Workflow automation is software that connects your apps so an event in one (a form submission, a new CRM row) triggers follow-up actions in the others without anyone touching a keyboard. The 2026 twist is that many of those actions now think: they classify, summarize, draft, and decide.
Pick a tool based on who will build the workflows and who will have to fix them when they break.
- Marketer who hates code: Zapier. The biggest app catalog, a builder you can grok in an afternoon, and AI steps that bolt onto any Zap.
- Technical ops person: n8n. Self-host it free, pay per workflow run instead of per step, and get real agent nodes with approval gates.
- Budget-maximizer: Activepieces or Make. Both start near $9–20 a month with AI included.
- AI-heavy marketing work (scraping, enrichment, research): Gumloop.
- An AI teammate in Slack: Lindy.
- A Microsoft 365 shop: Power Automate Premium.
One warning: Relay.app, which 2025 roundups still recommend, shut down in September 2026. This category moves.
What AI actually added (and what’s still hype)
Three things actually shipped: AI steps, agents that decide, and document understanding. Everything else is mostly the word “AI” painted on a pricing page.
AI steps are model calls inside a normal workflow, classify an inbound lead, summarize a long thread, draft a reply to approve. Every tool here has them. What differs is how they’re billed, and that’s where budgets get hurt.
AI agent is the industry’s word for software that decides its own next steps instead of following a fixed branch. Zapier runs a separate Agents product billed in “activities” (400 a month free, 1,500 on Pro, per Zapier’s pricing page, October 2026). Make ships AI Agents in beta on all plans. n8n, Activepieces, Gumloop, and Lindy all have agent builders; Lindy goes furthest, behaving like a coworker with a Slack handle that pauses for approval on anything with outside impact.
For extracting data from documents, Make’s AI Content Extractor pulls structure out of files, Gumloop layers in scraping and enrichment connectors, and Lindy sits on your calendar and inbox producing notes and drafts.
What’s still hype: “fully autonomous campaigns.” Lindy’s own pricing page charges 250–1,000 credits for a single deep-work task and pauses your agent when credits run dry, vendors built those guardrails because agents burn money fast. If nobody reviews what an agent sends, you’ll regret it.
The tools, one by one
What each does well, what it costs with the math shown, and the limitation the marketing page won’t tell you.
Zapier, best for marketers who never want to see code
Zapier is still the safest first choice: the app catalog is unmatched, and Copilot drafts Zaps from a plain-English description. AI steps call models mid-workflow; Zapier Agents research leads across the web.
Pricing is per task: a task is any successful action step; triggers, filters, and paths are free. Zapier’s own docs state that a five-step Zap consumes four tasks per run, and AI steps carry multipliers, Standard 1x, Advanced 3x, Premium 5x, an MCP tool call 2 tasks.
Worked example: a lead-routing Zap with four action steps running 500 times a month is 2,000 tasks, the Pro 2,000-task tier at $49 a month billed annually ($73.50 monthly). Swap one action for an Advanced AI step and it burns 2,500, next tier up.
Multi-step Zaps multiply fast, overages bill at 1.25x your per-task rate annually (2.5x monthly), and the free plan caps at 100 tasks and two-step Zaps.
Make, best visual builder for the money
Make draws your workflow as a diagram, which makes branching, loops, and error handling easier to see than anywhere else. Its AI Agent app got a new version in February 2026, and agents attach to scenarios like any other module.
Pricing is per credit: one module action equals one credit by default, and routers and error handlers cost nothing. Free gets 1,000 credits and two active scenarios; Core is $9 a month billed annually for 10,000 credits, then Pro at $16 and Teams at $29, all with unlimited users.
Worked example: the same four-module lead router at 500 runs costs 2,000 credits, inside Core’s 10,000 for $9, roughly a fifth of the equivalent Zapier tier.
Credits turn fuzzy when AI enters. On Make’s own AI provider, billing is token-based, Make’s help docs list GPT-5 nano at about 18,080 input tokens per credit versus roughly 723 for full GPT-5. Same workflow, 25x difference.
n8n, best for technical ops people
n8n is the developer’s pick: open, extensible, and priced in a unit that makes sense. The Community Edition is free to self-host with no execution limits, and cloud plans bill per full workflow execution, node count and runtime don’t matter. An August 2025 pricing overhaul made workflows, steps, and users unlimited on every plan.
Worked example: 500 runs of your lead router is 500 executions whether the flow has four steps or forty. Self-hosted, that’s $0 in license fees. On cloud, Starter is €20 a month billed annually for 2,500 executions, Pro is €50 for 10,000, and the self-hosted Business plan (SSO, environments, Git version control) runs €667 a month for 40,000.
