n8n vs Make vs Custom Code: Picking the Right Automation Stack in 2026

Automation tooling in 2026 has gotten genuinely good. Three years ago you had to choose between a polished SaaS that nickel-and-dimed you on every API call, or a self-hosted tool that needed a developer just to deploy. Today the gap has narrowed in both directions. Make is more flexible. n8n is easier to host. And writing custom code is faster than ever thanks to AI assistants generating most of the boilerplate.

That said, the wrong choice still costs you. Pick Make for everything and your operations bill creeps past your CRM. Pick n8n on day one and you spend three weeks fighting Docker before you ship a single workflow. Pick custom code too early and you end up maintaining a brittle Node.js service that nobody on the team understands six months later.

Here is how to actually think about it.

1. Make: The Best Place to Start, With a Ceiling

Make (the platform formerly known as Integromat) is still the cleanest visual builder on the market. Drag a Gmail trigger, connect it to a Notion module, drop in a router, fan out to Slack and HubSpot. A non-developer founder can build a working lead-routing scenario in 45 minutes and have it running in production by lunch.

Where Make wins:

• Speed to first workflow. Nothing else gets you from idea to running automation faster.
• Pre-built modules. Over 1,800 integrations in 2026, including the AI providers (OpenAI, Anthropic, Mistral) with structured output handling baked in.
• Visual debugging. You can replay a failed scenario step by step, see the exact payload at each module, and fix it without touching code.
• Team handoff. Your operations lead can edit and own scenarios without bothering engineering.

Where Make hurts:

• Pricing scales with operations. Every module run is one operation. A scenario that processes 200 leads per day with 6 modules each is 36,000 ops a month for one workflow. Stack five of those and you are at the Pro or Teams tier fast. We have seen client bills hit $400 to $700 monthly within a year of adoption.
• Data leaves your infrastructure. Every payload, including PII, transits Make's EU or US servers. For Canadian businesses subject to PIPEDA or clients with strict vendor reviews, that becomes a problem.
• Limited control over retries and concurrency. When a workflow needs custom retry backoff, distributed locking, or exactly-once semantics, you are working against the platform.
• Version control is an afterthought. There is no real git workflow. Cloning a scenario for staging is manual.

Make is excellent at what it does. The trap is treating it as your permanent automation layer when it was designed to be your fast prototyping layer.

2. n8n: The Sweet Spot for Most Growing Teams

n8n is open source, self-hostable, and in 2026 it has matured into the default choice for any company that wants serious automation without serious SaaS bills. It runs on a $20 a month VPS and handles tens of thousands of executions a day without breaking a sweat.

The visual builder is now genuinely close to Make in polish. The 2026 releases added native AI agent nodes (LangChain-style chains, vector store integrations, tool calling), proper credential encryption, and a workflow execution history that rivals what you get from a paid platform.

Where n8n wins:

• Flat-rate cost. Self-hosted, your bill is whatever your VPS costs. A Hetzner CX22 at $5 a month runs more workflows than most SMBs will ever need.
• Data residency. Customer data, API keys, payloads, all of it stays on infrastructure you control. For regulated industries this is non-negotiable.
• Code nodes when you need them. Drop a JavaScript or Python node into any workflow when the visual modules cannot quite do what you want. No need to leave the platform.
• Git-friendly. Workflows export as JSON. You can version them in a repo, code-review changes, and promote between staging and production properly.
• Active ecosystem. The community node library has exploded. There is now an n8n integration for almost everything, and the AI node coverage caught up to Make in the 1.7 release.

Where n8n hurts:

• You own the ops. Backups, upgrades, monitoring, SSL renewal, scaling. If your team has zero infrastructure capacity, this is a real cost.
• Self-hosted means self-debugged. When a webhook stops firing at 2am, there is no support line to call.
• The cloud-hosted plan is fine but pricier than you expect. If you go cloud-hosted, the cost advantage over Make narrows considerably.

For companies running our AI automation engagements, n8n is usually the recommendation once a client has more than five active workflows or processes any kind of customer PII.

3. Custom Code: For the Two Workflows That Actually Matter

Custom code (a TypeScript service on Cloud Run, a Python worker on Fly.io, a serverless function on Azure) is overkill for most automation. But there is always a small set of workflows where it earns its keep.

The rule of thumb we use: if a single workflow handles more than $10,000 a month of business logic, it deserves custom code. Examples: the Stripe webhook that provisions paid accounts. The lead-scoring pipeline that decides which prospects your sales team calls first. The billing reconciliation job that catches missed invoices.

What custom code gives you that no-code cannot:

• Real testing. Unit tests, integration tests, replay against production fixtures. When a payment workflow breaks, you do not want to find out from a customer.
• Real observability. Structured logs in your existing stack (Datadog, Grafana, Azure Monitor), distributed tracing, alerting on business metrics rather than just "scenario failed".
• Edge case handling. Custom retry logic, idempotency keys, circuit breakers, dead-letter queues. The boring stuff that prevents 3am pages.
• Performance. A well-written Node service can process 10,000 events per minute on hardware that costs $15 a month. No-code platforms are not designed for that throughput.

What custom code costs:

• Build time. Even with AI assistants generating most of the code, a production-grade workflow takes a developer two to five days to ship properly.
• Ongoing maintenance. APIs deprecate, libraries break, certs expire. Plan for at least one engineering hour per month per workflow.
• Knowledge silos. If only one person on the team understands the code, you have a bus-factor problem.

4. The Decision Tree

Here is how to actually choose, in order:

• Have you ever automated this kind of workflow before? If no, start in Make. Validate it works and is worth the effort before investing in anything else.
• Are you running more than five production workflows? Move to n8n. The cost math flips and the version control story matters.
• Does this workflow touch customer PII or financial data? Self-hosted n8n at minimum. Ideally custom code with proper audit logging.
• Does this single workflow drive more than $10k a month in business? Custom code, with tests and monitoring. Stop trying to make it work in a visual builder.
• Is this a one-off departmental glue task? Leave it in Make forever. Not everything needs to graduate.

5. The Blended Stack Most Growing Companies Land On

After two years of doing this with clients, we almost never see a single-tool stack survive. The companies that get automation right run a blend:

• Make for departmental workflows owned by ops, marketing, or sales. Simple stuff, low stakes, fast to change.
• Self-hosted n8n for the core operational backbone: lead enrichment, AI agents that handle inbound queries, internal data syncs between SaaS tools.
• Custom services for the two or three revenue-critical paths. Payment provisioning, billing, anything where downtime costs real money.

This is not over-engineering. It is matching the tool to the actual stakes of the workflow. A free trial signup confirmation does not need the same infrastructure as a Stripe webhook that grants paid access.

6. What to Do This Week

If you have nothing automated yet: open a Make account, pick one painful manual task (lead routing, invoice creation, internal Slack notifications), and ship a working scenario before Friday. Stop reading articles like this and go build something.

If you are already on Make and your bill is creeping past $200 a month: spin up n8n on a small Hetzner or DigitalOcean droplet this weekend. Migrate your three highest-ops scenarios first. You will pay back the migration time in two months.

If you have workflows handling real money in a no-code tool: audit them. Identify the one with the highest dollar value flowing through it. That is your first custom code candidate. Build it properly with tests, observability, and a runbook. Sleep better.

The tools are not the bottleneck in 2026. Knowing which workflow belongs in which tool is the actual skill.