Top 22 AI Workflow Automation Tools

Every platform in this guide can already spin up an "AI agent," and many now speak MCP (Model Context Protocol, the emerging standard that lets an agent call external tools). Eighteen months ago that was a selling point. Today it is table stakes. When agents, MCP, and a visual builder show up across most of the field, the feature checklist stops telling you which tool to standardize on. Two things that the marketing pages bury do: a price you can forecast at scale, and fit with the systems you already run.

Zapier tops the field on raw reach, with 9,000+ integrations and native AI agents. But because agentic capability is now universal, the right pick turns on buyer fit and cost predictability. Technical teams gravitate to n8n, Make, or Pipedream. Salesforce and Microsoft shops default to Agentforce and Copilot Studio. Regulated enterprises weigh Automation Anywhere, Workato, or UiPath despite sales-quote-only pricing. This guide ranks all 22, but read the two tables first: they answer the questions procurement actually asks.

The real dividing line is cost you can predict

A complaint that recurs across many of these platforms is not capability. It is a bill you cannot forecast. Roughly six platforms make you call sales for their paid or upper tiers, which blocks apples-to-apples comparison during procurement:

• Workato (every tier)
• Tray.ai (every tier)
• Automation Anywhere (paid Cloud and Enterprise tiers)
• UiPath (Standard and Enterprise editions)
• Relevance AI (paid tiers)
• CrewAI (Enterprise tier, its only paid plan)

Count the partials already in the pricing table and the set that is hard to price before a sales call pushes toward half: Pipedream publishes a free tier but no paid-tier dollar figures, and IBM watsonx Orchestrate lists prices only through its AWS Marketplace listing, not its own pages. Several of these tools do publish a free tier, so the problem is not the absence of any number. It is that the number governing your bill at scale is the one you cannot see.

IBM watsonx Orchestrate is the instructive case. Its price is public, but visible only through its AWS Marketplace listing ($6,360/year Essentials, $76,320 Standard, $216,000 Premium), not on IBM's own pages. It is priced, just not where you would look first.

The platforms that do publish a number meter it in incompatible units, so the same workload can cost very differently across two tools:

Usage metering is why unpredictable cost hits even the leaders. Capterra reviewers say Zapier's task pricing "escalates quickly" with volume. Agentforce runs "$2 USD per conversation," which G2 reviewers call hard to budget for at scale. Microsoft's own documentation states Copilot Studio disables custom agents once a tenant reaches 125% of prepaid Copilot Credit capacity. If you are choosing a platform to run at task or seat scale for the next two years, model your volume before the sales call, not after.

How each tool charges

This table answers the top procurement anxiety directly: what unit each platform meters, whether a permanent free tier exists, the published entry price, and whether you can price it at all without a sales call.

Master comparison: integrations, compliance, and buyer fit

Use this to filter 22 tools down to a three or four name longlist before reading any of the profiles below. Integration count is the vendor's own stated figure; compliance lists the certifications you can cite to procurement; buyer fit names the lane each tool serves best.

The 22 tools, ranked

1. Zapier, best for integrations and enterprise reach

Zapier connects 9,000+ apps, the largest ecosystem here, and pairs that reach with genuine agentic tooling rather than a cosmetic AI layer: autonomous AI Agents, AI Guardrails, and MCP support for Claude and ChatGPT. Satisfaction holds at volume, 4.5/5 across 1,754 G2 reviews and 4.7/5 across 3,023 Capterra reviews.

• Pick it if you need to connect many apps with no developer involvement and want strong enterprise governance behind it.
• Skip it if usage is heavy but budget-sensitive: Capterra reviewers repeatedly say task pricing escalates quickly with volume and lacks a well-priced mid-tier, and the Free plan's 100-task cap pushes real use into paid fast.

The cost trap to watch is the gap itself: there is no well-priced tier between the 100-task free plan and metered paid usage, so light-but-premium users get stranded exactly where they need a middle option.

