business

Best AI Automation Tools

The best AI automation tools in 2026, compared on list price, what each platform counts as a billable unit, how AI is metered on top, and which one fits which team.

Updated September 2026

Quick answer: The best AI automation tool for most teams is Make, because a single paid plan starts at $9 a month for 5,000 credits and rises with the credit tier you pick, and its visual builder handles branching logic that simpler tools cannot. For the widest app coverage, Zapier connects more than 9,000 apps and starts at $19.99 per month on annual billing, but it bills a task for every action step, which makes long workflows expensive. For technical teams and anyone self-hosting, n8n is the strongest option: its cloud Starter plan is €20 per month on annual billing for 2,500 executions, and one execution covers an entire workflow run no matter how many steps it contains. For Microsoft 365 organisations, Power Automate Premium at $15 per user per month is the cheapest route to governed automation inside an existing tenant, but Microsoft is removing the seeded AI Builder credits from that licence on 1 November 2026, so the AI portion becomes a separate budget line. For large enterprises, Workato is a Gartner Magic Quadrant Leader for the eighth time and is quote-only. For AI-heavy document and research workflows, Gumloop is the strongest of the newer AI-native builders, at $37 a month for 20,000 credits. The caveat that frames everything below: the sticker price is rarely the cost. Every platform on this page meters differently, per action, per workflow run, per credit, per CPU second, and AI features almost always draw on a second meter charged on top of the first.

This page ranks deterministic workflow automation tools: platforms where you define the steps and the platform executes them the same way every time, with AI called as one step among many. That is a different product category from autonomous AI agents, which decide their own steps. If you want a system that works out what to do on its own, see best AI agents instead; if you are building agents in code, see best AI agent frameworks. The dividing line matters commercially, because the two categories are metered and governed in completely different ways, and it is covered in full in the segmentation section below.

AI automation tools versus AI agents: the distinction that decides your bill

An AI automation tool runs a workflow you designed. The trigger, the steps and the order are fixed; you can drop an AI step into the middle of it to classify an email or summarise a document, but the path through the workflow is deterministic and auditable. Zapier, Make, n8n, Power Automate and Workato are all in this class.

An AI agent is given a goal and a set of tools, and chooses its own steps at run time. The path is different every time, which is what makes agents useful for open-ended work and what makes them hard to govern and hard to cost. ChatGPT Agent, Claude Code, Manus and Devin are in that class.

Three practical consequences follow from the split:

• Billing. Deterministic platforms bill a predictable unit per run. Agent products bill an unpredictable one. Zapier Agents runs on a separate meter from Zapier Zaps entirely, activities rather than tasks, precisely because agent consumption cannot be forecast the way a fixed workflow can.
• Failure mode. A broken workflow fails the same way every time and is fixable. A misbehaving agent fails differently every time. Gartner expects more than 40 per cent of agentic AI projects to be cancelled by the end of 2027.
• Audit. Deterministic workflows produce a repeatable execution log that a compliance team can read. Agent traces do not repeat.

Every major platform on this page now sells an agent product alongside its workflow product: Zapier Agents, Make AI Agents, the n8n AI Agent node, Microsoft Copilot Studio, Workato Genies. This page prices the workflow product and notes the agent surcharge. The agent products themselves are ranked on best AI agents, and the concepts are explained in what is agentic AI.

The current state of AI automation tools: September 2026

Four things define the category as of 20 September 2026.

The billing unit is now the main differentiator, not the feature list. Every serious platform connects to the apps most teams use, offers a visual builder and can call a large language model. What separates them is what they count. Zapier counts a task for each action step that runs successfully, so a five-step Zap firing 100 times consumes 500 tasks. n8n counts one execution for an entire workflow run, whatever its length. Make counts a credit per module action, with AI and code modules costing more. Activepieces counts one credit per flow run and two to twenty credits per AI step depending on the model. Latenode has left per-run billing altogether and charges for CPU seconds. Those models produce materially different bills for the same automation, and the difference grows with workflow complexity.

AI is metered separately almost everywhere. Make renamed its billing unit from operations to credits on 27 August 2025, converting one for one, specifically so that AI and code steps could consume more than one unit: its own documentation says AI modules consume credits according to tokens, file size or run time rather than a flat rate, and running code costs two credits per second of execution. Microsoft is removing the seeded AI Builder credits from Power Automate Premium and Process licences on 1 November 2026, after which AI Builder features draw on Copilot Credits. Activepieces charges two credits for a fast model, ten for a mid-tier one and twenty for a frontier model, dropping to one when you bring your own AI keys. Zapier Agents bills activities on a cap with no published overage rate. The pattern is consistent: the plan price buys the plumbing, and the AI is a second meter.

The Model Context Protocol has become the interoperability layer. Zapier’s MCP server exposes more than 30,000 actions across its 9,000-plus apps to any MCP-compatible model, and n8n ships MCP nodes in both directions, so an existing workflow can be published as a tool an AI host can call. This matters for buyers because it decouples the automation platform from the AI interface: the platform becomes the execution layer, and the model becomes the caller.