The AI story is strong: an AI Agent node, MCP support, human-approval steps, and bring-your-own keys.
You’re the sysadmin. Patching, backups, and security are your job, and the free edition lacks SSO and advanced access control. Self-hosting costs time before it saves money.
Gumloop, best for AI-heavy marketing work
Gumloop rebuilt itself in 2026: the drag-and-drop node canvas is retired, and you now describe what you want in chat while an agent assembles it. Its strengths are the marketer wish list (web scraping, contact enrichment, lead scoring, research) plus a “Company Brain” that syncs Slack, Gmail, Drive, Notion, and Salesforce.
MarketerMilk’s September 7, 2026 review lists these plans: Free gets 5,000 credits a month, Pro is $37 a month for 20,000 credits, and pay-as-you-go runs $1 per 200 credits.
Worked example: at $0.005 a credit, an enrichment flow costing 500 credits per run would eat Pro’s allowance in about 40 runs. AI-heavy work is never free.
Scraping and enrichment nodes use credits quickly. The plan carries an 8% orchestration fee, lacks live chat support, and pricing that has changed more than once this year.
Lindy, best if you want an AI teammate, not a builder
Lindy isn’t a classic integration platform; it’s an agent that lives where you work. It joins meetings and writes notes, drafts inbox replies “in your voice,” runs scheduled routines, and reaches 1,500+ integrations and MCP servers.
Pricing is per credit, and its page (October 2026) is refreshingly specific: everyday asks cost 2–250 credits, deep work like research reports runs 250–1,000, big builds 1,000–2,500. Team starts at $29.99 a month for 3,000 credits with top-ups at $10 per 1,000, and you get $50 in free credits for 7 days.
Worked example: a 3,000-credit plan covers roughly three to twelve deep-work reports a month, plus everyday asks. Credits only burn while Lindy works, and it pauses rather than overbills.
Monthly credits don’t roll over, deep work drains them fast, and there’s no visual builder to fall back on.
Activepieces, best budget pick, and it’s open source
Activepieces is the MIT-licensed dark horse. Self-host free, or take the cloud plans: Free forever with 1,000 credits a month for up to 5 users, Plus at $20 flat for 10,000 credits, Team at $200 for 50,000. Agents and MCP support are included even on Free.
Its billing unit is the friendliest I found: one run equals one credit regardless of step count. AI steps cost extra by model tier (2 credits fast, 10 smart, 20 frontier) but bring your own AI keys (Plus and up) and every AI step drops to 1 credit.
Worked example: twenty AI steps a day on the smart model costs about 6,200 credits a month at default rates, but about 620 with your own keys.
The app catalog is a fraction of Zapier’s, the polish is thinner, and the white-label Embed tier starts at $36,000 a year. For internal automation, the value is hard to beat.
Microsoft Power Automate, best inside Microsoft 365
If your company runs on Outlook, Teams, and Excel, Power Automate is already half-installed in your head. Cloud flows handle app-to-app logic, attended desktop flows cover RPA, and process mining ships on Premium.
Pricing (October 2026): Premium is $15 per user per month paid yearly. Unattended robots are where costs jump, the Process plan is $150 per bot per month, and Hosted Process with a Microsoft-managed VM is $215. A process-mining add-on lists at $5,000 per tenant per month.
Worked example: a ten-person ops team on Premium is $150 a month; add one overnight invoice-processing bot and you’re at $300.
Budget for user and bot licenses, pay-as-you-go flow runs, and a separate Copilot Studio bill. Read Microsoft’s licensing guide or you’ll guess wrong.
The same workflow, priced four ways
The billing unit matters more than the sticker price. Take one lead-routing workflow (form submission, AI scores the lead, CRM record created, Slack pinged) running 500 times a month.
Task-based pricing (Zapier) charges per action step: 500 runs x 4 actions = 2,000 tasks, landing at $49 a month on annual billing. Credit-per-module pricing (Make) counts the same four modules as 2,000 credits, covered by Core at $9. Execution pricing (n8n) sees 500 runs, period, free self-hosted. Run pricing (Activepieces) also sees 500 credits, which fits inside its free plan.
Same workflow, somewhere between $0 and $49 a month. Sketch your three busiest workflows and price them in each unit before you choose, the winner is rarely the one with the best landing page.