2. Make.com, best for complex visual workflows on a budget

Make handles complex, multi-step automations that simpler tools like Zapier cannot, at some of the lowest prices here, and Capterra reviewers (4.8/5 across 407 reviews) make value-for-price a recurring theme. Its 3,000+ integrations include deep native AI support (OpenAI, Claude, Gemini, Perplexity, DeepSeek), plus Make AI Agents, a Library of Agents, and a Make MCP Server.

• Pick it if you have someone comfortable with arrays, regex, and error handling and want power without enterprise pricing.
• Skip it if you need fast support or polish: reviewers flag slow support and missing quality-of-life features (no auto-save, no dark mode, no subfolders).

One fact the tables cannot hold: Make is now part of Celonis rather than independent, so weigh roadmap continuity before you standardize a business-critical process on it.

3. Automation Anywhere, best for power users

Automation Anywhere is the enterprise-RPA (robotic process automation) heavyweight here, and it earns the guide's best-for-power-users pick. Depth backs the power-user title: AI Agent Studio supports 26 predefined models across Anthropic, OpenAI, Google AI, DeepSeek, and Groq, plus bring-your-own-key (BYOK) connections to Azure OpenAI, OpenRouter, and AWS Bedrock. Track record backs dependability: 8 consecutive years as a Gartner Magic Quadrant Leader for RPA, named customers including SoftBank, KeyBank, NHS, and Cisco, and one of the deepest compliance stacks in the guide (ISO 42001 for AI management, ISO 27001/27017/27018, HITRUST CSF, SOC 1/2/3 Type 2, and a HIPAA report).

• Pick it if you run regulated enterprise RPA.
• Skip it if you need public pricing: only the free Community Edition (capped at 5 machines and 100 Document Automation pages/month) is listed, and users report high licensing cost and painful version upgrades.

Budget for the full cost of ownership up front: the quote-only licensing and steep learning curve mean the real number only surfaces after the sales call, not before you shortlist.

4. n8n, best for engineering-led teams

n8n pairs an unusually large catalog, 1,936 integrations including a full AI/LangChain node category (agents, chains, memory, vector stores, MCP), with real developer momentum: 195,600+ GitHub stars (a top-50 project) and named customers including Microsoft, Meta, NVIDIA, and Mercedes-Benz. It ships multiple pre-built agent architectures (Tools, ReAct, Conversational, Plan-and-Execute).

• Pick it if you want to self-host or combine visual building with custom JavaScript and Python.
• Skip it if your builders are non-technical: the node and code design carries a genuine learning curve, and support below Enterprise is forum-based.

Watch one pricing cliff: SSO and SAML sit behind the jump to the €667/month Business tier, so an identity-governance requirement can multiply your bill far beyond the €20 entry point.

5. Salesforce Agentforce, best for Salesforce shops

Agentforce runs natively on your Salesforce CRM data: the Atlas Reasoning Engine and ready-to-go agents (Service Agent, Sales SDR) extend through Flows, Prompts, Apex, and MuleSoft APIs, and the Einstein Trust Layer enforces zero-retention agreements and data masking. G2 rates it 4.3/5 across 1,194 reviews and named it the #1 Agentic AI Product in its 2026 Best Software Awards.

• Pick it if you are Salesforce-anchored.
• Skip it if you are not: the deepest value is locked to that platform, and reviewers call the metered pricing hard to forecast at scale.

Model the metering carefully before you commit: it charges two ways at once, "$2 USD per conversation" or Flex Credits at "$500 USD per 100,000 Credits" where one action equals 20 credits ($0.10 per action), and Enterprise Edition includes 100,000 complimentary credits via Salesforce Foundations.

6. Microsoft Copilot Studio, best for Microsoft 365 shops

Copilot Studio is the Microsoft-native choice: multi-agent orchestration, voice and phone agents, Computer-Using Agents, MCP server support, and 1,400+ external connectors, with native publishing to Teams, SharePoint, and Microsoft 365 Copilot, and Microsoft citing adoption by 90% of the Fortune 500.

• Pick it if you already license Microsoft 365 Copilot at $30.00/user/month (paid yearly), where building internal agents is included at no extra cost.
• Skip it if you sit outside the Microsoft stack, or if debugging load worries you: G2 reviewers report agents becoming complex and slow to debug as flows grow.

