About this guide

Contents

1. Who it is for
2. How it is made
3. Corrections
4. Ownership

An AI agent that drafts an email is a convenience. An agent that issues a credit note, changes a delivery date or updates a customer record is an employee with no judgement of its own. The difference lies in what it may touch and who approves it, and most rankings of agent builders do not ask. This guide does.

Who it is for

Operations, finance and customer-service leads at established companies who have been asked to "put agents to work" and are choosing a platform. They are accountable for what the agents do, often without an engineering team to build safeguards for them.

How it is made

Each agent builder is assessed from its own pricing, product, documentation and trust pages, read on the date shown, and from independent research by named publishers. All are compared on the same seven criteria, set out on the methodology page. Vendor-supplied figures are labelled as such.

Agent pricing and plan limits change frequently. We re-check every builder at least quarterly, and every page shows its last update and a change log.

Corrections

If a fact or a ranking looks wrong, contact us with the source. Verified errors are corrected and logged on the page.

Ownership

This publication is owned by the company that makes OpenLazarus, one of the platforms it reviews. OpenLazarus is ranked first for one stated situation, described in the methodology, and is assessed against the same published criteria and sources as every other entry. Its limits are listed as plainly as anyone else's.

Further reading

• Guide · ControlAI agent permissions and approvals: how to let agents act safelyAn agent's permissions are a job description written in advance. The platforms differ less in whether they offer approvals than in where those approvals apply and what record they leave.
• Guide · CapabilitiesWhat AI agents can do for a business today, and what they cannotAgents now answer mail and calls, key orders, chase invoices and prepare reports. They still fail about one task in three on hard benchmarks, which decides where they should start.
• Guide · CostWhat no-code AI agents cost: credits, actions, executions and the monthly billEvery platform meters agent work in its own unit. Converting those units into the cost of one task is the only way to compare them, and the only way to avoid the cost overruns Gartner expects to end many projects.