The mid-market is consolidating, and one platform has already closed. Relay.app shut down in September 2026, and its own notice states that all accounts, workflows and run history have been permanently deleted. TechCrunch reported that the wind-down was announced on 16 July 2026, that free accounts closed on 15 August and paid access ended on 14 September, and that founder Jacob Bank and several staff joined Google’s Chrome team. On the acquisition side, Workday agreed to acquire Pipedream on 19 November 2025 and closed the deal in February 2026, and Make has been owned by process-mining company Celonis since it acquired Integromat in 2020 and rebranded it in 2022. Activepieces replaced its per-flow pricing with a credit model in August 2026, which is a smaller change than an acquisition and a larger one than it sounds for anyone who budgeted on the old meter. Anyone choosing a platform in 2026 should weigh export options and vendor durability alongside price.

What the evidence actually shows

Adoption is broad; proven return is not

MIT’s Project NANDA study The GenAI Divide: State of AI in Business, published in July 2025, found that 95 per cent of enterprise generative AI pilots produced no measurable profit-and-loss impact, based on 52 executive interviews, surveys of 153 leaders and analysis of 300 public deployments (reported by Fortune). The study was not peer reviewed and its interview base is small, and critics have pointed out that “no measurable impact” partly reflects pilots run without a documented pre-deployment baseline rather than pilots that failed technically. The finding that matters most for anyone choosing an automation platform is a different one: the same study put the largest measurable returns in back-office automation, while more than half of generative AI budgets were going to sales and marketing tools.

Spending has not slowed in response. Gartner has raised its 2026 AI spending forecast three times this year, from $2.52 trillion in January to $2.59 trillion in May and then to $2.7 trillion on 16 September 2026, a 49.5 per cent year-on-year increase. The gap between what is being spent and what has been proven is the defining tension in this market.

The analyst view is concentrated at the enterprise end

The 2026 Gartner Magic Quadrant for Integration Platform as a Service, published on 16 March 2026, evaluated 18 vendors: Amazon Web Services, Boomi, Celigo, Frends, Google, Huawei Cloud, IBM, Jitterbit, Microsoft, Oracle, Salesforce (Informatica), Salesforce (MuleSoft), SAP, SEEBURGER, SnapLogic, Tray.ai, Workato and Zapier. Boomi was named a Leader for the twelfth consecutive time and placed highest for Ability to Execute; Workato was named a Leader for the eighth time and furthest in Vision for the third consecutive year; SAP was a Leader for the sixth time; Salesforce MuleSoft was also a Leader. Celigo, SnapLogic and Tray.ai were placed as Visionaries.

Two things follow for buyers. Zapier’s inclusion confirms that the self-serve and enterprise ends of this market are now judged in the same category, which was not true two years ago. And Make and n8n are absent, an absence that is a scope judgement rather than a quality one, since Gartner assesses enterprise integration platforms and both are sold primarily as self-serve builders. Buying an integration platform when you needed a $9-a-month scenario builder is a more common and more expensive mistake than the reverse.

Vendor-reported adoption figures should be read with care

n8n states that more than 80 per cent of workflows built on its platform now involve AI agents. That figure comes from the vendor and has not been independently verified. Reported customer counts for n8n also conflict across sources: Sacra puts it at 1,400-plus enterprise customers and 1.7 million monthly active builders, while n8n’s own Series C announcement and the trade coverage of it cite 3,000-plus enterprise customers. The larger figure is the fresher one and comes from the company itself; neither is a vendor filing. Treat all self-reported adoption numbers in this category as marketing until a regulator or an auditor sees them.

Pricing pages in this category move faster than the reviews of them

Every platform in the top five repriced or restructured its plans within the last thirteen months. Make consolidated its three paid tiers into one credit-scaled plan. Activepieces retired its $5-per-active-flow plan in August 2026, and third-party pricing guides published in July 2026 still quote it as current. Latenode moved from per-run pricing to CPU seconds. Lindy restructured at least once during 2026 and dropped its free plan, so reviews published in April 2026 describe tiers and a free allowance that no longer exist. IFTTT’s Pro price is quoted at $2.49, $2.99 and $3.49 a month across different third-party sources. This page names the conflict wherever one exists rather than picking a number quietly, and every figure below was checked against the vendor’s own pricing page on 20 September 2026.

The best AI automation tools compared

Tools are ranked on breadth of use, transparency of pricing, quality of AI integration and evidence of durability. Prices are list prices checked against each vendor’s pricing page on 20 September 2026, and are in US dollars except for n8n, which lists in euros.

1. Zapier — best for app coverage and non-technical teams

Zapier connects more than 9,000 apps, the largest published catalogue of any platform in this comparison, and since ZapConnect in September 2025 has positioned itself as an AI orchestration platform rather than a connector. That repositioning shipped real product: Copilot for building automations from a prompt, Canvas for diagramming a system before building it, Chatbots, Agents and an MCP server that lets Claude, ChatGPT or any MCP-compatible client execute actions across the catalogue.