The capacity cliff to plan around is enforced automatically: overage disables custom agents once a tenant hits 125% of prepaid Copilot Credit capacity, so size your credit packs before you scale, not after agents go dark.

7. Workato, the enterprise iPaaS with no public price

Workato is a mature enterprise iPaaS built for broad system integration plus agentic automation: 1000+ pre-built connectors (Salesforce, Slack, NetSuite, ServiceNow, Workday), native AI agent tooling (Agent Studio, pre-built "Genies," an MCP Gateway governing Claude, OpenAI, and Copilot models), and a deep compliance stack (SOC 1/2/3, ISO 27001/27701/42001, PCI DSS Level 1, HIPAA, and GDPR). G2 reviewers (4.7/5 across 778 reviews) praise support quality.

• Pick it if you need enterprise-grade integration breadth and can run a procurement cycle.
• Skip it if you need a public number to shortlist: every tier is a sales quote.

Two ceilings the compliance story hides can bite at scale: data-transformation caps (CSV parsing limited to 50,000 records, Lookup Table to 100,000 rows), and reviewer reports that costs grow unpredictable as volume climbs.

8. Pipedream, best for API-first developers (mid-acquisition)

Pipedream is developer-first: Connect provides managed OAuth across 3,000+ apps and 10,000+ pre-built tools, MCP servers expose those tools to agents, and custom code steps support npm packages.

• Pick it if your team writes code and wants tool-level agent access across a large catalog.
• Skip it if you need a firm price or a proven track record: exact paid-tier figures could not be extracted from the JavaScript-rendered pricing page during research, a Capterra reviewer cited a recent pricing change that made scaled use "very expensive," and the review base is thin (6 Capterra reviews).

The open question is ownership: Pipedream entered a definitive agreement to be acquired by Workday (announced November 19, 2025), with no public roadmap yet for the standalone product.

9. Latenode, best value

Latenode wins best value by pricing in CPU-seconds with full transparency and a real free tier: Pay-as-you-Go carries no base fee and bills from $0.00012 down to $0.00005 per CPU-second at higher volumes, which reviewers and the vendor both frame as far cheaper than Zapier or Make at scale. It ships a native AI Agent node and AI Code Copilot (not a bolt-on) plus 5,500+ connectors across 1,200+ apps and LLMs, and sentiment is strong (4.7/5 on G2, 4.9/5 on Capterra).

• Pick it if you are a cost-conscious technical team or SMB.
• Skip it if you need the widest native catalog or a long track record: the integration catalog trails Zapier and Make, and documentation lags.

The viability caveat is company signal, not product: Latenode discloses no funding, founding date, or independent audit data, so it carries thinner staying-power evidence than any incumbent here.

10. Relevance AI, best for enterprise agent teams with implementation support

Relevance AI targets mid-market and enterprise teams with a dedicated implementation cycle (the vendor cites 6-week deployments). The stack is broad: a no-code agent builder, multi-model routing (Claude/GPT/Gemini), RAG (retrieval-augmented generation) with hybrid retrieval, long-term agent memory, and a built-in evals framework, alongside named customers like Canva, KPMG, and Autodesk, with 1.24M tasks run monthly and a stated 96.4% eval pass rate.

• Pick it if you have implementation support and want a deep enterprise agent stack.
• Skip it if you want instant self-serve pricing: the public page shows only a "Talk to Sales" Enterprise tier, and the G2 review base is thin at 20 reviews.

Read the refund terms before you sign: reviewers specifically flag high pricing and no prorated refunds, which raises the cost of guessing wrong on capacity.

11. Gumloop, best for visual multi-model automations

Gumloop is a code-optional agent builder with 100+ integration nodes spanning CRM, docs, chat, and MCP servers, supporting multiple LLMs (Claude, GPT, Gemini, DeepSeek) with no vendor lock-in. The enterprise story is credible for a young company: a $50M Series B led by Benchmark and named customers including Gusto and Shopify.

