Arcade agent tooling: how businesses use it with AI assistants
What Arcade is, what a business uses it for, and how MILabs sets it up for AI assistants that need access to business systems.
Updated 2026-09-12
What Arcade is
Arcade is an agent tooling platform from the company of the same name. It gives AI agents a managed way to call outside services such as email, calendars and business software, so an agent can carry out a task rather than only describe one. The platform handles the authorisation step for supported services, keeps track of what each agent can reach, and exposes those actions as tools the agent can call. A business reaches for it when an assistant needs to do more than answer questions, such as looking up a record, creating a draft or updating a booking. Arcade is developer-facing, so the people who use it day to day are usually builders working with a model from a provider such as Anthropic or OpenAI. It sits between the model and the systems the business already runs, and keeps the access rules in one place.
Arcade in plain language
A language model on its own can only produce text. It can describe how to reschedule an appointment, but it has no way to move that appointment unless something links it to the calendar. Linking a model to outside services is the work that agent tooling covers. Each service brings its own sign-in method, permission model and rules about which fields may be read or changed, so a business that builds these links by hand must keep a separate set of credentials for every system.
Arcade places a layer between the model and those services. It turns each permitted step into an action the model can invoke, and it carries out the sign-in for the connectors it supports. Because the allowed actions are defined in one place, a team can read them, narrow them when a task changes and see which connections are live. Whether an action is exposed, and how it behaves, can depend on the connector and the plan, which is why the options are worked out together during setup.
What businesses use it for
- Letting a support assistant look up an order status in the business's own system before it replies to a customer.
- Drafting follow-up emails from a calendar entry or a quote record, ready for a team member to send.
- Opening a task or ticket when a client message needs a decision from a person.
- Checking availability in a booking calendar before offering a time to a customer.
- Updating a contact record once a sales call has been logged in the CRM.
Who it suits
- Operations leads who want an assistant to act on approved tasks rather than only suggest them.
- Small teams already using a CRM or accounting package who need a controlled link between those tools and an AI assistant.
- Developers at an agency or in-house team building agent workflows that need managed access to third-party services.
How MILabs sets it up
MILabs sets up Arcade for a business by starting with the owner or team lead. Together we choose the plan that fits the number of agents and services the business needs. Next we connect the systems the business already uses, such as its email, calendar, CRM or accounting software, and test each connection with a sample record. Permissions are set so each agent can read or write what its task requires, and access tied to a person is agreed with that person. We then settle who looks after the setup day to day, who approves changes to the list of allowed actions, and who receives alerts when a connection needs attention. The handover is a short written note that names those people and shows how the owner can open or close access.
Builds it fits
Industries it fits
Arcade in practice
Inside a business, the work usually starts with a request that arrives somewhere ordinary, such as a shared inbox, a web form or a chat window. The assistant reads the request, works out which step is needed, and calls the one tool that performs it. That tool runs under the access granted to the agent, so the same request might produce a lookup for one agent and an update for another, depending on what each was permitted to do. The result goes back to the assistant, which phrases a reply or hands a summary to a person.
A useful way to picture it is as a set of small, separate permissions rather than one broad login. A bookkeeping team, for example, might let an agent read invoice status across the ledger, while allowing changes only to a draft queue that a staff member reviews before anything is posted. An IT services team might give an agent read access to a ticket system and leave closing tickets to a person. Each boundary is written down, so when the rules change, the team edits one list rather than searching through prompts.
Frequently asked questions
- Is Arcade safe to use with customer data?
- Yes, when the setup gives each agent only the permissions its task needs and keeps test and live use apart. MILabs agrees those choices with the owner during setup, and sends only the personal details each step requires. The owner can review which records each agent touches, and can change them as the business grows, so customer data stays inside the tasks that need it.
- What are the limits of Arcade?
- The main limits are the services it supports, any usage limits set for the plan, and the reliability of the connections it depends on. Each third-party system has its own access rules, and an agent can act only where those rules allow. Some actions should wait for a person's approval, and the design can say which ones. Choosing the right plan and the right services keeps the assistant matched to the work.
- Who looks after Arcade after setup?
- The business does, and the handover note names the people responsible. One person watches alerts from failed connections, approves changes to the list of allowed actions, and removes access when a team member leaves. A second person is named as cover. If an agent starts behaving unexpectedly, the owner can pause that connection while the cause is looked into. A written list of responsibilities means every link has a clear owner.
- What should I check on the vendor's site before using Arcade?
- The useful reference points are the services Arcade supports, how user authorisation works for each one, and what data the platform stores or logs. MILabs covers those points during setup, so the owner can focus on which tasks the assistant should handle. The plan details and usage limits are matched to the number of agents and services the business needs. Starting from a clear task list keeps the choice simple.
- How do I know whether Arcade is paying off?
- Choose the measure before setup. For a support or booking assistant, that might be the time staff spend on a repeated task, the number of requests that reach a person, or the count of records that needed correcting. Record a baseline over a normal period, then compare the same measure once the assistant is running. The clearest proof is a saving or improvement visible in the business's own records.
