Mark's Thoughts

What does a CTO do? It depends — and that’s the point

markw@markwilson.co.uk (Mark Wilson) — Tue, 31 Mar 2026 00:00:00 +0100

Same title, different jobs

Ask ten people what a CTO (Chief Technology Officer) does and you will get ten confident answers. What’s interesting is that each answer is usually correct in its own context. In one organisation, the CTO runs engineering. In another, they shape product direction. In a third, they sit alongside the board and help make sense of technology choices. The title is consistent, but the role is anything but.

That variation can make the role feel vague from the outside but the shape of the job reflects the organisation around it. A startup trying to build something new needs a very different kind of CTO to a large enterprise trying to modernise decades of systems. A services business has different needs again, where credibility, narrative, and client engagement matter just as much as technical depth.

The common core

Despite that variation, there’s a common core. Strip away the organisational differences and the job usually comes down to three things: setting direction, making decisions, and developing leadership:

Direction is about ensuring there is a coherent technology roadmap that supports the business rather than competing with it.
Decisions are about navigating trade-offs that do not have neat answers — speed vs. resilience, standardisation vs. differentiation, build vs. buy.
Leadership is about creating the conditions for others to succeed.

This is where the role is often misunderstood. A CTO should not be the person with all the answers, or the most hands-on engineer in the room. The job is to help the organisation consistently arrive at better answers, and to turn individual pieces of work into something that adds up to a clear direction.

For a neutral definition of the role, MIT provides a useful overview of how CTO responsibilities typically span strategy, architecture, and innovation.

Turning technology into outcomes

Several years ago, I came across a line about the CIO (Chief Information Officer) that stuck with me. Gideon Kay observed on Twitter that:

“The mark of a strong CIO is the ability to turn a technology platform question into a business outcome answer.” — Gideon Kay (@gids), April 2017

That felt right, but I thought it could also work the other way round:

“The mark of a strong CTO is the ability to take a business problem and turn it into a technology solution.” — Mark Wilson, May 2017

Today, I would add that the solution needs to deliver something meaningful to the business — not a solution in isolation, but one that aligns to both business and technology strategies.

That distinction shifts the role away from technology for its own sake, and towards translation. The CTO sits between ambition and execution, helping ensure that what gets built is both technically sound and commercially relevant.

Why context matters

The reason the role varies so much comes down to a few factors. Stage is one:

Early-stage organisations are focused on building.
Growth-stage organisations are trying to scale.
More mature businesses are often optimising or modernising, while others are trying to reinvent themselves altogether.

Each of those requires a different emphasis.

The business model matters too, and a useful way to think about this is whether the organisation primarily builds technology or uses technology. A product company will expect its CTO to lean heavily into engineering, product development, and platform decisions. An organisation that is primarily a user of technology will lean more towards architecture, integration, and ensuring that technology choices support business outcomes. A services organisation often sits somewhere in the middle, with an added emphasis on client engagement and external credibility.

Add in the organisation’s risk profile — particularly in regulated industries — and the balance shifts again towards governance, resilience, and control.

CTO and CIO

This is also where the distinction between CTO and CIO often comes into play, although the boundaries are not always neat. A CIO is typically focused on internal systems and operations — keeping the organisation running and enabling people to do their jobs effectively. A CTO is more often outward-looking, concerned with how technology shapes products, services, and the organisation’s position in the market.

In practice, many organisations blur or combine these roles, which is fine as long as the underlying responsibilities are understood. Problems tend to arise when the ownership of strategic technology direction is unclear, regardless of what title is used.

There’s a useful article from McKinsey that explores how the two roles differ and where they overlap, along with a lot of advice on different approaches to the CTO role.

It’s also worth noting that both roles have evolved. There was a time when CIO was jokingly expanded as “career is over”, reflecting a perception of the role as operational and inward-looking. That is no longer true. The CIO role has shifted towards business leadership, just as the CTO role has broadened beyond pure technology.

A role that keeps being reinvented

A friend once advised me that I shouldn’t seek to become a CTO because it was a “dead job title”. The implication was that the role had either peaked or would be absorbed into something else.

Happily, it turned out that my friend was wrong and, if anything, the opposite is true. As organisations rely more heavily on technology, the need for clear technology leadership has increased.

That said, it’s worth being precise about the different titles that appear around it. A Chief Digital Officer, for example, is often focused on digital channels, customer experience, and business model change. There can be overlap with a CTO, but it’s not simply another flavour of the same role. Similarly, titles like Field CTO or Customer CTO tend to describe where the role operates — closer to customers, sales, or market engagement — rather than a fundamentally different purpose. You will also see the term vCTO (Virtual Chief Technology Officer), which can refer to the same strategic capability delivered on a fractional or shared basis.