The tools at a glance
The table compares starting prices for paid plans.
| Tool | Best for | Starting price | AI capability | Main limitation |
|---|---|---|---|---|
| Zapier | Marketers who want zero code | $19.99/mo (annual, 750 tasks) | AI steps with model tiers; Agents product | Multi-step Zaps burn tasks |
| Make | Visual builders on a budget | $9/mo (annual, 10,000 credits) | AI Agents beta on all plans | Token-based AI billing varies by model |
| n8n | Technical ops and developers | €20/mo cloud (2,500 executions); free self-hosted | AI Agent node, MCP, BYOK, approvals | You own the maintenance |
| Gumloop | AI-heavy scraping and enrichment | $37/mo (20,000 credits) | Chat-built agents, Company Brain, scraping | Credit burn on enrichment |
| Lindy | An AI teammate in Slack | $29.99/mo (3,000 credits) | Meeting notes, inbox drafting, approvals | No rollover; deep work drains credits |
| Activepieces | Budget maximizers | $20/mo (10,000 credits); free self-hosted | Agents and MCP on free plan; BYOK AI | Smaller app catalog |
| Power Automate | Microsoft 365 organizations | $15/user/mo (Premium, annual) | Copilot Studio integration, RPA | Licensing complexity; bots cost extra |
The first three automations I’d build
Don’t start with a grand AI strategy. Start with these three, in order, each pays for the tool within a month and teaches you how the pricing works.
- Lead routing with AI scoring. When a form fills or a trial starts, an AI step classifies company size, intent, and fit, writes a one-line summary, and routes: hot leads get a Slack alert with a booking link, everyone else enters nurture. In Zapier this is one Zap with four to five actions; watch the AI multiplier if you use an Advanced model.
- Content repurposing handoff. When a webinar recording lands in Drive, transcribe it, then have AI draft a LinkedIn post, a newsletter blurb, and an article outline into a review doc. End the workflow at a human approval step (n8n and Lindy have these built in) because AI-drafted social copy ships fine 90% of the time, and the other 10% is why you review.
- A weekly reporting digest. Every Friday morning, pull CRM pipeline, ad spend, and traffic numbers, then have AI write a short “what changed and why it matters” summary to Slack. Zapier’s Digest and scheduling built-ins cost zero tasks, so this one is nearly free, and it’s the one your boss will actually notice.
FAQ
What are the best AI workflow automation tools in 2026?
Zapier for zero code, Make for visual control at lower cost, n8n if you’re technical, Activepieces if budget rules, Gumloop for AI-heavy scraping and enrichment, Lindy for an agent teammate, Power Automate inside Microsoft 365. Pick by who builds, then check the pricing unit against your volume.
Zapier vs Make vs n8n, which should I pick?
Zapier for speed and app coverage, Make for visual workflows at roughly a fifth of the cost per action, n8n for execution-based pricing and self-hosting. A pattern across 2026 comparisons: companies end up running two, Zapier or Make for marketing’s self-service builds, n8n where volume, cost, or compliance demands it.
Can AI agents automate marketing tasks?
Yes, for classification, summarizing, drafting, enrichment, and research, with a human approving anything that sends. Agents score and route leads, turn one webinar into a week of drafts, and write reporting summaries. Fully autonomous campaigns aren’t something I’d trust with your brand yet; the vendors’ own approval gates and credit guardrails say they agree.
What does workflow automation cost?
Most small teams spend $0–100 a month. Realistic 2026 anchors: Zapier Pro from $19.99, Make Core at $9, Activepieces Plus at $20, n8n cloud from €20 or free self-hosted. The traps are units and overages, the same 5,000-run workflow can cost $0 on n8n or hundreds on Zapier.
Is n8n self-hosting worth it?
If you’re comfortable with Docker, run high volumes, or have data-residency requirements, yes, the license is free and executions are unlimited. If nobody wants to own patching and backups, no: the maintenance time costs more than n8n’s €20 Starter plan.
Do AI steps cost extra?
Always. Zapier multiplies AI steps by model tier (up to 5x), Make bills tokens when you use its provider, and Activepieces charges 2–20 credits per AI step unless you bring your own keys. Budget AI usage separately from your workflow count.
Sources
- Zapier pricing page, plans, task tiers, Agents allowances
- What is a task in Zapier?, official task rules and AI multipliers
- Make pricing page, plans, credits, AI Agents
- Make help: Credits, fixed vs. dynamic usage, token rates
- n8n pricing page, cloud plans, self-hosted Business plan
- n8n blog: n8n’s new pricing, August 2025 pricing overhaul
- Lindy pricing page, credit costs by task type
- Activepieces pricing page, plans, AI step credit costs
- Microsoft Power Automate pricing, Premium, Process, Hosted Process
- MarketerMilk: Gumloop review, September 7, 2026
- Parseur: n8n vs Zapier vs Make, October 2026; the “companies run two tools” pattern
- Relay.app homepage, shutdown notice, September 2026
