Mark's Thoughts

Bring your own network

markw@markwilson.co.uk (Mark Wilson) — Wed, 5 Aug 2026 00:00:00 +0100

Bring your own network

Most of us have heard of bring your own device. We expect our laptops, phones and tablets to work wherever we are.

But what if, instead of trusting somebody else’s network, you could simply take your own with you? That was the idea behind one of my more recent impulse purchases: the UniFi Travel Router (UTR).

It doesn’t look particularly special. It’s a small white box that slips easily into a laptop bag. It isn’t even self-powered, so you’ll need a USB-C cable and a power source. That feels like an odd omission for something designed for travel.

After using it on a few trips, though, I realised it wasn’t really a travel router. It was a way to carry my home network with me—making public Wi-Fi safer, simpler and, in some cases, almost invisible.

It’s now one of the gadgets that lives permanently in my laptop bag.

The problem with public Wi-Fi

Whether you’re in a hotel, airport lounge, coffee shop, conference venue or on a train, public Wi-Fi is often the easiest way to get online.

Modern websites encrypt almost everything using HTTPS, so this isn’t as risky as it once was. Even so, you’re still placing your trust in someone else’s network. You don’t know how it has been configured, who else is connected, or whether there’s a convincing rogue hotspot using a similar name.

I’d rather not have to think about any of that. Instead, I’d rather treat every unfamiliar network as simply a way of reaching one that I already trust.

One network, everywhere

Here’s how it works:

Instead of connecting each of your devices directly to the local Wi-Fi, you connect the travel router.
Using the UniFi app on your phone, you tell the travel router which local Wi-Fi network to join. If there’s a captive portal, such as the familiar hotel “accept our terms” page, you complete that authentication once from your phone.
The router then establishes a secure UniFi Teleport VPN connection back to your home network. From that point on, every device connected to the router behaves as though it’s sitting in your house.

Your traffic is encrypted between the travel router and your home, so, in effect, you’ve taken your home network with you.

Many guest Wi-Fi networks require you to authenticate each device individually through a captive portal. Some even limit the number of devices you can connect.

With the travel router, I authenticate once using my phone. After that, every other device simply connects to my own Wi-Fi network, exactly as it does at home.

My laptop, phone and tablet don’t know they’re connected to a guest network. As far as they’re concerned, they’re connected to the same network they always use.

To the venue, it looks like a single connected device. Behind the travel router, I can connect as many of my own devices as I like.

More than just security

I bought the travel router because I liked the idea of taking a trusted network with me. And, if I’m honest, because I’m a bit of a geek.

In practice, convenience has become the bigger benefit.

Everything behaves exactly as it does at home. My devices reconnect automatically, local services continue to work, and I don’t spend the first ten minutes in every hotel room joining Wi-Fi networks and accepting terms and conditions.

Websites should also see my home IP address rather than the IP address of the network I’m connected to. Although I haven’t travelled abroad with the UTR yet to test this, services that are normally only available in the UK should continue to work without needing a separate VPN subscription.

I’m simply using my own internet connection, wherever I happen to be.

It’s not perfect

The lack of an internal battery means you need to remember a USB-C power source. In practice, I’ve just plugged mine into the same charger that powers my phone overnight, but it would have been nice to have an hour or two of battery life for situations like trains or airport lounges.

It also assumes you already have a UniFi gateway at home capable of running Teleport. I do, but if you don’t, this probably isn’t the travel router for you.

Bringing your own network

I’ve written before that the best technology quietly disappears into the background.

The UniFi Travel Router does exactly that.

Every time I travel, it removes a little friction. I authenticate to guest Wi-Fi once, my devices reconnect automatically, and I know my traffic is heading back through a network that I already know and trust.

I can’t see myself travelling without it now.

Powerless: the invisible foundations of modern life

markw@markwilson.co.uk (Mark Wilson) — Tue, 4 Aug 2026 00:00:00 +0100

Powerless: the invisible foundations of modern life

As I write this, the electricity supply to my home is switched off.

Every few seconds, the house alarm lets out an apologetic beep. The electrician is outside, upgrading the communications module on my smart meter and installing an EV charging point. To do that, the house has to be without power for about 45 minutes.

It shouldn’t make much difference. But I feel strangely trapped.

I can’t head to a café because the electrician still needs access to the house. I can’t even take the dog for a walk in case I’m needed at home. So I stay put, waiting for the power to come back.

Disconnected

It’s surprising how quickly a home office stops being an office.

I work from home most of the time, and I’ve invested a fair amount of effort into making that work well. Reliable broadband. A solid UniFi network. A NAS. Multiple screens. Everything quietly doing its job in the background. Until someone turns the electricity off.

The Wi-Fi disappears. My NAS is unreachable. Home Assistant stops assisting. Even making a cup of coffee is off the table.

My laptop still has plenty of charge, and I can tether to my phone over 5G if I need to access the Internet. I haven’t lost the ability to work. I’ve lost the environment that makes it effortless.

Without really noticing, my mindset changes. Can I complete this task? Should I join this Teams call now or wait? And, because my mind works as it does, it’s just really hard to focus.

For just 45 minutes, my focus shifts from getting work done to managing the interruption.

Invisible foundations

The electrician isn’t here because anything has gone wrong. Quite the opposite. He’s making the house more connected. A smarter smart meter. An EV charger in preparation for when my new car arrives. More technology woven into everyday life. To get there, though, everything has to stop for a while.

It makes me wonder how much of modern life depends on things we rarely think about. Electricity is the obvious one, but it’s only part of the picture. Broadband. Mobile networks. DNS. Cloud services. The infrastructure that quietly fades into the background because it simply works.