The pricing is the problem. The Free plan allows 100 tasks a month with unlimited Zaps. Professional starts at $19.99 per month on annual billing, or $29.99 monthly, for 750 tasks, and scales on a slider to two million. Team is $69 per month on annual billing for 2,000 base tasks and up to 25 users. Enterprise is quote-only. Annual billing saves 33 per cent, and Zapier prices overages as a multiple of your base task rate: 1.25 times on annual plans and 2.5 times on monthly ones, so the cost of going over is set by the tier you bought rather than by a flat published rate. Because a task is one successful action step, a five-step automation firing 100 times consumes 500 tasks, where the same work costs n8n 100 executions and Activepieces 100 credits.

Best for: teams with many apps, few technical staff and short workflows. Worst for: long multi-step workflows at volume.

2. Make — best all-rounder and best value for complex workflows

Make offers the strongest visual builder in the category: branching, iterators, aggregators and error handlers are first-class, not bolted on. The Free plan gives 1,000 credits a month, two active scenarios, a fifteen-minute minimum interval between runs and a five-minute cap on any single execution.

Make consolidated its old Core, Pro and Teams tiers into a single paid plan, priced by the credit allowance you choose on a slider. It starts at $9 a month for 5,000 credits and runs to $16 for 10,000, $29 for 20,000 and $53 for 40,000, with tiers continuing into the millions; Make advertises a saving of 15 per cent or more for annual billing, and Enterprise is quote-only. Every paid tier includes unlimited active scenarios, a one-minute minimum interval, forty-minute executions and access to Make’s API.

Make renamed operations to credits on 27 August 2025, converting one for one. Most module actions still consume one credit; AI modules consume credits according to tokens, file size or run time, and running JavaScript or Python in the Make Code app costs two credits per second of execution. Make AI Agents first launched in beta in April 2025; the next generation, which moves agents into the scenario builder with a live reasoning panel and in-canvas chat, shipped on 2 February 2026. Agents run on Make’s own AI provider on every plan, or on your own model key on paid plans. Make does not publish a typical credit cost for an agent run, so measure one before you build on a small tier.

Make is owned by Celonis, which acquired Integromat in 2020 and rebranded it Make in 2022. The Celonis connection gives Make an enterprise process-mining parent, which is unusual for a tool priced at $9.

Best for: teams that need real branching logic at a low price. Worst for: anyone who wants a one-click, no-thought setup.

3. n8n — best for technical teams, and the best economics at scale

n8n bills one execution per workflow run regardless of how many nodes the workflow contains. A two-node workflow and a fifty-node workflow cost exactly the same. For any automation with more than three or four steps, that is the single largest cost difference in this market.

n8n is a Berlin company and lists its cloud prices in euros, which is why third-party guides quoting “$667” for its Business tier are reproducing a euro figure with a dollar sign. On annual billing, Starter is €20 per month with 2,500 executions, Pro is €50 with 10,000, Business is €667 with 40,000 plus SSO, Git sync and multiple environments, and Enterprise is custom. Monthly billing is €24, €60 and €800 respectively, so annual saves roughly 17 per cent. The permanent free cloud plan has been discontinued in favour of a trial. Every plan includes unlimited active workflows.

The self-hosted Community Edition is free with unlimited executions, and managed hosting for it runs from a few dollars a month. One important correction to a claim repeated everywhere: n8n is not open source. It ships under the Sustainable Use Licence, a source-available fair-code licence that permits internal business use but restricts commercial use of the source without a separate embed licence. The Open Source Initiative definition does not permit use restrictions, and n8n does not claim the label itself.

n8n 2.0 shipped in January 2026 with task runners on by default, sandbox isolation, an improved AI Agent cluster and human-in-the-loop tool gates. Commercially, n8n raised a $180 million Series C led by Accel at a $2.5 billion valuation, with NVIDIA’s NVentures among the investors; Tech.eu reported in May 2026 that a strategic investment from SAP more than doubled that to $5.2 billion, alongside a multi-year commercial deal that puts n8n inside SAP’s Joule Studio.

Best for: engineering-adjacent teams, high-volume workflows, data-residency requirements. Worst for: non-technical users with no one to maintain it.

4. Microsoft Power Automate — best for Microsoft 365 estates

Power Automate Premium is $15 per user per month paid yearly and is the governed default for organisations already inside a Microsoft tenant: it inherits Entra identity, data loss prevention policies and admin controls that no standalone tool can match without integration work. Unattended automation is licensed per bot instead, at $150 a month for Power Automate Process or $215 for Hosted Process, which includes a Microsoft-managed virtual machine. There is a 30-day free trial and no free tier.

The material change for 2026 buyers is that the seeded AI Builder credits included with Premium and Process licences are being removed on 1 November 2026. Microsoft’s own licensing documentation states that each Power Automate Premium user licence carries 5,000 seeded AI Builder credits until that date, after which only credits bought through the AI Builder add-on remain active and everything else is removed. The add-on itself, sold in units of one million credits, is closed to new customers, who must buy Copilot Credits instead; existing customers can renew or top up until 1 November 2026. Any business case modelled on the seeded allowance needs rebuilding before that date.