• Pick it if you want visual, multi-model automations with real access controls.
• Skip it if you need heavy concurrency or proven sentiment: free and Pro tiers cap concurrent runs low (2 and 5 respectively), and third-party review volume is very thin (2 Capterra reviews).

The concurrency caps are the constraint the price tag does not signal: at 2 and 5 concurrent runs on the lower tiers, high-throughput workloads hit a ceiling before you feel the cost.

12. Lindy.ai, best for individual pros automating email and CRM

Lindy sits in one clear lane, personal-productivity automation: pick a trigger, connect apps, give natural-language instructions, and let it run email, meeting, calendar, and CRM follow-ups across hundreds of connectors (Gmail, Slack, HubSpot, Salesforce, Zendesk), with human-in-the-loop review before sending. G2 satisfaction is high at 4.9/5 across roughly 170 reviews.

• Pick it if you are an individual pro or small team automating your own inbox and CRM.
• Skip it if you want a free tier or heavy volume: there is only a 7-day trial before $49.99+/month, and users call the credit caps restrictive.

Two reliability flags are worth heeding beyond the pricing: reviewers report AI instruction-following drifting over time, and some cite billing problems including being charged after account deletion.

13. UiPath, best for enterprise RPA with dedicated teams

UiPath spans a vast product surface under one platform: Studio, Orchestrator, Maestro, Agent Builder, Autopilot, IDP/IXP document processing, Process Mining, Task Mining, and Test Cloud, with 164+ Marketplace connectors and deep SAP and Microsoft ties. The review base is large (Capterra 4.6/5 across 723 reviews, one cited case cutting invoice processing from 3 hours/day to 20 minutes).

• Pick it if you have a dedicated automation team and enterprise RPA needs.
• Skip it if you want simple self-serve pricing or rock-solid stability: only the entry Basic tier is public (from $25/month, Europe region only), Standard and Enterprise are quote-only, and reviewers report frequent crashes and version upgrades that break compatibility.

The stability tax is the underrated cost here: reviewers repeatedly cite crashes, unclear error messages, and upgrade breakage, which a dedicated team absorbs but a lean one cannot.

14. Tray.ai, best for hybrid IT integration

Tray.ai aims squarely at technical and IT teams: 700+ (elsewhere 474+) prebuilt connectors plus a universal HTTP client and an on-prem agent for hybrid and legacy systems, dual build modes (visual Tray Build and code-based Tray Headless for Claude Code), and purpose-built agent tooling (Merlin Agent Builder, an Agent Gateway for MCP governance). G2 rates it 4.5/5 across 158 reviews.

• Pick it if you need to orchestrate complex, hybrid, or legacy integrations.
• Skip it if you want a public price or a gentle onboarding: pricing is fully quote-based with only a free trial, and reviewers flag a cluttered UI and slow support.

Factor onboarding as a real line item: the platform assumes API fluency, and reviewers cite a steep curve for anyone unfamiliar with APIs, so ramp time is a cost lean teams underestimate.

15. Relay.app, best for non-technical operators

Relay.app leads with chat-first, natural-language workflow building backed by multi-model AI steps (Anthropic, OpenAI, Gemini, and others), plus sequences, tables, MCP servers, human-in-the-loop approvals, and custom JavaScript steps. User sentiment is excellent on G2 (4.9/5 across 70 reviews), especially for support responsiveness.

• Pick it if you are a non-technical operator or small team wanting agentic automation without technical setup.
• Skip it if you need broad integrations or proven longevity: the 350+ integration catalog is smaller than Zapier or Make, a recurring reviewer complaint.

The bet you are placing is on an early seed-stage company ($8.2M raised from Khosla Ventures, a16z, and others), so weigh long-term continuity against the strong day-one experience.

16. Taskade, best for an affordable all-in-one workspace

Taskade combines task and project management with app generation, automation workflows, and multi-model agents (15+ models across OpenAI, Anthropic, and Google, assignable per agent), with 145+ connectors and MCP/Developer API support, at one of the lowest entry prices here. Ratings are strong (4.7/5 on both G2 and Capterra) with praise for collaboration and value.

