other direction. Building a real system for a task a connector tool already handles is paying deep expertise for shallow work. It is the first situation in the refusal list below, and we turn that work down instead of taking the fee.

Which is better, n8n or Zapier?

n8n or Zapier is the wrong question, and we do not rank vendors, because we build on these tools and a vendor ranking from a builder is marketing dressed as a review. What matters for your decision is that both are connector tools. They sit on the same side of every line on this page: right for repetitive, deterministic workflows, wrong for non-deterministic problems, and subject to the same subscription arithmetic as your automation grows.

Choosing between them is a configuration-level decision about integrations, hosting and pricing that an afternoon of reading settles. Choosing whether a connector tool is the right category is the decision that moves money.

Can you build AI workflow automation in n8n?

Yes. n8n can orchestrate AI workflow automations, and the published examples, such as summarise this, classify that, draft a reply, work exactly when the AI step behaves deterministically enough to wire into a canvas.

Use the test this page opened with on any such example. If every step’s output is predictable enough that the next step never needs to exercise judgment, the connector tool is the right home for it. The moment that stops being true, you are no longer configuring a workflow. You are operating an AI system, and it deserves to be built, tested and handed over like one.

We will tell you when your task is an afternoon in a connector tool, and we turn the build down when it is. When your task needs a real system, the people who sign the proposal do the work. There is one tier, we designed, made and launched our own products first, and we hand the system over running to a team we have trained.

The first step is a free expert audit: a senior practitioner reviews your setup, with no RFP and no pitch deck. You can read how Foxnut Consulting works first. If you are not sure which side of the line your own workflow sits on, tell us what it does and we will say honestly whether it is an n8n afternoon or a real build.

The part most pages leave out

When not to choose Foxnut Studios

Situations where another option is the better call, and where we say so in the first conversation rather than the fourth.

  • You want a superficial AI setup, something you could buy off the shelf or get running in an afternoon, and it does not need our expertise. An off-the-shelf tool, or a generalist implementer. If the problem does not need deep expertise, deep expertise is the wrong thing to pay for. We say so and point you at the simpler option.
  • Your senior leadership does not have conviction in making AI work for the organisation. Nobody, yet. Without conviction and commitment from the top, these projects tend to fail despite the best implementation. The honest move is to decline until leadership is committed, because the alternative is building something that will be abandoned.
  • Some of your organisation's data is offline or simply undocumented. Your organisation, doing the documentation work first. Data that is unstructured and scattered across many tools and databases is workable, and we work with that all the time. Data that exists only in someone's head or on paper is not, and no consultant can fix that from outside.

Foxnut Studios takes on work like this from Bengaluru and Paris. If you want to see how we do it before you talk to anyone, here is what an AI engagement covers and what you keep.