For conversational and autonomous agents, Copilot Studio is the sibling product, and it is now sold two ways: pay-as-you-go billing in Copilot Credits, or a pre-purchase plan of Copilot Credit Commit Units that Microsoft says saves up to 20 per cent. Both require an Azure subscription, and Microsoft no longer publishes a per-credit rate on the pricing page, so the meter is real but the unit cost is not public. A Microsoft 365 Copilot licence at $30 per user per month includes agent building in the Copilot Chat and Standard harnesses; autonomous agents and agents published to external channels need the standalone Copilot Studio billing.

Best for: IT-governed Microsoft organisations. Worst for: small teams outside the Microsoft stack, where the licensing model is disproportionate.

5. Workato — best enterprise integration platform

Workato was named a Leader in the 2026 Gartner Magic Quadrant for iPaaS for the eighth time and furthest in Vision for the third consecutive year. It sells recipes rather than seats, which suits enterprises rolling automation across departments without per-user licensing, and its Agent Studio builds agents it calls Genies, with the Workato One edition adding agentic orchestration.

Pricing is quote-only and Workato publishes no list prices at all: not an entry tier, not a per-recipe rate, not a free tier. Figures circulating in third-party pricing guides are resale estimates rather than vendor numbers, and this page does not repeat them. Recipe packs, premium connectors, high-volume recipes, on-premise access and extra seats are all charged separately, so the shape of the quote matters as much as its headline.

Best for: enterprises with integration governance requirements and a procurement function. Worst for: anyone who needs to know the price before a sales call.

6. Gumloop — best AI-native builder for document and research workflows

Gumloop is built AI-first rather than AI-added: the canvas assumes model calls, and its sub-agent feature runs multiple agents in parallel inside a single pipeline, which most competitors cannot do. Its published pricing is now two lines. Pro starts at $37 a month and includes 20,000 credits, unlimited agents, unlimited seats, access to more than 35 models, one hosted MCP server and a 14-day free trial; Enterprise is custom and adds custom credit volumes, a virtual private cloud option, role-based access control, SCIM and SAML, audit logs and custom data retention. There is no free plan. Both tiers carry an 8 per cent orchestration fee on top of credit consumption, with discounts available on Enterprise, and you can bring your own model API keys on either.

Gumloop raised a $50 million Series B led by Benchmark, which it advertises on its own site, and names Shopify, Instacart, Ramp, Gusto, Samsara and Webflow as customers. Total funding is reported inconsistently across trackers and is not stated here.

Best for: teams whose workflows are mostly model calls over documents and web data. Worst for: high-volume, low-AI plumbing, where credits are the wrong meter, and anyone who needs to trial a platform without a card.

7. Lindy — best assistant-style automation for non-technical operators

Lindy builds automations that behave like a delegated assistant, handling meeting scheduling, inbox triage and lead follow-up rather than presenting a flowchart. It is the only platform on this page priced per user: Plus is $29.99 per user per month with 3,000 credits per user, Pro is $99.99 with 15,000 and Max is $199.99 with 35,000, with Enterprise adding HIPAA compliance, a signed business associate agreement, shared usage, audit logs and dedicated support. Credits pool across the team.

Lindy publishes credit bands rather than a per-action rate, and they are wide: everyday requests such as a summary, a lookup or a drafted reply cost 2 to 250 credits; deep work such as researching a competitor and writing the report costs 250 to 1,000; large builds such as a multi-source dashboard cost 1,000 to 2,500. On the Plus tier, a single big build can therefore consume most of a month’s allowance for that seat. That is the honest reason this ranks seventh rather than higher: the product is good and the unit cost is unforecastable.

Lindy repriced at least once during 2026 and removed a free plan that earlier reviews still describe.

Best for: small commercial teams who want outcomes without building. Worst for: anyone needing predictable unit costs.

8. Pipedream — best for developers who want to write real code

Pipedream sits closer to a serverless function platform with a workflow interface than to a traditional automation tool: every step can be Node.js, Python or Go, across more than 3,000 integrated apps. The Free plan gives 100 credits a month, three active workflows, three connected accounts and one million AI tokens. Basic is $29 a month with 2,000 credits, ten active workflows and 20 million AI tokens; Advanced is $49 with 2,000 credits, unlimited workflows, unlimited connected accounts, control-flow operators, premium apps and GitHub sync; Connect is $99 with 10,000 credits and managed authentication for up to 100 external users at $2 per additional user; Business is custom.

The governance point for buyers: Workday announced a definitive agreement to acquire Pipedream on 19 November 2025 and closed the deal in February 2026. Ownership by an enterprise HR and finance vendor changes the roadmap risk profile for independent users, and it is the reason this sits eighth despite being the best developer experience here.

Best for: developers who want code-first workflows without managing infrastructure. Worst for: non-technical teams.

