Another week, another mix of building things, testing things, and tidying up the bits that make automation actually reliable. Here’s what I’ve been up to.
Putting an AI travel assistant through its paces
I spent a good chunk of the week stress-testing an AI-powered trip finder for a travel client (SkiGB ski website). Rather than just describing what it can do, I ran it live with real, tricky briefs — a family of five with a long wishlist and a tight budget, and separately a solo snowboarder watching every penny.
The interesting bit wasn’t that it found matches. It was watching it push back honestly when nothing quite fitted, suggest sensible trade-offs, and then land on genuinely good options once priorities were clear. That’s the difference between a flashy demo and something that actually earns trust with customers.
If you run a business where customers have to choose between lots of options (property, holidays, finance, insurance — anything with trade-offs), there’s a lesson here: a tool that’s honest about “here’s what won’t work and why” often builds more confidence than one that just says yes to everything.
Growing the local news site’s business coverage
Cheltenham Times — my ongoing project building a local, AI-assisted news and information site — got a proper business section this week. I put together the framework for local business profiles and interviews, a monthly “state of the high street” style feature, and a tracker that watches for early signs of businesses moving in or out of town (new company filings, planning applications, footfall data).
None of this is glamorous, but it’s the kind of steady, useful local content that’s genuinely hard to keep on top of by hand. It’s a good example of how a bit of automation can quietly keep a resource up to date without someone having to remember to do it every week.
On the sport side, results from the Prescott British Hill Climb Championship went live, alongside a full route and road closures guide for the Cheltenham Half Marathon. Small, steady additions like these are what turn a site from “a few articles” into somewhere people actually check regularly.
A big week for my WordPress plugins
A large chunk of this week went into improving and releasing the WordPress plugins I build and maintain for clients and other site owners:
Message Manager — fixed a bug where stock-level messages were still showing inside orders after items had actually sold out, added the ability to reset the stock counter when editing a message, added a new checkbox option, and tidied the settings so users only see the location options that are relevant to them.
JG Secure — fixed a bug where address and postcode verification was checking an entire company account as one, rather than checking each individual user separately. That’s now working correctly.
Website Analytics — optimised the plugin to meet WordPress.org’s guidelines and uploaded the latest version to the official WordPress plugin repository, so other site owners can now find and install it directly from within WordPress.
JG Sync — fixed the syncing of failed orders so they no longer get missed, finished off category syncing, and added the ability to sync theme and plugin details from one site to another.
JG Search & Search Lite — kicked off a brand new search plugin project, with the first milestone confirmed: blog posts are now correctly included in search results.
Fixed a checkout fault for a client this week, and got to the bottom of a WordPress publishing issue that was stopping posts from saving or going live on another site. I’m also continuing work on new checkout functionality for another client, which is progressing well.
What’s new in WI1, the AI CRM behind all of this
Most of my day-to-day work — client tasks, plugin projects, this very blog post — runs through WI1, the AI-powered CRM I’ve been building. A couple of useful features landed there this week: email signatures are now available, and there’s a new option to whitelist links that were being wrongly flagged as spam. Work has also started on security guardrails to stop unauthorised changes being made to the platform.
One nice bit of progress: more free AI models are now downloaded and running locally on one of the servers, which means more jobs can be handled without relying on a paid AI service at all.
Coming soon on WI1: posting directly to LinkedIn and Facebook, letting each person set their own working style and preferences for how the AI helps them, using pre-built “skills” straight from within a task, a page to see AI token usage at a glance, and a proper way to let customers know when something new has shipped.
Smoothing out the automation behind the scenes
I also spent time on the less exciting but important stuff: making sure a batch of scheduled jobs (things like publishing queues, price comparisons, and content pipelines) were actually running reliably rather than silently failing or getting stuck. A few had built up backlogs or were tripping over edge cases nobody had hit before.
This is the unglamorous reality of running automation for a business — it’s not “set and forget”, it’s “set, then keep half an eye on it and fix the small things before they become big things”. If you’ve ever had a spreadsheet macro or an automated report quietly stop working without anyone noticing for weeks, you’ll know exactly why this matters.
On a personal note
Also kept an eye on the household fuel and running costs tracking I do for fun — nothing exciting to report there, just steady background upkeep.
If any of this sounds familiar — a process in your business that’s held together with good intentions rather than a proper system, or a tool you wish was a bit more honest with your customers — get in touch and I’ll happily have a chat about what a sensible first step might look like.