Until it doesn’t.

Normal service resumes

Forty-five minutes without electricity isn’t a hardship. It’s barely an inconvenience.

Eventually the power comes back. The Wi-Fi reconnects. Teams signs me back in. The day carries on as though nothing had happened, and I can finally make some lunch.

The feeling of being trapped disappears almost as quickly as it arrived.

By tomorrow, I’ll have forgotten all about it.

Process debt: the hidden challenge in digital transformation

markw@markwilson.co.uk (Mark Wilson) — Mon, 3 Aug 2026 00:00:00 +0100

Process debt: the hidden challenge in digital transformation

A different kind of check-in

A few weeks ago, I stayed at a large hotel that’s part of a well-known international chain.

Before I arrived, I’d been encouraged to check in using the hotel’s app. That’s fairly common these days, so I didn’t think much of it. I assumed it would save me a minute or two before I collected a key card from reception.

Instead, as soon as my room was ready, the app assigned it to me and activated a digital key on my phone. While a colleague queued at reception to collect a key card, I went straight to my room, held my phone against the lock and let myself in.

Checking out the following morning was equally straightforward. A couple of taps in the app, my invoice arrived by email and I was on my way.

It wasn’t until later that I realised why the experience had stood out. The hotel hadn’t invented any new technology. Smartphones have supported NFC for years, and electronic locks are hardly revolutionary. What had changed was the process.

Someone had looked at the traditional hotel check-in experience and asked a simple question: why does every guest need to visit reception?

Process debt

In technology, we spend a lot of time talking about technical debt. I wonder whether organisations also accumulate something I’d call process debt.

Processes are usually introduced for good reasons, but they’re rarely revisited. Over time, they become accepted as “the way we do things”. Instead of questioning them, we build systems around them, automate them and eventually forget why they existed in the first place.

Once you start looking for it, you see it everywhere: a chatbot that eventually tells you to call the contact centre, an online form that still results in someone re-keying the information into another system, or a polished mobile app that’s really just a more modern front end for an old process. Technology has improved the experience around the edges, but the underlying process remains unchanged. The technology looks modern, but the thinking behind it hasn’t changed.

Start with the outcome

Good digital transformation starts by asking what you’re trying to achieve, not how technology can improve the current process.

Too often, organisations start with the process they’ve always had and look for ways to make it faster, cheaper or more efficient. They don’t go far enough because they never stop to ask whether the process is still the best way of achieving the outcome.

That’s what the hotel had done. It hadn’t asked, “How can we make check-in more efficient?” It had asked, “What does a good arrival experience look like?” Once you ask that question, the reception desk becomes optional rather than inevitable.

Looking beyond technology

That’s why I think we should spend as much time thinking about process debt as we do technical debt. Sometimes the right answer is to automate a process, sometimes it’s to simplify it and sometimes it’s to remove it altogether.

Good digital transformation doesn’t start with the process. It starts by asking what success looks like. Once you’ve answered that question, you can decide whether the process should be improved, simplified or removed altogether.

LinkedIn is a network, not an audience

markw@markwilson.co.uk (Mark Wilson) — Thu, 30 Jul 2026 00:00:00 +0100

LinkedIn is a network, not an audience

I’ve been on LinkedIn for almost twenty years. I was one of the first million users in the UK.

When I joined, it wasn’t a social media platform. It was a professional networking site where you kept in touch with colleagues, customers and people you’d met at conferences. If you changed jobs, LinkedIn helped you stay connected.

Since then, LinkedIn has changed enormously. Today it’s full of creators, influencers, newsletters, videos and people who’ve built businesses around their LinkedIn presence.

Which makes me wonder if I’m the one who’s got it wrong.

When LinkedIn was just LinkedIn

A few months ago, someone I know in the AI community suggested I was missing a trick.

She pointed out that I don’t accept every connection request. If I did, my posts would reach far more people. And she’s right. If my goal was to maximise the reach of my posts, then accepting every invitation would almost certainly help.

It made me realise that we were using LinkedIn for completely different things. She sees it as a platform for building an audience. I still think of it as somewhere to stay in touch with people I know and trust.

A connection isn’t a relationship

As I write this, I have 169 unanswered connection requests sitting in my LinkedIn inbox.

All are from people I don’t recognise. Some are clearly hoping to sell me something. Others may have mutual connections, but no explanation of who they are or why they’d like to connect. Most contain no message at all.

I sort of understand the lack of a message because I don’t subscribe to LinkedIn Premium, so I’m limited in the number of personalised invitations I can send each month. That isn’t really a problem because, most of the time, we’ve already had the conversation.

If I’ve met someone at a conference, after a presentation or during a customer meeting, I’ll usually mention that I’ll connect with them on LinkedIn. Sometimes we’ll do it there and then. Sometimes I’ll send the invitation later that day. Either way, it’s simply a continuation of a conversation we’ve already had.

That’s why so many unsolicited connection requests feel odd to me.

Imagine receiving an email that simply said:

Hello.

Will you reply?

No introduction. No context. No explanation of why the sender had contacted you. You’d probably ignore it. So why is that considered acceptable on LinkedIn?

LinkedIn makes it easy to collect connections. That’s not the same as building a network.

The coffee test

For many years, I was very active on Twitter. Like many people, I shared all sorts of things: technology, photography, family life, holidays and the occasional random observation. It was social media in the broadest sense.

LinkedIn was different. It was where I connected with colleagues, customers and people I’d met at events.

