AI · Service

AI agent development services for the systems you already run

An agent is only useful if it can act in your systems and only safe if it is limited, logged and owned. We build agents that do real work in the software you already have, with a person in the loop where it matters.

Short answer

AI agent development is the design, build and operation of software that uses an AI model to carry out multi-step tasks in business systems. A sound project picks a bounded task, connects the agent to the systems through scoped access, tests it on real cases, and launches it with logging, limits and a human approval step. We build on the major model platforms and connect to your systems through their own APIs.

  • Anthropic
  • OpenAI
  • Microsoft 365
  • Salesforce
  • HubSpot
  • ServiceNow
  • Slack
  • NetSuite

Last reviewed . General guidance, not legal, tax or accounting advice. Consultants in a Box is a Make partner and independent of the other vendors named.

Steps in an agent build, by phase

Counted from the table on this page.

Show as table
ItemSteps in an agent build, by phase
Define3
Build3
Prove2
Run2

Where the numbers come from

These counts come straight from the lists on this page. Nothing here is a survey result or a client outcome.

What kinds of agents are worth building first?

Ones that handle frequent, rule-bound work where a person already checks the result: triage, research, drafting and data entry between systems.

  • Sales: account research, call notes into the CRM, follow-up drafts.
  • Service: ticket routing and reply drafts from the knowledge base.
  • Finance: matching and coding with exceptions sent to a person.
  • Operations: daily summaries and schedule preparation.

How do you keep an agent safe?

Treat it like a new employee with system access: its own account, the minimum permissions, a record of what it did and a manager.

  • Scoped credentials per system, never a shared admin login.
  • A log of every action, readable by the business owner.
  • Hard limits on what it may send, spend or change without approval.
  • A pause switch more than one person knows how to use.

Anthropic's guidance on agents and the NIST AI Risk Management Framework point the same way: start with the simplest design that works, and add autonomy only when the record supports it.

Do you build on one model or several?

Whichever fits the task and your existing contracts. The design that matters is around the model: the tools it can call, the data it can reach and the checks on its output.

  • We connect through each system's own API and through Model Context Protocol connectors where they exist.
  • Your data stays under the terms of your own vendor agreements.
  • The agent is documented so your team, or another firm, can maintain it.
FAQ

Questions buyers ask about this

How long does an AI agent take to build?

It depends on the task and on how ready the systems and data are. A bounded agent on systems with good APIs is a matter of weeks, not months; we scope it on a call after looking at the systems.

Do we need a developer on our side?

No. You need a business owner for the task and someone who can grant system access. We do the build and document it.

Who owns the agent and the code?

You do. It runs on your accounts with your vendors, and we hand over the instructions, tools and tests.

Can an agent work across several systems?

Yes, that is where agents earn their keep. Each connection gets its own scoped access and its own entry in the log.

Vendor documentation to check against

Product details, plans and pricing change. Confirm current details in each vendor's own documentation. Consultants in a Box is a Make partner and otherwise an independent consultancy; this page is not published or endorsed by any vendor named. Names and logos are trademarks of their owners.

Want to talk this project through?

Book a call and we will look at your systems, what the work involves and what it should cost. Or start with the free business analysis.

Book a free consultation

Tell us about your team and the software you need help with. We'll follow up to schedule a time to talk.

Open the form on its own page

Talk to a person

The fastest way is to pick a time. Prefer to write? Leave a message and we will email you back, usually the same business day.

Book a free call

By sending this you agree to our Terms and Privacy Policy.