Automation & AI Agents: Frequently Asked Questions

Straight answers to the questions I get asked most often before a project starts.

How AI Agents Actually Fit Into Business Automation

An AI agent, in the sense that matters for a small business, is not a chatbot bolted onto a website. It is a system that can look at a task, decide what steps are needed, call the tools required to complete them, and check its own work, largely without a human directing every individual step. That distinction matters because it is what separates a genuinely useful automation from a scripted flow that just happens to call an AI model at one point in the chain. For a concrete look at what one of these looks like end to end, see how to build a real n8n AI agent, tested on real messages before it went live.

Anthropic, the company behind Claude, has published one of the clearer public explanations of where this genuinely helps versus where a simpler workflow is the better choice. It is worth reading if you want the technical version rather than the marketing one: Building Effective Agents.

In practice, most of what a small business actually needs is closer to the “simpler workflow” end of that spectrum: a handful of well-defined automations, each doing one job reliably, with an AI step used only where genuine judgement is required. Reading an inbound email and deciding what it is about is a good example, rather than running everything through a single do-it-all agent. The questions below go into the specifics.

What This Means for Your Business, Practically

The honest answer to “should we be using AI agents” is usually: probably not yet, and not as the first thing to automate. Most businesses get far more value from automating the boring, well-defined, repetitive parts of their workflow first: routing a lead, sending a reminder, updating a record. Those tasks have a clearly correct answer every time, and a simple automation gets it right 100% of the time for a fraction of the cost and complexity of an AI-driven one.

Where an AI step genuinely earns its place is at the point in a workflow where a human would otherwise need to read something and make a judgement call: classifying an inbound email, summarising a long document, or deciding which of several possible next steps applies to a specific situation. Bolting an AI agent onto a task that already has one obvious correct answer usually just adds cost, latency and a new failure mode without adding any real capability. The case study on a voice agent answering calls 24/7 is a good example of an AI step used only where judgement was actually needed, not everywhere at once.

The practical starting point for most businesses is a handful of simple, reliable automations covering the repetitive 80% of the work, with one or two AI-assisted steps used narrowly, only where genuine judgement is required. That combination tends to outperform a single ambitious “AI agent that does everything” build, both on reliability and on how quickly it can actually be shipped and trusted.

A Simple Way to Decide Where AI Actually Belongs

A useful test before adding an AI step anywhere in a workflow: could a straightforward rule get this right at least 95% of the time? If yes, that rule will almost always be cheaper, faster and more predictable than an AI call, and it is worth building it that way even if an AI-based version would technically also work. AI earns its place specifically where that simple rule breaks down: where the input varies too much for a fixed set of conditions to handle reliably, and genuine judgement is actually required to get a good result.

One more practical marker: if a task can be written down as a short, finite checklist that rarely changes, it almost never needs an AI agent to handle it well. A plain automation will do it faster and more predictably. Save the AI step for the parts of the work that genuinely resist being written down as a checklist in the first place.

Related reading: for questions specific to n8n, see the n8n FAQ.