Over the last few years, those worlds have started to overlap. As Twitter became X and declined, many people moved to LinkedIn and brought their posting habits with them. The platform changed, and perhaps the expectations changed too.

These days, I have a simple test before I post something on LinkedIn. Is it something I’d naturally talk about over coffee with a business contact?

Sometimes that’s a technology observation. Sometimes it’s a conference I’ve attended, or an article I’ve written. Occasionally it’s something more personal, because working relationships are built between people, not job titles.

But there are plenty of things that don’t pass that test. Years ago, I’d happily have shared photos of my sons’ sporting achievements on social media. Today, I wouldn’t normally post those on LinkedIn (though there may be some exceptions as they take their steps into the world of work). Not because I’m trying to keep my personal life secret, but because they belong somewhere else.

It’s entirely possible that I’m using LinkedIn as it was, rather than as it is.

My role certainly involves sharing ideas. I write articles, speak at events, represent Node4 and help organisations make sense of technology. LinkedIn is an important part of that.

But I don’t think those conversations are valuable just because they reach more people. They’re valuable because they’re built on relationships. The network comes first.

Perhaps one day I’ll decide I’ve been looking at LinkedIn the wrong way all along. Perhaps the platform has changed and I haven’t changed with it.

But I’m OK with that.

I’d still rather have a network than an audience.

You can't beat a cable

markw@markwilson.co.uk (Mark Wilson) — Fri, 17 Jul 2026 00:00:00 +0100

For years, I had one simple rule for my home network: keep it simple.

As our family’s dependence on reliable internet grew – for working from home, streaming TV, online gaming and, back then, homework – I deliberately removed complexity wherever I could. The old home servers disappeared. Exchange Server gave way to Microsoft 365. Files moved to OneDrive, with a Synology NAS providing local storage for anything that didn’t belong in the cloud. More recently, after a bout of Azure bill shock, that was joined by a second, smaller NAS in a separate building to provide an additional backup.

There are still pockets of complexity. Home Assistant runs much of our smart home, for example (and the Zigbee pains are a whole seprate topic). But I’ve always tried to keep the underlying infrastructure as simple and as reliable as possible.

Keeping it simple

Back in 2018, after converting our loft and spreading the house across three floors, I invested in an AmpliFi HD mesh Wi-Fi system. If you don’t know AmpliFi, it’s Ubiquiti’s consumer networking range. Most IT professionals will probably be more familiar with UniFi, which targets business users.

The AmpliFi system was excellent. It covered the house well and, when something did occasionally go wrong, everyone in the family knew the fix: unplug the nearest mesh point, plug it back in, and carry on. It wasn’t glamorous, but it worked.

Then one of the mesh points failed.

At that point I decided it wasn’t worth replacing a five-year-old Wi-Fi 5 system. Instead, I’d “upgrade” to UniFi. Looking back, that’s where the problems began.

The upgrade that wasn’t

Before anyone thinks this is a rant about UniFi, it isn’t. I like the platform a lot. It gives me enterprise-grade networking features at home for a relatively modest cost. The mistake wasn’t buying UniFi. The mistake was buying it before fixing the foundations.

I started with a handful of UniFi Express (UX) devices. They looked perfect: compact, unobtrusive and, perhaps most importantly, aesthetically acceptable around the house. They could connect over Ethernet where it was available or create a wireless mesh where it wasn’t.

Unfortunately, I ignored one piece of advice that seemed to appear in almost every online discussion: don’t use a UniFi Express as your main router if you have a busy network.

Combined with a patchwork of wireless meshing and Powerline Ethernet adapters, the result was… disappointing.

The network became slower, less reliable and generally more frustrating than the AmpliFi system it had replaced. Even today, my sons still remind me, “You should have kept the AmpliFi, Dad”.

They weren’t entirely wrong.

Chasing the wrong problem

Over the following months I upgraded the network piece by piece. One of the UniFi Express units eventually failed, which gave me the excuse to replace it with a UniFi Cloud Gateway Ultra (UCG Ultra). That not only removed the limit on managed devices but also provided a significantly more capable router. My ISP’s router remains in place purely to terminate the broadband connection, with the UniFi Cloud Gateway handling the routing.

Migrating the configuration mostly went smoothly because I found I could restore my UX backup to the UCG, although I inherited an odd problem with the guest network that refused to go away, even after factory resets. The captive portal simply wouldn’t behave properly. I spent months exchanging logs with Ubiquiti support before eventually solving it myself by rebuilding that part of the configuration and moving the guest network onto a pair of dedicated U6+ access points.

Each upgrade improved something. The UCG made routing faster. The U6+ access points improved Wi-Fi coverage. Performance steadily crept upwards.

Eventually I admitted what I’d been trying to avoid for years. Wi-Fi wasn’t really the issue. I’d built an increasingly capable wireless network on top of an increasingly inadequate backbone.

Copper beats radio

Our house is now around 35 years old and was built long before anyone imagined every room might need multiple network connections. Indeed, back then, volume home builders only put single electrical sockets in rooms - not even double sockets. As a result, there are very few convenient routes for new cables without lifting carpets, taking up floorboards or opening walls. There are certainly no risers and there’s no structured cabling.

When we extended the house in 2009, I asked the electrician about installing Ethernet while the walls were open.

“You won’t need it,” he told me. “Wi-Fi is the future.”

Foolishly, I listened to him. Wireless certainly got better. But I’ve yet to find a radio signal that beats a piece of copper.

The frustrating part is that I missed the opportunity twice. We extended the house again in 2018 (converting the loft) and, once again, no Ethernet went in.