• Pick it if you want an inexpensive all-in-one workspace rather than a single-purpose automation specialist.
• Skip it if you are a regulated buyer or need automation depth: reviewers note missing automation depth and no offline mode.

The disqualifier for regulated buyers is concrete: no SOC 2 or ISO certification is disclosed on official pages (SAML SSO and BYOK appear only at Enterprise), so Taskade cannot clear a compliance floor most enterprise procurement now requires.

17. IBM watsonx Orchestrate, best for IBM-ecosystem enterprises

IBM watsonx Orchestrate fits larger organizations already invested in the IBM ecosystem: 700+ integrations spanning CRM, ERP, and native AWS services (S3, Lambda, API Gateway, IAM, CloudWatch), a no-code builder alongside a pro-code Agent Development Kit, and a prebuilt Agent Catalog for HR, sales, procurement, and customer service. It rates 4.4/5 across 387 reviews per its own listing.

• Pick it if you are IBM-anchored and can absorb its $6,360/year entry point.
• Skip it otherwise: reviewers report a steep ecosystem learning curve, pricing seen as expensive at scale, and vague support.

The pricing quirk to know before you compare: the price is public only through the AWS Marketplace listing ($6,360/year Essentials), not IBM's own pages, which returned HTTP 403 during research.

18. Flowise AI, best for open-source, self-hosting developers (Workday-owned)

Flowise is an open-source visual builder for AI agents and LLM workflows, structured in three tiers (Assistant, Chatflow, Agentflow) from simple chatbots through multi-agent orchestration and RAG, with 100+ LLMs and vector databases, MCP client and server nodes, and on-premise or air-gapped deployment for data-sensitive teams.

• Pick it if you want open-source, self-hosted flexibility.
• Skip it if you rely on independent validation: there is no G2 listing and zero Capterra reviews, so ease-of-use and support signal is effectively absent, and no explicit SOC 2 or ISO certification was found.

It was acquired by Workday on August 14, 2025 (42,000+ GitHub stars at the time), which signals viability but ties the roadmap to a parent company, so plan for direction set outside the open-source community.

19. CrewAI, best for developer-built multi-agent workflows

CrewAI is a deep multi-agent framework (Agents, Crews, Flows) with memory, guardrails, and state-persistent long-running workflows, plus enterprise connectors for Gmail, Slack, Salesforce, Outlook, Teams, OneDrive, and HubSpot, and Amazon Bedrock Agent integration. It is backed by $18M in funding (Insight Partners-led Series A) and claims 60% of the Fortune 500 as users.

• Pick it if you have Python engineers building custom multi-agent systems.
• Skip it if you need transparent SaaS pricing or proven support: the only paid tier is Enterprise with fully custom, undisclosed pricing and no mid-tier, and independent review coverage is extremely thin at 3 G2 reviews.

This is a code-first commitment: the framework requires Python (3.10 to 3.13), so the no-code path other tools here offer simply does not exist.

20. Manus, a broadly capable general agent under ownership uncertainty

Manus is a broadly capable general-purpose agent spanning research, browser automation, app/site/slide generation, and native connectors (Gmail, Slack, Notion, GitHub, Stripe, HubSpot), with a strong compliance posture (SOC 2 Type I and II, ISO 27001:2022, ISO 27701:2019).

• Pick it as a capable individual tool if you want one general agent.
• Skip it as an enterprise standard: the reliability signal is weak (53 Trustpilot reviews, among the lowest net sentiment here), users report credits consumed or deleted unexpectedly (one cited $2,380 lost), the agent making unauthorized changes, and unresponsive support.

Ownership is genuinely unsettled: Meta's announced acquisition (December 29, 2025) was ordered unwound by China's regulators on April 27, 2026, leaving the product operating but its roadmap in limbo.

21. Bardeen AI, best for GTM and sales scraping

Bardeen is a focused browser-automation and web-scraping tool for go-to-market (GTM) teams, with four purpose-built capabilities (Web Scraper, Web Search, Enrichment, AI Qualification), roughly 100 integrations across major CRMs and outreach platforms, and $22M in funding including strategic investment from Dropbox Ventures and HubSpot Ventures. Reviewers value the roughly 10-minute no-code setup.

