Zapier built the modern integration category. It made "if this, then that" logic accessible to non-technical teams and helped businesses connect hundreds of apps without writing code. In 2026, that category has been outgrown. The processes businesses need to automate today involve unstructured data, multi-step reasoning, and decisions that cannot be reduced to a trigger and an action. We published a list of 15 AI-native tools in June 2026. On 18 October we checked every one again on its own website. Eleven are still worth your time, and four are not, for reasons explained at the end.
| Tool | Best for | Free option | Paid from | Self-host |
|---|---|---|---|---|
| Gumloop | Non-technical teams building AI agents | 14-day trial only | $37 | No |
| n8n | Developers who want code inside a visual builder | Self-hosted Community Edition | €20 (cloud, annual) | Yes |
| Make | Low cost multi-step automation | 1,000 credits a month | $9 | No |
| Relevance AI | Teams of specialist sales and support agents | Not listed | Enterprise, via sales | Not stated |
| AirOps | AI search visibility and content production | 14-day trial | Not listed | Not stated |
| Power Automate | Microsoft 365 organisations | 30-day trial | $15 per user | No |
| Domo | Workflows triggered by business data | Free trial | Not listed | Not stated |
| UiPath | Legacy desktop systems with no API | Community plan | $25 | Yes |
| LangChain / LangGraph | Custom builds by engineering teams | Open source; LangSmith free for 1 seat | $39 per seat (LangSmith) | Frameworks yes |
| Retool | Internal tools generated from a description | Up to 5 users | $10 per builder (annual) | Yes |
| CrewAI | Multi-agent workflows | 50 runs a month | Enterprise, via sales | Framework yes |
1. Gumloop: AI agents for non-technical teams
Gumloop targets operations teams who want to build AI automation without a developer. In 2026 it has moved its focus from workflows to agents that work across your business tools, and its original visual workflow canvas is now labelled "Legacy". There is no free plan any more: Pro starts at $37 a month after a 14-day trial. If your team wants an agent working across its existing apps without involving engineering, it is still one of the quickest ways to try one.
2. n8n: Custom logic without the ceiling
When standard workflow tools hit their limits, developers reach for n8n. It is source-available under a "fair-code" licence, free to self-host through its Community Edition, and supports full JavaScript and Python alongside its visual builder. Cloud plans start at €20 a month, billed annually. The practical benefit is that when an AI flow needs to do something no pre-built node covers, you write the code inside the platform rather than routing around it. Self-hosting also makes it one of the better options when data has to stay on your own infrastructure.
3. Make: Multi-step architecture at lower cost
Make, formerly Integromat, has held its position by undercutting competitors on complex multi-step workflows. It now bills in credits rather than operations: the free plan includes 1,000 credits a month and paid plans start at $9. Agencies managing automation for several clients often run dozens of Make scenarios at a cost that would be prohibitive elsewhere, and its visual builder handles branching and error routing clearly enough for non-developers to maintain. If cost per workflow is your main constraint, benchmark Make first.
4. Relevance AI: An AI workforce of specialist agents
Relevance AI now positions itself as a platform for "AI workforces": teams of specialist agents for sales, customer success, marketing and HR. Public self-serve pricing has gone, and new customers are routed to an Enterprise conversation with sales. It suits mid-market and larger teams that want several agents sharing context across a function, rather than one-off automations.
5. AirOps: AI search visibility and content operations
AirOps has shifted from classic SEO to AI search visibility: helping brands appear in answers from ChatGPT, Perplexity and Google AI Overviews, with bulk content production built on top. For a brand trying to grow its presence with a small team, it automates repetitive production work without a developer wiring up each step. Pricing is not published; there is a 14-day trial. If you first want to know where your brand stands in AI answers, our free AI Visibility Score checks that before you buy any tool.
6. Microsoft Power Automate: The enterprise default
For organisations already committed to Microsoft 365, Power Automate is the path of least resistance. It covers cloud workflows and desktop (RPA) automation, with Copilot Studio alongside for agents. Premium starts at $15 per user a month and a Process licence at $150 per bot. The licensing is complex and the tool is not the most intuitive for non-technical users, but its integration with SharePoint, Teams and Dynamics is unmatched in that ecosystem.