A weekend with a drill

Last weekend, my long-time friend Stuart – a former IT Manager who has retrained as an electrician – gave up much of his Saturday to help me fix the problem properly.

Rather than trying to route cables internally, we ran external-grade CAT6 around the outside of the house, following the path of an old telephone cable before bringing it back inside where it was needed. Three cables now leave what I affectionately refer to as “the datacentre” – an IKEA wall cabinet in the garage.

One link feeds the Man Cave, another terminates in the dining room where an old coaxial TV socket once lived, and a third is simply spare. Knowing me, it probably won’t stay that way for long.

Then, this week I finished the job by replacing the last of my ageing unmanaged Netgear switches with UniFi switches. The UCG now connects to a UniFi USW Lite PoE switch, which powers the U6+ access point in the hallway as well as providing wired connectivity to the rest of the house. A USW Flex Mini switch lives in the Man Cave alongside the backup NAS, while the remaining UX units now do what they were always meant to do: provide wireless coverage in the more awkward corners of the property, rather than trying to compensate for a weak backbone.

Finally, a network that makes sense

For the first time since I started this journey, the network architecture actually makes sense.

The house now has a gigabit wired backbone between the key locations, with Wi-Fi used where it adds value rather than where it has to compensate for missing infrastructure. There are still a couple of Powerline links serving the loft, but they’re now the exception rather than the rule. And the U6+ AP in my office is powered with a PoE injector and connected over a mesh to its peer in the hall, but that seems to work well - even if it is technically inferior to a wired connection.

The difference is immediately noticeable. Roaming between access points is smoother, latency is lower and large file transfers are finally limited by the broadband connection rather than by the network inside the house.

I’ve also added a UniFi Travel Router, which deserves a post of its own. It uses the UniFi Teleport VPN, which allows me to connect to public Wi-Fi while securely tunnelling back to my own network. That means I can browse through my own internet connection, access private resources safely and avoid relying on whatever security the hotel or café happens to provide.

The journey

The network didn’t change overnight. Looking back, each upgrade solved one problem. Unfortunately, I fell into the trap where we often optimise the thing we can see instead of fixing the thing that matters.

2009: House extension
❌ Didn't install Ethernet on ground and 1st floors.

2018: Loft conversion
➡️ Installed AmpliFi HD
❌ Missed additional Ethernet cabling opportunity for 2nd floor

~2023: Move to UniFi
➡️ Replaced Amplifi HD with 4x UniFi Express
➡️ Replaced Powerline Ethernet adapters for "up to 1Gbps"

~2024: Fibre broadband, more capable routing
➡️ Upgraded WAN from FTTC to FTTP
➡️ Replaced 1x UniFi Express with UniFi Cloud Gateway Ultra

~2025: Improved Wi-Fi
➡️ Installed 2x U6+ access points, turned off 5Ghz radio on 2x UniFi Express to direct modern devices to more capable APs

2026: Fixed the foundations
➡️ Upgraded WAN link from 100 to 150Mbps down (18Mbps up)
➡️ CAT6 installed = wired backbone at 1Gbps (10Gbps capable)
➡️ UniFi switches
✅ Network finally behaves

The evolution of my home network in recent years.

Foundations first

It’s taken several years, more money than I’d originally intended to spend, and 100m CAT6 cable to get here (and I should have bought more).

The irony isn’t lost on me. I spent years buying smarter networking equipment, tweaking configurations and chasing new features, only to discover that the biggest improvement came from putting the right foundations in place.

There’s a wider lesson too, because, in technology, we often assume the answer is another appliance, another feature or another layer of software. Sometimes the real answer is to fix the infrastructure underneath.

That’s true in my home network. It’s just as true in the organisations I work with every day. In my case, it turned out to be couple of lengths of CAT6 cable.

My sons haven’t been home to comment on the Wi-Fi yet. We’ll see what they think next time - or whether they still say “You should have kept the AmpliFi, Dad”.

Current network

Device	Purpose
Vodafone router	Modem only: Wi-Fi disabled
UniFi Cloud Gateway (UCG) Ultra	Router and UniFi controller
UniFi Lite PoE	Core switch
2 × UniFi U6+	Primary wireless access points (guest and private SSIDs)
3 × UniFi Express (UX)	Additional access points (private SSID only)
UniFi Flex Mini	Man Cave switch
UniFi Travel Router	Secure travel networking

Still on the list

Replace the remaining Powerline Ethernet links by running another CAT6 cable into the loft.
Use the spare external CAT6 run (potentially for more CCTV or maybe the loft connection).

Future-proofing

CAT6 (rather than 5E) means I can upgrade the wired backbone to 2.5Gbps or even 10Gbps one day.

AI isn’t killing consulting. It’s killing time as a proxy for value

markw@markwilson.co.uk (Mark Wilson) — Thu, 16 Jul 2026 00:00:00 +0100

“Why am I paying a consultant for something AI can do in 15 minutes?”

I’ve heard variations of that question several times over the past few months. On the face of it, it’s a perfectly reasonable one.

Generative AI can draft documents, analyse data, write code and create presentations in a fraction of the time they once took. If the work now takes minutes instead of hours, why should customers still pay the same?

The answer is that they were never buying your time in the first place.

Rewind a decade or so in IT consulting

About ten years ago, I was regularly delivering a consulting engagement for risual. We called it a Vision and Scoping Document and it typically took around ten days to complete.

My sales colleagues would often describe it to customers as a “ten-day engagement”. Every time I heard that, it grated.