• Pick it if you are a solopreneur or small revenue team needing scraping and enrichment.
• Skip it if you need reliability or a free tier: there is no permanent free plan, reviewers flag credit-based pricing as confusing and expensive, and workflows that break without clear error feedback.

One maintenance flag colors the sentiment data: the G2 profile appears unmaintained, so lean on the funding and integration facts rather than review momentum when judging staying power.

22. BuildShip, best for custom backend automation

BuildShip is an AI-assisted visual backend builder: a natural-language "AI Builder" generates flow logic, with 100+ integrations across major AI model providers (OpenAI, Anthropic, Gemini, Groq, Cohere, and more), databases, and messaging, plus cloud or self-hosted deployment and JavaScript export.

• Pick it if you are a technical team needing custom backend automation with export flexibility.
• Skip it if you need stability or want to gauge sentiment first: G2 reviewers cite a poorly executed v2 migration that caused bugs and downtime, support that goes unanswered once problems persist, no permanent free plan, and an extremely thin review base (3.9/5 from just 4 G2 reviews, 0 on Capterra).

The migration history is the caution that matters most: a forced v2 cutover that broke workflows, on a near-zero review base, means you are buying with little independent signal that the next change goes smoothly.

Your procurement-readiness checklist

Before you book a single demo, answer these five questions. They convert the anxieties above into a filter that shortens your longlist fast.

1. Do you need a public price? If procurement requires an apples-to-apples comparison before any sales call, cross off the quote-only platforms now: Workato, Tray.ai, Automation Anywhere's paid tiers, UiPath Standard/Enterprise, Relevance AI's paid tiers, and CrewAI's Enterprise tier.
2. Are you anchored to an ecosystem? If you standardize on Salesforce, Agentforce is the native path; on Microsoft 365, Copilot Studio; on IBM, watsonx Orchestrate. Each delivers its deepest value inside that ecosystem and is described as restrictive outside it, so weigh the lock-in deliberately.
3. What compliance certs are mandatory? For a hard regulated floor, the deepest stacks are Automation Anywhere (ISO 42001, HITRUST, SOC 1/2/3, HIPAA) and Workato (SOC 1/2/3, ISO 27001/27701/42001, PCI DSS L1, HIPAA). At the other end, Taskade, Flowise, and IBM (page unverified) disclose no certification you can currently cite.
4. What is your task or seat volume forecast? Model it before you commit. Metering unit determines your exposure: task (Zapier), credit (Make, Gumloop, Taskade, Bardeen, Manus), conversation/action (Agentforce), Copilot Credit (Copilot Studio), CPU-second (Latenode), or annual Resource Unit (IBM). This is where "unpredictable cost at scale" comes from.
5. Is the vendor independent and viable? Flowise is Workday-owned and Pipedream has agreed to be acquired by Workday (it operates independently until the deal closes), Make sits inside Celonis, Manus's ownership was left unsettled after regulators unwound Meta's deal, and several newcomers (Relay, Gumloop, Latenode, Bardeen, BuildShip) have thin review bases. If you need confidence the platform stays supported and independent for two-plus years, favor the established public vendors.

Frequently asked questions

Which tools fit best if you already run Salesforce or Microsoft 365?

For Salesforce, Agentforce is the native path; for Microsoft 365, Copilot Studio. The specifics sit in their profiles above and in checklist item 2: each delivers its deepest value inside its home ecosystem, so weigh the lock-in if you also run systems outside it.

Do you need engineers to build automations, or can non-technical staff do it?

It splits by builder skill:

• No-code, business-user friendly: Zapier, Relay.app, Lindy, Gumloop, Taskade, and Bardeen all lead with no-code or natural-language building.
• Engineering-oriented: n8n, Pipedream, Tray.ai's Headless mode, CrewAI (Python), and BuildShip reward technical teams and carry a steeper curve for everyone else.

Make sits in between: visual, but with a real learning curve for arrays and regex.