AI automation for small business: the kind that breaks, and the kind that adapts

There are two ways to build AI into a business, and the difference decides whether what you buy is still working in six months. One kind is a fixed script: it does exactly what it was told, in the exact order it was told, and it breaks the moment your process changes. The other kind reads the tools it has available and works out which one fits the job, so when your process changes, it adapts. Most small businesses have only ever been sold the first kind. This is how to tell them apart before you pay for either.

What actually breaks when your process changes?

Picture a simple bit of automation: when a quote gets accepted, create the job, add the client to your accounting system, and send a welcome email. If someone built that as a fixed script, they hardcoded every step. Which button, in which system, with which fields, in which order.

That works beautifully on day one. Then your bookkeeper switches you to a different accounting package, or your supplier changes their portal, or you add one new step to how you onboard a client. Now a person has to go back into the automation and rewrite it. Until they do, it either does the wrong thing or stops altogether. You are paying for something that needs a developer every time your business evolves, which for a growing small business is constantly.

The reason it breaks is not that the AI is not clever enough. It is that it was never allowed to think. It was handed a script and told to run it.

What "adapts" actually means

The other approach hands the AI something different: not a script, but a description of the tools it can use and what each one is for. The AI reads that, understands what is available, and reasons about which tool to use for the job in front of it. When the job changes, it picks differently. Nobody has to rewrite anything.

The plumbing that makes this possible is a standard called the Model Context Protocol, or MCP. You do not need to understand it, any more than you need to understand how email routing works to send an email. But it is worth knowing it exists, because it is the thing that separates an automation that adapts from one that just follows orders.

Here is the test we use at Handiwork, and it is the only sentence in this article worth writing down:

Does this break if my process changes next month, or does it adapt? If the honest answer is "it breaks," you are buying a script, not an AI setup.

Is MCP just another integration?

Fair question, because the word "integration" has been used to sell the fixed-script kind for years. MCP is genuinely different in one specific way: instead of a developer hardcoding how your AI talks to each tool, each tool describes itself in a way the AI can read and reason over. The AI discovers what it can do rather than being told, step by step, what to do.

The clearest way to think about it is the way one Google Cloud explainer put it: MCP did for AI tools what HTTP did for the web. Before HTTP, different corners of the internet spoke different protocols and did not interconnect. A single standard made them interoperable. MCP is trying to be that single standard for the tools an AI can use.

Is this a real standard, or one vendor's bet?

This is the part that matters for a small business owner deciding whether it is safe to build on. MCP is not a Claude-only idea or a passing fad. It was created by Anthropic and released as an open standard in late 2024. Through 2025 it was adopted by OpenAI, Google and Microsoft across their main AI products. In December 2025 Anthropic handed it to the Linux Foundation's Agentic AI Foundation, which is the same kind of neutral, shared home that governs a lot of the open technology the internet already runs on. In plain terms: it stopped being one company's project and became shared infrastructure.

And it is being used in earnest, not just talked about. A 2026 industry survey found around 41% of software organisations were already running MCP tools in production, not experiments. That is not everyone, everywhere, and it is honest to say adoption is still growing. But it is well past the point of being a bet. Building on it is building on something the major AI providers have all agreed to.

Why this changes what you should buy

Most small business owners have been sold "automation" that is really a fixed script wearing an AI label. It works on the first day and quietly breaks the first time reality moves. The whole reason we at Handiwork lean on the tool-discovery approach for the workflows we build is that it survives change, which is also why we can add a new tool to a client's setup in an afternoon instead of a rebuild.

That is the practical difference for you. Not the acronym, not the protocol. The difference between an automation you have to keep paying to repair, and one that keeps working as you change how you do things. If you want to go a level deeper on which jobs are worth automating in the first place, our use cases and workflow implementation service are the practical next read.

Ready to find out where you stand?

If you want to know whether your current setup would survive a process change, the free AI Readiness Check is a good place to start. It takes about five minutes and there is no pitch at the end of it.