The customer wasn’t buying ten days. They were buying clarity — a clear understanding of where they were, where they wanted to be and a roadmap for getting there. The ten days were simply our estimate of what it would cost us to deliver that outcome.

If we found a better way of working and delivered it in eight days, we’d make a little more profit. If unforeseen issues meant it took twelve days, that was our problem, not the customer’s. That’s how consulting should work.

Yet, for years, we’ve become comfortable describing knowledge work in terms of days, hours and rates. Time became a convenient way of pricing expertise, even though it was only ever a proxy for the value being delivered.

It’s not just consulting

A couple of weeks ago, I collected my car after some repair work. The invoice listed one hour of labour at £95 plus VAT, alongside the cost of the replacement parts. The only problem was that I’d only been at the garage for around forty minutes. So I queried it, and was told “It’s a plug in.”

That was the garage’s shorthand for a fixed-price job. They weren’t charging me for exactly sixty minutes of someone’s time. They were charging a standard amount for carrying out that repair. In other words, it was already value-based pricing.

The irony is that they were already selling value — but presenting it as time and materials. Time was simply an easy way of arriving at a price. By describing it as an hour of labour, they’d encouraged me to question the price instead of the outcome.

The locksmith test

Imagine calling a locksmith because you’ve locked yourself out of your house. They arrive, examine the lock for a few moments, insert a couple of tools and, within thirty seconds, the door is open.

Do you argue that the bill should be lower because it only took thirty seconds? Of course not.

You’re paying for the years it took them to learn their trade, the specialist tools they’ve invested in and the confidence that they’ll open the door without damaging the lock. You’re paying for the outcome.

Indeed, if they spent two hours trying different techniques before eventually getting you in, you’d probably have less confidence in their expertise, not more.

We rarely question paying for expertise when we can see the outcome immediately, so why should consulting be any different?

AI changes the economics, not the value

Warren Buffett famously observed that “Price is what you pay. Value is what you get.”

For years, we’ve treated time as though it were synonymous with value. AI is forcing us to revisit that assumption. Because customers aren’t buying typing. They aren’t buying prompt engineering. They aren’t buying someone to spend ten days producing a document.

Customers are actually buying judgement, experience and confidence. They’re buying better decisions, reduced risk and faster progress.

AI changes the cost of delivering those things, but it doesn’t diminish their value. If anything, it increases the importance of knowing when to trust AI, when to challenge it, and how to turn its output into something genuinely useful.

AI is already an extraordinary amplifier of expertise. Put it in the hands of someone who understands the problem they’re trying to solve, and they’ll often deliver better work, faster than ever before. Put it in the hands of someone without that experience, and they’ll often produce something that looks convincing but lacks the judgement that comes from years of practice.

AI compresses effort. It doesn’t compress experience.

Anyone can ask AI to generate a strategy document, but producing one that reflects an organisation’s objectives, culture, constraints and appetite for change is something else entirely.

The commercial shift

I don’t think AI is going to kill consulting firms. I do think it’s going to accelerate the move away from charging for effort and towards charging for outcomes.

The firms that continue to sell “ten-day engagements” may find customers asking increasingly uncomfortable questions. Those that sell measurable business value, and use AI to improve the efficiency of delivering it, should become more competitive, not less.

Perhaps that’s the biggest commercial shift AI will bring. Not the end of consulting — but the end of pretending that time is a good proxy for value.

The future of autonomous driving is not on public roads

markw@markwilson.co.uk (Mark Wilson) — Mon, 13 Jul 2026 00:00:00 +0100

Earlier today, I shared a post on LinkedIn after reading about The Port of Felixstowe expanding its autonomous truck fleet to 100 vehicles.

I realised this isn’t really a story about ports. It’s a story about where autonomous driving delivers value first — and why that probably isn’t where most people are looking.

Last year I wrote about software meeting steel: the idea that AI is escaping our screens and beginning to interact with the physical world. From delivery robots and airport service robots to autonomous systems in warehouses and factories, we’re already surrounded by examples. They just don’t look like the self-driving cars we’ve been hearing about from billionaire (trillionaire) technology executives.

The expansion at Felixstowe is another step along the journey. This time, though, the robots aren’t delivering takeaway meals or moving pallets around a warehouse. They’re shifting shipping containers around one of Britain’s busiest ports.

And this isn’t a pilot or a proof of concept. It’s a fleet expansion. The technology has demonstrated enough value to justify significant further investment.

Looking in the wrong place

Normally, whenever autonomous vehicles are discussed, the conversation is centred around self-driving cars on public roads.

How will they cope with city traffic? Motorways? Country lanes? School runs? Snow? Roadworks? Cyclists? Pedestrians?

They’re fascinating problems, but I think we’re looking in the wrong place.

If you want to understand where autonomy will succeed first, don’t look at the public roads.

We already trust cars to drive themselves — sometimes

In truth, today’s cars already contain elements of autonomous driving.

Adaptive cruise control maintains a safe distance from the vehicle ahead. Lane centring gently steers the car. Automatic emergency braking can intervene if a collision appears imminent. Some systems will even change lanes or park the vehicle with minimal driver input.

These features are impressive, and they’re becoming increasingly commonplace. But they also reveal how difficult the remaining challenge is.

Anyone who has driven a modern car has probably experienced phantom braking, where the car reacts to a perceived hazard that isn’t really there. Equally, there are situations where the technology struggles to recognise something unusual because it doesn’t match the patterns it has been trained to detect.

A plastic bag blowing across the carriageway, a police officer directing traffic around an accident, or a child chasing a football into the road all demand judgement, not just object recognition.