9. Activepieces — best genuinely open-source option

Activepieces ships its Community Edition under the MIT licence, which makes it the most permissively licensed credible platform in this category, a real distinction from n8n’s fair-code licence and the reason it ranks here at all. The distinction has a limit worth stating plainly: the MIT licence covers the Community Edition, while the enterprise features are released under a commercial licence, so self-hosting is unrestricted but not feature-complete.

Its pricing changed in August 2026 and most published guides are now wrong about it. Activepieces retired the $5-per-active-flow plan with unlimited runs and moved to credits: Free gives 100 credits a day with unlimited flows, Plus is $20 a month with 10,000 credits a month and up to five users, Team is $200 with 50,000 credits, 25 users, unlimited projects and SSO, and Ultimate is custom. Annual billing saves two months. One flow run costs one credit; an AI step costs two credits on a fast model, ten on a mid-tier one and twenty on a frontier model, falling to one credit on any model if you bring your own AI keys, which Plus and above allow. Overage is $0.007 per credit. Embedding the builder in your own product starts at $36,000 a year.

The change matters directionally as well as numerically. Under the old meter, a high-frequency automation was effectively free once the flow was paid for. Under the new one, frequency is the cost, and a single flow calling a frontier model on every run can exhaust a monthly allowance quickly.

Best for: teams that need a permissive open-source licence or unrestricted self-hosting. Worst for: anyone who budgeted on the pre-August 2026 per-flow pricing and has not re-modelled.

10. Latenode — best low-cost AI-first low-code platform

Latenode combines a visual editor with inline JavaScript, a headless browser, more than 5,500 integrations and access to more than 335 models. It is the only platform here that has abandoned per-operation billing entirely: it charges for CPU seconds, so the number of steps in a workflow does not multiply the bill, and only the time the workflow actually runs does.

The free tier is 10,000 CPU seconds a month with five active workflows, which the vendor equates to about 10,000 one-second runs. Beyond that there is no base fee and rates fall in brackets from $0.00012 per CPU second to $0.00005 at the top of the scale. Latenode’s own published examples put 100,000 CPU seconds a month at $10.80, whether that is 100,000 one-second runs or 10,000 ten-second ones, 500,000 at $48.80 and one million at $83.80. Those are the lowest published figures in this comparison at equivalent volumes. It ranks tenth rather than higher because it is a newer entrant with a shorter track record than the platforms above it, and because CPU-second billing rewards efficient workflows and punishes slow ones in a way that is hard to forecast before you build.

Best for: cost-sensitive teams running many short workflows. Worst for: enterprises needing an established vendor, and anything with long waits inside a run.

11. UiPath — best for desktop and legacy-system automation

UiPath is the robotic process automation incumbent and the right answer when the systems you need to automate have no API: mainframes, desktop applications, legacy ERP screens. It repositioned as an agentic automation platform at its FUSION conference in September 2025, with Maestro as an orchestration layer coordinating robots, AI agents and human approvals in one process.

Pricing runs on two frameworks, credit-based Unified pricing and licence-based Flex pricing. Agent Units meter AI agent execution in production, while design-time development and testing do not consume them. UiPath does not publish list prices, and the cost estimates that circulate for platforms in this class come from consultancies selling implementation services, so this page does not repeat them. The one durable generalisation is that RPA is priced for programmes, not teams, and that an agentic add-on bought on top of an existing RPA estate is a second line item rather than a replacement for the first.

Best for: enterprises automating systems without APIs. Worst for: anything an API-based tool can already reach.

12. IFTTT — best for consumer and smart-home automation

IFTTT remains the simplest automation tool and the only one on this page built primarily for consumer devices. The Free plan allows two applets; Pro is $2.99 a month for 20 applets, multi-action applets, faster execution and webhooks; Pro+ is $8.99 a month for unlimited applets and additional features. Those are the figures IFTTT’s own documentation gives; third-party sources also circulate $2.49, $3.49, $8.49 and $14.99, which appear to be stale or regional. Check the plans page before committing.

Best for: smart-home and personal automation. Worst for: business workflows of any complexity.

Enterprise integration platforms worth naming

Three platforms serve the integration end of this market rather than the workflow-builder end, and all three are 2026 Gartner iPaaS Leaders: Boomi, a Leader for the twelfth consecutive time and highest for Ability to Execute; SAP Integration Suite, a Leader for the sixth time and now embedding n8n inside Joule Studio; and Salesforce MuleSoft. All three are quote-only, all three are priced for programmes rather than teams, and none is a sensible purchase for a department that wants to connect a form to a spreadsheet.

How each platform meters, and why it matters more than the price

The billable unit determines whether a platform is cheap or ruinous for your particular workload. This is the table to read before the pricing table.

