Skip to content

AI Agent Builders for Non-Developers: Gumloop, Relevance AI, Dust and Lindy

Four platforms let a business team build AI agents without writing code. Here is what each is for, what it costs in effort and attention, and the catch we found in each.

9 min read

AI Agent Builders for Non-Developers: Gumloop, Relevance AI, Dust and Lindy

Plenty of teams now want an AI agent that does a job on its own. Prepare the notes before a sales call. Sort the support inbox. Chase overdue invoices. Pull a weekly report together from three tools.

A fixed automation can do some of that, and we cover those tools in our guide to Zapier, Make and n8n. An agent is different. You describe the job, it decides the steps, uses the tools you connect, and asks you before it does anything you have told it to check.

Four platforms are built so that someone without a developer background can set one up: Gumloop, Relevance AI, Dust and Lindy. They overlap, but they are not interchangeable. This guide draws on our own reviews of the four, each vendor's current site, and a UK government security note on the main risk, with the sources linked at the end.

Quick answer: Lindy if you are a small team that wants one assistant working inside Slack, email and your calendar. Dust if the job is answering questions and drafting from your own company documents, with admin controls. Gumloop if you want to build agents and workflows across your tools on a visual canvas and can spend time learning it. Relevance AI if you are building a whole team of agents for sales, support or operations and are ready to talk to the vendor about pricing. Whichever you pick, start with one narrow job and keep a person approving anything that leaves the building.

What an agent builder is, and what it is not

A builder gives you four things: a way to describe the job, a set of connections to your tools, a choice of AI models, and controls over what the agent may do without asking. You are not writing code. You are writing instructions and deciding permissions.

It is not magic. An agent is only as good as the instructions and the data it is given, and it can be wrong with great confidence. The sensible way to use one is the way you would treat a new junior hire: a clear job, access to only what the job needs, and a check on early work before you trust it.

If the job is "when X happens, do Y" every single time, you probably do not need an agent. A plain automation is cheaper, more predictable and easier to debug. Gumloop's own page says as much about jobs that do not need an agent, and our review of Gumloop makes the same point.

The four, one by one

Lindy: an assistant that lives where you already work

Lindy describes itself as an AI teammate. The vendor's examples are the small-team jobs: a morning brief, replies to new leads, follow-ups on overdue invoices, a draft client report each Friday. It works through Slack, email and messages, and it connects to a long list of other tools.

Our review says it is good for an assistant that handles meetings, inboxes and routines inside Slack, email and messages. Skip it if you want a free plan to run it on, because there is only a short trial.

The catch, from our review: you pay per seat plus credits, anyone who first uses it in Slack can create a billable seat, and unused monthly credits expire. Before you roll it out, decide who is allowed to start using it. The vendor says actions that send or write wait for your approval by default, which is the setting to leave on while you learn what it does.

For a worked example of Lindy next to two other tools, see our guide to AI sales and CRM assistants.

Dust: agents grounded in your company's own material

Dust is built around your internal knowledge. It connects to tools such as Slack, Notion, Google Drive and GitHub, lets you pick between models from several providers, and lets you share the agents you build across a team. The vendor lists SOC 2 Type II certification, a choice of US or EU data residency, single sign-on and audit logs.

Our review says it is good for teams that want AI agents working on their own company data and tools, with admin controls. Skip it if you are one person who only needs a chat assistant.

The catch, from our review: its pricing has changed over time and published sources disagree, so confirm the current plan on its site before you budget. That is worth doing before anything else, because the plan decides how many credits each person gets.

Gumloop: a canvas for building agents and workflows

Gumloop is the most hands-on of the four. You describe work in plain language or build it on a drag-and-drop canvas, connect your apps, and the agents run on triggers. The vendor lists typical jobs such as lead enrichment, CRM updates from calls, support ticket triage, SEO research, competitor monitoring and scheduled reports. IT can control access, and the enterprise tier adds single sign-on and audit logging.

Our review says it is good for building AI agents and workflows across your tools on a drag-and-drop canvas. Skip it if you want a permanent free plan, because recent sources describe only a trial.

The catch, from our review: credits are billed with a service fee on top, and work that does not need an agent is costly to run through one, so keep simple jobs in a simpler tool. In practice, expect to spend your first week learning the canvas and the credit model, not shipping ten agents.

Relevance AI: a team of agents, sold as a project

Relevance AI calls its product an AI workforce. It offers prebuilt agents for sales tasks such as account research, pre-meeting briefs, post-call follow-ups and meeting booking, and aims at sales, customer success, marketing, support, operations and HR teams. The vendor lists SOC 2 Type II, role-based access, human approval steps and a policy of not training on customer data. It is an Australian company, which suits some businesses on its own.

Our review says it is good for building a team of AI agents for sales, support and operations. Skip it if you want one flat monthly price whatever your volume.

The catch, from our review: usage is metered two ways, actions and model costs, so the bill grows with how much your agents do, and its plan lineup has changed, so check what is on offer now. When we last read its pricing page it pointed to talking to sales rather than listing self-serve plans, so plan for a conversation.

Choosing between them

Your situation Start with
A team of two to ten that lives in Slack and email Lindy
Staff keep asking questions the company documents already answer Dust
Someone enjoys building and wants one place for agents and workflows Gumloop
A sales or support function with several jobs to hand to agents Relevance AI
One fixed "when X, then Y" job A plain automation tool, not any of these

Put the four through the same test before you commit. Pick one real job, write down what a good result looks like, run it for two weeks on a trial, and count how many times a person had to fix the output. That number tells you more than any feature list.

The risk to understand before you connect anything

An agent that reads email, web pages or documents can be tricked by what it reads. This is called prompt injection: text hidden in content that tells the model to do something you did not ask for. The UK National Cyber Security Centre says language models do not enforce a boundary between instructions and data, that the risk cannot be fully removed, and that you should be wary of any product claiming to stop it.

Its advice translates into three habits for a small business:

  1. Give the agent the least access that does the job. An agent that drafts replies does not need permission to send them or to read the whole shared drive.
  2. Keep approval on for anything that sends, pays or deletes. All four platforms say they have approval or human-in-the-loop controls. Check that the setting is on for the jobs that matter.
  3. Keep a record. Know what the agent did and when. Audit logs tend to sit on higher plans, so check that before you rely on one.

None of this is legal advice. If the agent will handle personal information about customers, privacy rules differ by country and state, so check with the regulator, such as the Office of the Australian Information Commissioner, before you connect it to customer data.

When to bring in help

Agent builders are built for non-developers, and a good first agent is a few hours of work. The point where an agency earns its fee is not the first agent. It is the fifth, when agents share data, touch customer records, or need testing and monitoring. If that is where you are heading, browse AI agencies and read how to brief one so the quote means something.

Where to start

Sources

Share this article

All articles