Human drivers don’t just recognise cars, buses and lorries. We interpret context. We notice unusual behaviour. We anticipate what someone might do next.

That level of hazard perception remains extraordinarily difficult to replicate reliably across every possible road, weather condition and traffic scenario.

A port is a different world

Compare that with a controlled industrial environment, such as a port.

The roads are private. Routes are carefully defined. Speeds are low. Traffic flows are predictable. The vehicles all have a specific purpose.

That doesn’t make the engineering easy. It simply makes the problem more constrained. The unexpected still happens, but it happens within a much narrower set of possibilities.

More importantly, the commercial case is obvious.

Container shipping is built around efficiency. Every minute a ship spends alongside the quay costs money. Every unnecessary delay moving containers between cranes, storage yards and rail terminals affects the throughput of the entire port.

Autonomous vehicles don’t need to solve every edge case on Britain’s road network. They simply need to move containers safely, consistently and efficiently around a known environment.

If they can do that, they reduce bottlenecks, improve utilisation and generate a measurable return on investment.

We’ve seen this pattern before

Ports aren’t unique. Warehouse robots have been doing similar work for years. Delivery robots operate within carefully defined areas. Autonomous agricultural machinery works in fields with known boundaries. Mining companies have deployed autonomous haul trucks in remote sites where routes and operating conditions are tightly managed.

These systems all share the same characteristics.

The environment is constrained.
The rules are predictable.
The business value is obvious.
The risks are understood and manageable.

That’s why they succeed.

This isn’t unique to ports, warehouses or agriculture. It’s how new technologies generally find their footing. They succeed first where the environment is controlled, the business case is compelling and the risks can be managed. Only later do they move into the messy, unpredictable outside world.

Look behind the fence

We tend to judge the success of autonomous vehicles by asking whether they can drive us to the supermarket or the station.

Instead, let’s look at where autonomous driving is creating value today.

The answer lies behind the fences of ports, factories, warehouses, distribution centres and industrial estates. Places where software is increasingly controlling the physical world, and where shaving seconds from a process can save significant money each year.

One day, fully autonomous vehicles may become an everyday part of life. But before that happens, they’ll quietly become an everyday part of ports, factories, warehouses and distribution centres.

The hidden timer in my car

markw@markwilson.co.uk (Mark Wilson) — Fri, 26 Jun 2026 00:00:00 +0100

“It just needs to get me across France and back, and then I’ll replace it.”

Those were my words after a rather large bill that sealed my Volvo V60’s fate.

A few months ago, I wrote about the hidden cost of complexity in modern cars. I argued that today’s vehicles are increasingly defined by software, and that this complexity ultimately makes them more expensive to own and maintain. I hadn’t expected my own car to provide another perfect example quite so soon.

It’s been a great car, but it’s eight years old now and things are starting to go wrong. A new steering rack, along with fitting, replacement tyres, tracking and brake pads, came to almost £3,000 of unexpected expenditure. To be fair, the tyres and brakes were expected, but I’d already started thinking about replacing the car when another warning appeared on the dashboard.

A warning

Halfway through a road trip across France, the message appeared:

Volvo On Call. Service required.

My first thought was some dodgy fuel from an unattended rural petrol station where I’d filled up earlier that day. The car was running perfectly normally though, so I kept driving. Remote support was only likely to result in an expensive visit to a franchised dealer. From that point on I filled up with premium diesel, just in case, but nothing changed. The warning remained, while the car continued to drive without issue.

Back in the UK, I called Avalon Cars, the independent Volvo specialist I’ve used ever since the car came off warranty. They use OEM parts and the same diagnostic equipment and software as the main dealers, but the labour rate is merely expensive rather than extortionate.

They asked what message I’d seen and immediately told me what the problem would be. They didn’t even need to see the car.

Spoiler: it wasn’t the fuel.

An eight-year timer

It turned out to be the backup battery for the Volvo On Call (emergency and remote services) system. Not because it had failed. Not because it had been tested and found to be at the end of its life. Simply because the software decides that, eight years after the car was built, it’s time to replace it.

I could ignore it, but I’d have to dismiss the warning every time I start the car. And since I want to sell it in a few months, I’d rather not advertise it with a warning message on the dashboard.

The fix? £222 for the battery, an hour’s labour and a software update.

Nobody tells you when you buy a car that one day it will celebrate its eighth birthday with a service bill.

It also made me wonder how many more arbitrary timers are quietly counting down in the background, waiting for their turn.

Rethinking ownership

More importantly, it reinforced the conclusion I’d already reached. Cars are becoming software platforms first and mechanical products second. Parts are coded to the vehicle. Components are increasingly replaced as complete assemblies rather than individual parts. Even seemingly straightforward maintenance often requires dealer software. Some features are offered on a subscription basis. Whether intentionally or not, the ownership experience is becoming one that’s difficult to separate from the manufacturer’s service network.

Perhaps that’s why this doesn’t receive more attention. Many new cars are financed and replaced every three or four years. By the time arbitrary eight-year maintenance timers appear, the original owner is long gone. As my brother put it, that’s not something you mention when selling the car. It’s someone else’s problem by then.

Yesterday, I ordered an electric vehicle on a lease.

It’s expensive — more expensive than I would have liked — but I know exactly what I’ll be paying each month. The insurance is included, servicing is covered and the legal costs (VED, etc.) are taken care of. All I need to worry about is charging it, and that should cost less than diesel.

I’ll be tied into a subscription for years to come, but at least the costs won’t be hidden.

