What is an AI agency, and how does it differ from an AI consultancy?

An AI agency is a firm that builds AI-powered tools for other businesses. The work is usually a defined piece: a customer-service chatbot, a content or marketing tool, a lead-qualification assistant, or an AI agent that carries out a task such as answering enquiries or updating records. Many agencies work from templates and components they have built before, which is why they can move quickly.

The name causes confusion. An AI agent is software, not a supplier. The UK government's AI Playbook describes agentic AI as autonomous systems that can make decisions and perform actions with minimal human intervention [1]. An AI agency may build agents, but it is a company you hire, and the question on this page is which kind of company that should be.

An AI automation agency is a kind of AI agency that focuses on automation. It connects the software you already use, such as your inbox, CRM, accounts package and spreadsheets, and adds AI steps in between: reading an email, classifying it, drafting a reply, updating a record. The work is usually built on a no-code workflow platform rather than written as software.

The commercial model often has two parts: a fee to build each automation, then a monthly fee to host, monitor and fix it. That suits small, contained jobs well. The catch is that the workflows often run in the agency's own platform account, so what you are paying for each month includes the right to keep using them.

An AI consultancy starts from the business problem rather than the tool. It works out where AI is worth using, checks whether your data and systems can support it, settles what to build and what to buy, and designs the controls that let risk, legal and IT approve it.

Consultancies split into two kinds. Some advise only: the output is a strategy or roadmap, and the build is bought from someone else. Others advise and build: the same people who designed the system write it, put it into production and hand it over for your team to run. When you compare quotes, find out which kind you are talking to.

The three differ less in the technology than in where they start, how they charge and what they leave behind. These are the typical patterns; any one supplier can differ, so ask.

The three price different things, so their quotes do not compare directly. A fee per automation and a monthly running charge price a tool you rent. A fixed-scope consultancy fee prices a defined output. Time charged by the day prices effort, and leaves the risk of overrun with you.

The National Audit Office's lessons-learned report on how government buys consultancy says contracts should focus on outcomes and outputs rather than inputs [8]. The same logic helps a private buyer. Whatever the shape, ask three questions: what exactly does each fee pay for, what does it cost to keep the system running for a year, and what happens to it if you stop paying.

This is the question that separates suppliers more than any other, and the answer is in the contract, not the pitch. Under UK law a computer program is a literary work [2], and the author of a work is the first owner of its copyright [3]. The UK Intellectual Property Office's guidance applies that to commissioned work: the creator, not the commissioner, owns the copyright unless you agree otherwise in writing [5]. An assignment of copyright is not effective unless it is in writing and signed by or on behalf of the supplier [4].

The table sets out the default position for each asset an AI build leaves behind, and what to ask for instead. It describes the law and published guidance as they stood on 29 September 2026. It is not legal advice: have your own adviser review the contract.

Controls follow the same rule. If the checks on what an automation may read, change or send exist only in the supplier's head, you do not own them. An AI governance framework written as enforceable controls, with an owner and a record for each, is something you can keep.

Start from the job, not the supplier. Four situations cover most buyers, and in the first two an agency is the better choice.

If you cannot tell which of these you are in, that is itself an answer: the problem is not yet defined well enough for an agency to build against. A short piece of discovery, or a self-assessment, is the cheaper first step.

1AYM is a consultancy that also builds. The people who write the strategy write the code, and the architecture, code and operating model are documented and handed over, so nothing depends on us staying. We sell three shapes of engagement: fixed-scope statements of work, a retained implementation engagement, and engineers embedded in your own programme. 1AYM is an OpenAI Select Partner.

We take small fixed-scope statements of work as well as larger builds. Our engineering and controls matter most in the third situation above: work that touches money, customer records, regulated decisions or several systems at once. Where the first question is where AI pays back, the usual first step is the AI Opportunity & Feasibility Sprint, a fixed-scope engagement of typically two to four weeks. You can see every service we offer, with what each one delivers, before you book a call.