7. Domo: Business intelligence triggering workflows
Domo combines data analytics with workflow orchestration. Instead of waiting for a user action or a schedule, Domo alerts, workflows and now AI agents can respond when a KPI crosses a threshold. If your business intelligence data should drive operational responses, Domo closes the gap between "the data says something changed" and "the team does something about it". It is enterprise-priced, with no public prices.
8. UiPath: RPA extended with AI agents
UiPath started as a Robotic Process Automation platform, automating desktop applications by mimicking clicks and keystrokes. It now calls itself "agentic automation": the same robots plus AI agents that handle unstructured data and make decisions inside those automations. There is a free Community plan and paid plans from $25 a month, and Standard and Enterprise can run on your own infrastructure. If your workflows involve systems that can only be reached through a desktop interface, UiPath handles that in a way most AI-native tools cannot.
9. LangChain and LangGraph: Building custom from the ground up
LangChain is a development framework, not a no-code platform. Together with LangGraph for stateful agents and the newer deepagents library, it is the usual starting point for engineering teams that need more control than any off-the-shelf tool provides: document retrieval, custom agent logic, human approval steps and tool integration. The frameworks are open source; LangSmith, the paid platform for tracing and evaluation, is free for one seat and $39 per seat on Plus. It takes developer time, but it has no ceiling. It is what we use for our own LangGraph agent builds.
10. Retool: Internal tools generated from a description
Retool lets teams build internal tools quickly, and AI is now at the core of the product rather than an add-on: you describe the app and Retool generates it, with Agents and Workflows alongside. It is free for up to five users, Team starts at $10 per builder a month billed annually, and it can be self-hosted. If your business needs bespoke internal tooling rather than an off-the-shelf application, Retool cuts the build time considerably.
11. CrewAI: Multiple AI agents working as a team
CrewAI is built around the multi-agent model: different AI roles, such as a researcher, a writer and a reviewer, working towards a shared goal. Its main documented modes are sequential (agents hand work along a chain) and hierarchical (a manager agent delegates and checks). For tasks that benefit from separate reasoning perspectives, that structure can produce better output than a single chain. The framework is MIT-licensed and free; the hosted platform (AMP) has a free tier of 50 runs a month and Enterprise pricing beyond that. It needs more design work than most tools here to use well.
What changed since June: four tools we removed
Our June version listed 15 tools. Four no longer belong on a list like this. Relay.app shut down in September 2026 and deleted its accounts. OpenAI is retiring Agent Builder on 30 November 2026; teams on the OpenAI stack should look at the Agents SDK instead. Vellum has pivoted to an open-source personal AI assistant rather than a business workflow tool. Cloudsquid now sells AI agents that audit ERP transactions for finance teams, and we found no self-hosted option, so it no longer fits our "regulated data" recommendation.
This is the real lesson of the list. In four months, a quarter of the tools changed enough to matter. If an automation is central to how your business runs, choose a platform you can leave, keep your logic and data portable, and avoid building on a product whose roadmap you cannot see.
Which tool fits which situation
No single platform is the right answer for every use case. The decision comes down to four factors: who is building the workflow, how complex the logic needs to be, how sensitive the data is, and whether you are automating a single task or a multi-step process that needs reasoning.
Non-technical teams with simple to moderate complexity: Make or Gumloop. Developer teams needing custom logic: n8n or LangChain and LangGraph. Microsoft environments: Power Automate. Regulated data that must stay on your own infrastructure: self-hosted n8n, or a custom LangGraph build with data residency controls designed in. Legacy systems with no API: UiPath. Multi-agent workflows: CrewAI.
The tools that will not serve you well in 2026 are the ones built purely around API triggers with no AI reasoning layer. The processes worth automating today are the ones that could not have been automated two years ago. If the hard part of your process is reading documents such as invoices, contracts or forms, see our document processing work before picking a general workflow tool.
If you are trying to work out which approach fits a specific process in your business, whether a platform or a custom build, an AI Readiness Assessment maps the process, data and integration requirements before any tool or architecture decision is made.