Copilot Cowork just got a price tag

markw@markwilson.co.uk (Mark Wilson) — Thu, 18 Jun 2026 00:00:00 +0100

OK, so there is a significant risk that this post will just be lost as noise in a sea of outcry and “analysis” of the recent Copilot charging changes, but here goes… this is my 4 penn’th.

First up, what’s this all about?

Microsoft has just taken Copilot Cowork from Frontier status — beta for those of us who remember when software was tested before being released to an unsuspecting public — to general availability (GA). And now the meter is running, with charges taking effect from next month.

If you don’t use Copilot Cowork, then it’s off by default. But if you used it in Frontier, then it will remain enabled and start accruing usage charges from 1 July. And that’s in addition to your base Microsoft 365 Copilot subscription.

An important caveat

Quoting one LinkedIn post:

“Your everyday Copilot has not changed. Word, Outlook, Teams, Chat. Still flat, per-user, predictable. […] The meter only runs when you send Cowork off to do multi-step work in the background.”

Why is this important?

Copilot Cowork has become a huge part of what I do. I previously found Microsoft 365 Copilot to be of limited use. It was OK, and had the advantage of being grounded in my business data, but the results were patchy. Copilot Cowork is built on Anthropic’s Claude models and, in my experience, the difference in output quality is night and day.

Since I got access in April, I have been delegating parts of my work to it — including daily analysis of my emails, messages and meetings, and helping me to draft content. At Node4, we even have skills that allow it to create content using the company brand.

How does the pricing work?

Each Copilot Cowork task is metered on four inputs and billed in Copilot Credits:

Model use — which model runs, and for how long
Context retrieval — how much of your data it reads
Tool calls — how many actions it takes
Runtime — how long the task runs end-to-end

A light task might be 125 credits (approximately $1.25), a medium task 500 credits (approximately $5), and a heavy task 2,500 credits (approximately $25).

Microsoft’s own model assumes 22-38 tasks per user per month. Spend is uncapped by default, and many tasks are scheduled — they run unattended with no-one watching the meter.

I used an estimation tool from Microsoft MVP Josh Cook to try and understand my costs. It was quite a lot — well into three figures for the last month — and I was away from work for two weeks!

[Edit: my colleague Jay Fitzhenry later told me about the /cost skill in Cowork, which provides details of credit usage in-session. These estimates were more reassuring (about half the number), though I’m now confused about which estimate to believe. Nevertheless, and regardless of the actual number, variable pricing means brings new challenges to be addressed.]

Now, we also have to consider that I have a cost as an employee. If Copilot Cowork saved time, then that may still have been a good investment. That’s a decision that needs to be made by my managers, though it would be good if I had the tools and insights to make informed choices about my use of AI tools. I have some latitude on other expenses that I incur for the company (e.g. travel), and there are some things I just cannot do (e.g. buy equipment). I’d like to think that we will reach a point where AI spend is similarly simple to justify or not.

A grace period

Microsoft has been extremely generous. It has to pay Anthropic for use of its models and we’ve all had free access during the Frontier phase. We’re still not paying, but the cost of running Copilot Cowork is suddenly visible.

I’d go further and say that the age of free or low-cost generative AI tool access is coming to an end. The financial (and environmental) costs are huge and the business model is still unclear.

Someone has to pay for all those GPUs eventually.

Using the right tools for the right jobs — and understanding which tasks will burn tokens

My analysis gave some other insights too. The analysis itself accounted for around 10% of my token usage. And some tasks that were particularly expensive were ultimately unsuccessful. I used the content from the chat, but the presentation slides it created were not up to scratch.

If I’d paid an agency to create something unusable, I’d be asking awkward questions about rebates. AI billing doesn’t work like that.

So, what can we do to avoid a huge bill?

It shouldn’t be a surprise that my employer, Node4, is all-in on AI. We have an extensive AI transformation plan in place and, like many organisations, we’re actively thinking about how to balance innovation with governance and cost control.

Without giving away specifics, uncontrolled Copilot Cowork usage could add significant costs to any organisation. Clearly that can’t be allowed — but there are some straightforward steps that can help.

Control	Visibility	Efficiency
Set tenant, group and user spend limits	Configure usage alerts with named recipients	Choose appropriate plans once the run-rate is known
Apply hard caps so spend cannot run away	Review usage weekly by user, group and feature	Use cheaper models for routine work where appropriate
Keep Copilot Cowork scoped to approved groups	Enable per-task pricing visibility when available	Coach heavy users on cost-aware prompting
Apply per-user budgets inside group policies	Monitor scheduled and unattended tasks closely	Right-size who actually needs Copilot Cowork

As Josh Cook says in his post, the takeaway is that:

Prompt design matters.

Context matters.

Skills matter.

Reusable instructions matter.

Clear task boundaries matter.

If you hand Cowork vague work with no structure, you are going to burn more [tokens].

If you build skills, give clean context, and tighten the task, the experience becomes way more efficient

The grace period ends on 1 July 2026, so you need to act now.

None of this means Copilot Cowork isn’t valuable. Quite the opposite. The fact that we’re now having conversations about cost rather than capability really tells us something important about where these tools are heading.

Footnote: this post was written by a human who likes em dashes — I was too scared by the token cost to get an AI to do this for me, though I did use ChatGPT for the image.

From Bletchley to a nationwide AI community — and a vibe-coded website

markw@markwilson.co.uk (Mark Wilson) — Thu, 21 May 2026 00:00:00 +0100

From Bletchley to a nationwide AI community — and a vibe-coded website

I’m pretty stoked today because, after a very late night vibe coding session last night, I finally got the UK AI User Groups website to a state where I felt it was ready to go live. And, now the DNS changes have propagated, we are in business.