The worked example that makes the difference concrete: a five-step workflow running 1,000 times a month consumes 5,000 Zapier tasks, 5,000 Make credits, 1,000 n8n executions and 1,000 Activepieces credits. Same automation, same volume, four different consumption profiles. On Zapier’s Professional slider that workload needs a 5,000-task tier; on Make it exactly exhausts the $9 entry tier, so you would buy the $16 one; on n8n Starter it uses 40 per cent of a 2,500-execution allowance; and on Activepieces Plus it uses a tenth of the monthly credits. If each run takes about a second, the same job sits inside Latenode’s free 10,000 CPU seconds.

Pricing comparison: what you will actually pay

Figures were checked against each vendor’s pricing page on 20 September 2026. Annual-billing prices are shown where the vendor discounts for annual commitment. All prices are in US dollars except n8n, which lists in euros.

Best AI automation tool by use case

Best overall for most teams: Make

Make at $9 a month for 5,000 credits, or $16 for 10,000, gives the best combination of price, branching logic and error handling, and its AI Agents are available on every plan.

Best for the widest app coverage: Zapier

Zapier connects more than 9,000 apps and is the only platform here where the answer to “does it integrate with this obscure tool” is almost always yes.

Best for technical teams and self-hosting: n8n

One execution per workflow run, unlimited active workflows on every plan, and a free self-hosted Community Edition under the Sustainable Use Licence. Cloud starts at €20 per month on annual billing.

Best for a Microsoft 365 organisation: Microsoft Power Automate

Premium at $15 per user per month inherits tenant identity and data loss prevention policies. Re-model the AI portion before 1 November 2026.

Best for a large enterprise with governance requirements: Workato

A Gartner iPaaS Leader for the eighth time and furthest in Vision for three consecutive years, with recipe-based rather than seat-based licensing.

Best for AI-dense document and research pipelines: Gumloop

Parallel sub-agents and an AI-first canvas, at $37 a month for 20,000 credits, with a virtual private cloud option for enterprise buyers.

Best for developers: Pipedream

Real Node.js, Python or Go in every step, from a free tier to $49 a month for unlimited workflows, with the Workday acquisition as the roadmap caveat.

Best genuinely open-source platform: Activepieces

An MIT-licensed Community Edition, materially more permissive than n8n’s fair-code Sustainable Use Licence, with free self-hosting and cloud plans from $20 a month since the August 2026 move to credits.

Best for legacy systems and desktop automation: UiPath

The only platform here designed for systems with no API, with Maestro orchestrating robots, agents and human approvals in one process.

Best on price at volume: self-hosted n8n, with Latenode as the managed alternative

Self-hosted n8n Community Edition is free with unlimited executions on hosting that costs a few dollars a month. Among managed platforms, Latenode’s own published examples put 100,000 CPU seconds a month at $10.80.

Best for personal and smart-home automation: IFTTT

Two applets free, and a paid tier in the low single digits of dollars per month.

Best free option: Make, or self-hosted Activepieces or n8n

Make’s Free plan gives 1,000 credits and two active scenarios indefinitely. Activepieces and n8n both ship Community Editions that are free to self-host, though n8n’s is under a fair-code licence rather than an open-source one and Activepieces reserves its enterprise features for a commercial licence. Note that n8n no longer offers a permanent free cloud tier, and Gumloop, Lindy and Workato have no free tier at all.

What buyers and builders actually report

The meter, not the feature, is what surprises people. The most common complaint across this category is a bill that scaled faster than the workload, and it is almost always a meter mismatch: a long workflow on a per-action meter, or an AI agent on a credit plan sized for plain automation. The clearest published example is Activepieces, where a frontier-model step costs twenty times a plain flow run.

AI steps are used, but narrowly. The pattern reported most often is classification, extraction and summarisation inside an otherwise deterministic workflow: reading an inbound email and routing it, pulling fields out of an invoice, summarising a transcript into a record. The MIT NANDA finding that the measurable returns sit in back-office automation is consistent with that pattern.

Self-hosting is a real cost, not a free lunch. n8n Community Edition and Activepieces Community Edition are free as software. Upgrades, backups, monitoring and the engineer who owns them are not. Teams that self-host to save $50 a month and then spend an afternoon a month maintaining it have not saved anything.

Vendor durability has become a purchasing criterion. Relay.app announced its wind-down on 16 July 2026 and ended paid access on 14 September, roughly two months later, and has since deleted all workflows and run history. That is a reasonable wind-down by industry standards, and it is still two months to migrate everything you built. Ask about export formats before you build.

