A lot of business owners are still being shown AI as a clever add-on: a chatbot on the website, a button inside software, or a tool that writes a quick first draft.
Those things can be useful. But they are not the most interesting change.
The bigger shift is that AI is starting to behave less like a separate tool and more like business infrastructure. Not magic. Not a replacement for people. Infrastructure: something that sits inside a workflow, watches for the next job, helps move work along, and reports back when it needs judgement.
That is the lesson I keep coming back to from two of my own experiments: Cheltenham Times, the local AI-assisted news site, and an AI-first CRM project built around agents rather than just records and reminders.
Both started as experiments. Both taught me that the value of AI appears when it is given a real business job, not when it is asked to perform a party trick.
The Cheltenham Times lesson: give the agent a useful job
Cheltenham Times began as an experiment on a spare PC. The task was simple to describe, but not simple to run well: build a useful local news and information website for Cheltenham.
That immediately made the AI agent do more than write paragraphs. It had to find local leads, check whether something was actually relevant to Cheltenham, decide whether a story was useful, create or suggest WordPress content, think about categories, and support useful local pages such as fuel prices, planning applications, property information, school catchment areas, clubs and events.
That matters for SMEs because it changes the question.
The question is not, “Can AI write content?”
The better question is, “What useful job could AI do repeatedly in this business?”
For a professional services firm, that job might be reviewing enquiries, preparing client briefing notes, spotting missing documents, or keeping internal knowledge up to date. For a trades business, it might be triaging incoming jobs, drafting responses and preparing daily schedules.
The useful bit is not the AI itself. The useful bit is the work that gets done.
The CRM lesson: AI should sit inside the workflow
The AI-first CRM experiment came from a different problem. Most CRMs are built around storing information. You add contacts, projects, tasks, notes and emails. Then the business still has to remember what to do next.
I wanted to test a different idea: what if the CRM was not just a database, but a control centre for work?
In the case study, I gave Anthropic’s Fable 5 a large brief: help build a CRM where AI is part of the centre of the product, not bolted on afterwards. The brief included projects, tasks, email handling, role-based views, AI assistants for staff, and the ability to connect to local machines running tools such as Hermes and OpenClaw.
The interesting part was not that it produced screens. The interesting part was the direction of travel. The agents could take action, work through tasks, split work down, improve workflows, and help in a way that felt closer to a junior colleague than a normal chatbot.
That is what many existing business systems are missing. They hold data, but they do not do much with it. AI changes that. A CRM, helpdesk, quoting system or operations dashboard can start to notice, suggest, prepare and act within clear limits.
That is when AI becomes infrastructure.
Experiments only become systems when you add guardrails
There is a catch, and it is an important one.
The first version of an AI workflow is rarely the final version. Cheltenham Times made that clear. Some content needed steering. Some images were wrong. Some stories needed better routing because Cheltenham, Gloucester and wider Gloucestershire are not the same thing. The answer was not to abandon the system. The answer was to improve the rules.
That is where business owners need to be realistic.
An AI agent should not be treated as an all-knowing machine. It should be treated as a fast assistant inside a managed process. You still need:
- clear instructions about what good looks like
- data and source rules
- privacy and access controls
- human approval points for sensitive work
- cost limits
- a way to measure whether the workflow is actually helping
Without those, AI can create noise very quickly. With them, it can become genuinely useful.
What this means for a Gloucestershire SME
You do not need to start with a huge “AI transformation” project. In fact, most businesses should not.
A better starting point is to find one recurring job that is valuable, slightly messy, and currently takes up too much time. Not the most dangerous job in the business. Not the most complex. Something practical.
For example:
- preparing a daily list of priority enquiries
- turning meeting notes into follow-up tasks
- checking public information sources for opportunities or risks
- drafting client updates from existing project records
- creating a first version of reports, summaries or proposals
- keeping internal knowledge articles current
Then you design the workflow around that job. What information can the AI see? What can it do automatically? What must a person approve? What would count as success after 30 days?
The real shift: from clever demo to working habit
The strongest lesson from these experiments is that AI becomes useful when it disappears into the working day.
Cheltenham Times is not just “an AI wrote some articles”. It is a set of workflows, sources, checks, pages, tools and human decisions around a local information service.
The CRM experiment is not just “AI built some software”. It points towards business systems that do not simply store work, but help move it forward.
That is where I think practical AI is heading for SMEs: smaller, clearer, managed systems that do useful work every day.
Not hype. Not replacing the business owner. Not removing judgement.
Just better infrastructure for the jobs that keep repeating.
If you run a business in Cheltenham or Gloucestershire and you are wondering what AI could sensibly do inside your own workflow, I can help you find the right first project, design the guardrails, and test whether it is worth building further.
Book a short AI consultancy call or send me a message with the question: “What could AI do in my business every day?”