I’ve been involved in the Bletchley AI User Group from the sidelines for a while, but then my friend Phil Kermeen got involved and I became a regular attendee. After I was invited to speak last year, I suddenly found myself “voluntold” onto the committee. So, with strict instructions from my family not to take on yet more projects/responsibilities, I agreed to photograph the Bletchley events and create a website.

That escalated quickly.

I am not a web developer

OK, so, I’m not a web developer. I have a Computer Studies degree (so technically, “I can code”), but I haven’t written a line of code professionally since my internship in 1992/3 — except perhaps some MS-DOS batch files or a bit of bad PowerShell. I’ve had my own website since at least 1998 though, so dabbled in HTML/CSS, but mostly used WordPress (until I switched to Hugo earlier this year — that’s another story/post).

The point about the UK AI User Groups site is that I wanted to write something that:

is easy to maintain;
separates presentation from site logic and from data;
looks OK, using the group’s visual identity (I’m not a designer either);
feels “modern”;
is accessible and easy to navigate.

Cutting code

And this is where the vibe coding comes in…

I know some people will tell me there are better tools for coding, but I have a personal ChatGPT Plus subscription. That was enough for me to create a simple single-page site using HTML, CSS, and JavaScript. And then I got the memo that this was not just for Bletchley AI, but for all of the UK AI User Groups, adding a extra page for each group.

I used the W3.CSS framework with Font Awesome icons — in common with my own websites — and ChatGPT handled much of the app logic and presentation. Data (events) are read from a single file, and presented in various forms across the site.

There’s light and dark modes, and the site is responsive so it should gracefully adjust to different devices — though the best user experience is still in a desktop browser. I have my limitations (testing/access to devices and skill — as well as time).

A companion script pulls images from the actual event management platforms (currently Meetup and Luma) so we can bring the event cards on the site to life. And then, on the train home last night I decided to see if I could show the various groups as either a list or on a map — so Leaflet does all the heavy lifting for the maps, with the page offering a toggle between the two views.

Hosting it for free

Oh yes, and did I say this is free? Yep, hosted in a GitHub repository with CI/CD to push it to Azure Static Web Apps on the free tier.

The last challenge was the DNS setup for the custom domain — not difficult, but one of those tasks where everything feels slightly more complicated than it should because of the quirks of the GoDaddy domain registration.

So that’s it. It went live today. A vibe-coded website at https://www.ukaiusers.co.uk/

“Anyone can do that”

Now, some will be reading and saying “whoop-de-do Mark, anyone can do that” — and that’s my point. I’m not taking a professional developer’s work away — proper sites still need proper skills. Not just to write code but to perform proper analysis, design, creation, testing, and support. The code was never the whole job.

What generative AI (and just enough skill to be dangerous) has allowed me to do is create something functional, for a group with no budget for professional services (any sponsorship revenue goes to running the events). And it’s allowed me to exercise my inner geek.

And that’s where I see value in generative AI. Bouncing ideas around. Helping to draft content. Assisting people. And letting projects that would never get past the business case come to life.

The support question

Let’s discuss the elephant in the room though — support. A vibe-coded site is a bit of a black box. I understand enough to say “this is what I have created and this is how it works at a high level”. I don’t understand every line of JavaScript though, and the CSS had a lot of fixes applied to get it to behave properly.

I’ve no doubt that this could do with a complete refactor. Maybe I can use another AI assistant for that some time.

And that’s the main point I want to get across. Vibe coding is not magic. It doesn’t remove the need for skill, judgement or support. But it can lower the barrier enough for an idea to become real.

Sometimes that’s enough.

Especially when the alternative was another good idea sitting on the “maybe one day” pile.

Mark's Thoughts

Bring your own network

Bring your own network

The problem with public Wi-Fi

One network, everywhere

One login, all your devices

More than just security

It’s not perfect

Bringing your own network

Powerless: the invisible foundations of modern life

Powerless: the invisible foundations of modern life

Disconnected

Invisible foundations

Normal service resumes

Process debt: the hidden challenge in digital transformation

Process debt: the hidden challenge in digital transformation

A different kind of check-in

Process debt

Start with the outcome

Looking beyond technology

LinkedIn is a network, not an audience

LinkedIn is a network, not an audience

When LinkedIn was just LinkedIn

A connection isn’t a relationship

The coffee test

You can't beat a cable

Keeping it simple

The upgrade that wasn’t

Chasing the wrong problem

Copper beats radio

A weekend with a drill

Finally, a network that makes sense

The journey

Foundations first

Current network

Still on the list

Future-proofing

AI isn’t killing consulting. It’s killing time as a proxy for value

Rewind a decade or so in IT consulting

It’s not just consulting

The locksmith test

AI changes the economics, not the value

The commercial shift

The future of autonomous driving is not on public roads

Looking in the wrong place

We already trust cars to drive themselves — sometimes

A port is a different world

We’ve seen this pattern before

Look behind the fence

The hidden timer in my car

A warning

An eight-year timer

Rethinking ownership

Copilot Cowork just got a price tag

First up, what’s this all about?

An important caveat

Why is this important?

How does the pricing work?

A grace period

Using the right tools for the right jobs — and understanding which tasks will burn tokens

So, what can we do to avoid a huge bill?

From Bletchley to a nationwide AI community — and a vibe-coded website

From Bletchley to a nationwide AI community — and a vibe-coded website

I am not a web developer

Cutting code

Hosting it for free

“Anyone can do that”

The support question