How to choose in September 2026

1. Count your steps before you compare prices. Multiply steps by expected monthly runs. If the answer is large, a per-run meter such as n8n’s execution or Activepieces’ credit, both of which cover a whole workflow, will beat a per-action meter at any list price.
2. Price the AI separately. Assume the plan price covers the plumbing only. Ask what an AI step costs in the platform’s own units, and what happens at the cap. Where a platform lets you bring your own model keys, as Activepieces, Gumloop and Make do, that is usually the cheaper route at volume.
3. Check whether you actually need an agent. If the steps are knowable in advance, a deterministic workflow is cheaper, auditable and more reliable. Gartner’s expectation that over 40 per cent of agentic projects will be cancelled by the end of 2027 is a reason to start deterministic and add autonomy only where it earns its place. See best AI agents if it does.
4. Match governance to the organisation you actually are. A Microsoft tenant with a compliance function should look at Power Automate first. A five-person team should not.
5. Ask how you get your workflows out. One platform in this comparison closed during 2026 and another changed owner. Export capability is a feature.
6. Start on a free tier or trial and measure real consumption for a month. Zapier, Make, Activepieces, Pipedream, Latenode and IFTTT all have permanent free tiers; n8n, Power Automate and Gumloop offer trials only; Lindy and Workato offer neither. Modelled consumption is consistently wrong. Measured consumption is not.

Frequently asked questions

What is the best AI automation tool in 2026?

Make is the best AI automation tool for most teams as of September 2026, because its single paid plan starts at $9 a month for 5,000 credits and scales by credit tier, and its visual builder handles branching, iteration and error handling that simpler tools cannot. Zapier is the better choice when app coverage matters most, since it connects more than 9,000 apps. n8n is the better choice for technical teams, because it bills one execution per workflow run regardless of the number of steps.

How much do AI automation tools cost?

Entry pricing for self-serve AI automation platforms in September 2026 runs from $9 a month for Make’s 5,000-credit tier to $37 a month for Gumloop Pro, with Zapier Professional at $19.99 on annual billing, Activepieces Plus at $20, n8n Starter at €20, Pipedream Basic at $29 and Lindy Plus at $29.99 per user. Latenode charges for runtime instead, with no base fee beyond a free allowance of 10,000 CPU seconds a month. Enterprise platforms including Workato and UiPath publish no list prices at all.

What is the difference between an AI automation tool and an AI agent?

An AI automation tool executes a workflow you designed, in a fixed order, with AI available as one step inside it. An AI agent is given a goal and chooses its own steps at run time. The practical consequence is cost and auditability: deterministic workflows bill a predictable unit per run and produce a repeatable log, while agents bill unpredictably and take a different path each time. Zapier bills its Agents product on a separate activity meter from Zap tasks for exactly this reason.

Is n8n open source?

No. n8n ships under the Sustainable Use Licence, a source-available fair-code licence that allows free use for internal business operations and non-commercial purposes but restricts commercial use of the source code without a separate embed licence. The Open Source Initiative definition does not permit restrictions on use, and n8n does not describe itself as open source. Activepieces, whose Community Edition is MIT-licensed, is the genuinely open-source alternative, though its enterprise features sit under a commercial licence.

Which AI automation tool is cheapest at high volume?

Self-hosted n8n Community Edition is cheapest at high volume because it is free software with unlimited executions, leaving only hosting costs of a few dollars a month. Among managed platforms, Latenode publishes the lowest figures: its own worked examples put 100,000 CPU seconds a month at $10.80 and one million at $83.80, because it bills runtime rather than operations. Activepieces is no longer the cheap answer for high-frequency workflows: it retired its $5-per-active-flow plan with unlimited runs in August 2026 and now charges one credit per run.

What is a Zapier task and how many do I need?

A Zapier task is one action step that runs successfully; triggers and filters that stop a Zap do not count. A five-step Zap that fires 100 times a month consumes 500 tasks. Estimate your requirement by multiplying the number of action steps in each automation by how often it runs. Zapier’s Free plan allows 100 tasks a month and Professional starts at 750 and scales to two million, with overages charged at 1.25 times your base task rate on annual plans and 2.5 times on monthly ones.

What changed in Make’s pricing?

Two things. Make renamed its billing unit from operations to credits on 27 August 2025, converting one for one, which allowed AI and code modules to consume more than one unit per call: standard module actions still cost one credit, AI modules consume credits according to tokens, file size or run time, and running code costs two credits per second. Make has since consolidated its Core, Pro and Teams tiers into a single paid plan scaled by credit allowance, starting at $9 a month for 5,000 credits and rising to $16 for 10,000 and $29 for 20,000. Guides still quoting three named tiers are describing the old structure.

Is Power Automate losing its AI credits?

Yes, in part. Microsoft’s licensing documentation states that the seeded AI Builder credits included with Power Automate Premium and Process licences, 5,000 per Premium user licence, will be removed on 1 November 2026. From that date only credits bought through the AI Builder add-on remain active, and AI Builder features otherwise draw on Copilot Credits. The add-on, sold in units of one million credits, is already closed to new customers, who must buy Copilot Credits instead; existing customers can renew or top up until 1 November 2026. Anyone who built a business case on the seeded allowance should re-model the AI portion before that date.

Did any automation platform shut down in 2026?

Yes. Relay.app’s own site states that it shut down in September 2026 and that all accounts, workflows and run history have been permanently deleted. TechCrunch reported that the wind-down was announced on 16 July 2026, that free accounts closed on 15 August and paid access ended on 14 September, that annual customers received prorated refunds and every workspace was given an additional 25,000 steps and 10,000 AI credits a month through the transition, and that founder Jacob Bank and several staff joined Google’s Chrome team.

