The Agent Directory and Deals

Build, buy or bring in an implementer: three ways to get an AI agent

You can build an agent yourself, buy one ready-made or pay a specialist to set one up. Each route suits a different business. How to choose.

The Infercline Desk · Friday, 6 November 2026 Sign in to save

Once you have decided a job suits an AI agent, the next question is how to get one. There are three broad routes. You can buy an agent that comes ready-made, usually inside software you already use. You can build your own, using the tools that model makers and cloud providers offer. Or you can bring in an implementer, a consultancy or specialist firm that designs and sets up the agent for you. Many businesses end up using more than one.

Buying is the fastest route and usually the right first step. Many business software vendors now include agents: help desk tools such as Intercom and Zendesk include customer service agents, accounting packages such as Xero and QuickBooks are adding agents for bookkeeping and payments, and Salesforce and Microsoft offer agent tools inside their wider platforms. A bought agent already understands the software it sits in and can see your data there. You pay a subscription or a usage fee, and the vendor handles updates and security.

The drawback of buying is fit. A ready-made agent does the job the vendor designed, in the way the vendor designed it. If your process is unusual, or the job crosses several systems from different vendors, a bought agent may cover only part of it. You also depend on the vendor's roadmap and pricing, which can change.

Building gives you the most control. Model makers and cloud providers offer tools for this. OpenAI offers an Agents SDK, a set of building blocks for developers, which its AgentKit page recommends for workflows that should run as code. Amazon describes Bedrock AgentCore as a platform for running agents in production, with tools for securing access and evaluating behaviour. Integration platforms such as Zapier offer a middle path, letting less technical staff build agents that connect thousands of apps without writing code.

Building is not cheap, even when the tools are. You need people who understand both the technology and your process, time to test the agent on real cases and someone to maintain it as models, prices and your business change. Anthropic, in its guide to building effective agents, advises starting with the simplest solution possible and only adding complexity when it clearly improves results. That advice applies doubly to a business building its first agent.

Building also carries platform risk. In June 2026 OpenAI added a note to its AgentKit launch page saying it was winding down its Agent Builder and Evals products, which would no longer be available from 30 November 2026. Businesses that had built on those tools had to move. Whatever you build on, keep your instructions, test cases and data in a form you can take elsewhere.

Bringing in an implementer suits businesses that want something tailored but do not have the skills in-house. A good implementer maps your process, chooses the tools, builds and tests the agent, trains your staff and hands it over with documentation. Implementers range from large consultancies to small specialist firms, and many partner with platforms such as Salesforce and Microsoft. The risks are cost, dependence on the implementer for changes, and firms that are new to the field. Gartner warned in June 2025 about agent washing among vendors, and the same caution applies to service firms.

When you hire an implementer, agree on what you will own at the end: the agent's instructions, the code if there is any, the test cases, the logs and the accounts with the underlying platforms. Ask for references from clients with a similar process, and ask to speak to them. Agree a fixed scope for a first agent before any larger programme, and make sure the contract covers data protection in each country where you operate.

A simple rule helps choose. If a product you already use offers an agent for the job, try it first. If the job crosses several systems or is central to how you compete, consider building or bringing in an implementer. If you lack the skills to build and maintain an agent, an implementer is usually cheaper than learning by trial and error on live work.

Why it matters to your business

The route you choose decides your cost, your speed and how much control you keep. Buying is quick but may not fit. Building fits but needs skills and carries platform risk. An implementer brings skills but adds cost and dependence. Choosing deliberately, and keeping ownership of the parts that matter, avoids being locked into an agent you cannot change or afford.

What to do on Monday

Check whether any software you already pay for offers an agent for the job you have in mind, and try it before anything else.
If you consider building, name the person who will maintain the agent after launch. If there is no one, do not build yet.
If you consider an implementer, ask three firms for references from clients with a similar process, and call them.
Whatever the route, write into the agreement that you own the instructions, test cases and logs, and can export them.

Build, buy or bring in an implementer: three ways to get an AI agent

Why it matters to your business

What to do on Monday

Sources