Which automation platforms are Gartner Magic Quadrant Leaders?

In the 2026 Gartner Magic Quadrant for Integration Platform as a Service, published on 16 March 2026 and covering 18 vendors, the Leaders were Boomi, Workato, SAP and Salesforce MuleSoft. Boomi was a Leader for the twelfth consecutive time and placed highest for Ability to Execute; Workato was a Leader for the eighth time and furthest in Vision for the third consecutive year; SAP was a Leader for the sixth time. Celigo, SnapLogic and Tray.ai were positioned as Visionaries. Zapier was one of the 18 vendors evaluated; Make and n8n were not included, because Gartner assesses enterprise integration platforms rather than self-serve builders.

Can AI automation tools connect to ChatGPT or Claude directly?

Yes, through the Model Context Protocol. Zapier operates an MCP server that it says exposes more than 30,000 actions across its 9,000-plus apps to any MCP-compatible client, including Claude and ChatGPT, and n8n ships MCP nodes in both directions so an existing workflow can be published as a tool an AI host calls. Activepieces, Make, Gumloop and Latenode all ship MCP support as well. The practical effect is that the automation platform becomes the execution layer and the model becomes the caller.

Do AI automation tools actually save money?

The evidence is mixed and the honest answer is that most organisations have not proven it yet. MIT’s Project NANDA study found that 95 per cent of enterprise generative AI pilots produced no measurable profit-and-loss impact, while also finding that the largest measurable returns came from back-office automation rather than the sales and marketing tools that absorbed more than half of generative AI budgets. That study is not peer reviewed and rests on a small interview base, so it is a signal rather than a settled finding. The implication for buyers is to automate internal processes with clear before-and-after metrics first.

What should a small business choose?

A small business with no technical staff should start on Make’s Free plan, which gives 1,000 credits and two active scenarios indefinitely, and move to the $9 tier when it runs out. If the workflows depend on a long tail of niche applications, Zapier is worth the higher price for its catalogue of more than 9,000 apps. Microsoft Power Automate is only worth it if the business is already paying for Microsoft 365 licences.

How do I avoid a surprise automation bill?

Measure before you commit. Build the workflow on a free tier, run it for a month, and read the actual consumption in the platform’s own units rather than modelling it. Then check three things: the overage rate, whether AI steps draw on a second meter, and what happens when you hit the cap. Zapier charges overages as a multiple of your base rate rather than a flat fee, Activepieces charges $0.007 per credit over, and Zapier does not publish an overage rate for Agents activities at all, so that cap is a hard stop rather than a meter.

Conclusion: how to choose in September 2026

Three sentences should govern this purchase. The platforms have converged on features, so the meter is the product. AI is a second meter almost everywhere, and the plan price tells you nothing about it. And the measurable returns, on the best evidence available, sit in dull internal processes rather than the customer-facing workflows that attract most of the budget.

• Most teams: Make, from $9 a month for 5,000 credits or $16 for 10,000.
• Widest app coverage: Zapier, Professional at $19.99 per month on annual billing, more than 9,000 apps.
• Technical teams: n8n, Starter at €20 per month on annual billing, or free self-hosted.
• Microsoft estates: Power Automate Premium at $15 per user per month, re-costed for 1 November 2026.
• Enterprises: Workato, a Gartner iPaaS Leader for the eighth time, quote only.
• AI-dense pipelines: Gumloop, $37 a month for 20,000 credits.
• Developers: Pipedream, free to $49 a month for unlimited workflows.
• Open-source requirement: Activepieces, MIT-licensed Community Edition, Plus at $20 a month or free self-hosted.
• Legacy systems: UiPath, priced for programmes rather than teams.
• Personal and smart home: IFTTT, from a few dollars a month.

Related reading: best AI agents for autonomous systems that choose their own steps, best AI agent frameworks for building agents in code, what is agentic AI for the concepts, and best AI for business for the wider commercial stack.

Prices on this page were checked against each vendor’s own pricing page on 20 September 2026 and change frequently. All figures are US list prices except n8n’s, which the vendor lists in euros and which are shown in euros here rather than converted. Five platforms repriced or restructured within the last thirteen months and stale figures are widely republished elsewhere: Make moved from operations to credits on 27 August 2025 and has since consolidated its three paid tiers into one credit-scaled plan; Activepieces retired its $5-per-active-flow plan in August 2026 in favour of credits; Latenode moved from per-run pricing to CPU seconds; Lindy repriced and removed its free plan; and Microsoft removes seeded AI Builder credits on 1 November 2026. Workato and UiPath publish no list prices; third-party estimates circulate for both and are not repeated here. IFTTT’s Pro and Pro+ prices are quoted differently across current third-party sources and both the figure used and the alternatives are shown. n8n’s adoption figures conflict between sources and are attributed rather than asserted, as is its claim that more than 80 per cent of workflows on the platform involve AI agents. Relay.app closed in September 2026 and is listed here as closed, not as an option.