There’s also a practical question about the word “Chief”. In theory, there should only be one Chief Technology Officer. In practice, that’s not always the case. As Matt Watts (himself, a former Field CTO) put it in a recent LinkedIn post, “how many CTOs do you have?” — pointing out that some organisations end up with dozens of CTO-titled roles, each with different scopes and levels of influence.

It might feel messy, but it reflects how the role has evolved. What matters is not the purity of the title, but whether the responsibilities are clear and whether someone is accountable for setting direction and making the key technology decisions.

The titles may shift slightly, and new variations appear, but the underlying need remains.

Why the CTO role matters

Most organisations are not short of technology activity. There are always projects, initiatives, and new tools being introduced. What is often missing is coherence. Without a clear direction and someone accountable for the decisions that shape it, work can become fragmented. Individual teams may make sensible choices in isolation, but those choices do not always add up to a clear strategy.

This is where the value of a CTO becomes visible. It isn’t about knowing the technology better than anyone else. It’s about helping the organisation make better decisions about how technology is used, and ensuring those decisions are aligned with what the business is trying to achieve.

You can usually tell when that role is missing, even if the title is not. Repeated debates about priorities, uncertainty about where to invest, and a sense that there are lots of initiatives but no clear direction are all indicators. In those situations, the organisation doesn’t necessarily need more projects or more tools: it needs clarity, ownership, and a way of turning activity into something coherent.

That, in essence, is what the CTO provides.

ABBA Voyage and the future of live performance

markw@markwilson.co.uk (Mark Wilson) — Sun, 22 Mar 2026 00:00:00 +0000

ABBA last performed in London in 1979.

And yet, yesterday afternoon, I was in an East London arena watching them perform. Except they weren’t really there.

ABBA Voyage is not a reunion tour. It’s a new form of live performance, built on a blend of digital production, physical staging, and carefully engineered illusion.

Many reviews focus on the music and the spectacle, but that’s only part of the story. As a technologist, what makes ABBA Voyage notable is not that it recreates a band, but that it challenges what we mean by a “live” performance.

A mixed reality performance, not a hologram

ABBA Voyage is often described as a hologram show. That’s convenient, but not really accurate. It’s not just a projection of flat images onto a stage, nor a simple visual trick. The show combines motion capture, visual effects, real-time rendering, and a physical stage environment. The digital performers, known as ABBAtars, were created by Industrial Light and Magic in collaboration with the band. The original members spent weeks in motion capture suits, effectively re-performing their younger selves.

Alongside that, there is a live band on stage. The music is real. The lighting is real. The environment is real. The result is a mixed reality performance where digital and physical elements are presented as one.

The illusion is engineered

What makes this work is not just the quality of the digital characters. It is how the entire experience is designed to guide perception.

Lighting plays a central role, shaping what you can see clearly and what you cannot. It directs your focus and, at times, deliberately hides the edges of the illusion. The live band anchors the experience, giving your brain something that is unquestionably real and making it easier to accept what you are seeing.

I spent a good part of the show trying to work out what was real and what was rendered. That lasted until a moment where the illusion did something that felt impossible. I won’t spoil it, but it was enough to make me exclaim “Wow!” out loud.

As Benny Andersson himself puts it:

“I think the only way to understand what this is you have to come and see it. It’s sort of non-explainable […] you need to go and see it.” — Benny Andersson, ABBA

Solving the uncanny valley problem

Digital humans often fall into what’s known as the uncanny valley. Close to real, but not quite right. The ABBAtars mostly avoid that.

Presenting ABBA as their 1970s selves helps. It removes the expectation of a perfect, present-day likeness and allows for a slightly stylised interpretation. More importantly, the performances feel human. Movement, timing, and interaction are convincing enough that you engage with them as performers rather than as effects.

That is what makes the illusion sustainable over an entire show, rather than just impressive for a few minutes.

Why this matters beyond ABBA

It would be easy to dismiss ABBA Voyage as a one-off built around a globally recognised band. I see it as a proof of concept for a different model of live entertainment.

Performances are no longer constrained by the physical presence of the artist. Shows can run continuously in a fixed location. The experience is consistent and repeatable. Performances can be preserved, recreated, or reimagined.

From a technology perspective, this is what happens when film production, game engine thinking, and live event design converge. The business model is just as interesting, built around a permanent venue and a long-running production rather than a touring schedule.

Culturally, it raises a simple question: what does “live” actually mean?

A brilliant show, built on something bigger

It is worth saying this clearly: ABBA Voyage is a brilliant show.

(It also messed with my head a little.)

The technology is impressive, but it never feels like the point. It serves the performance rather than distracting from it. The music, the energy, and the atmosphere all land exactly as you would want them to. Without that, none of the rest would matter.

There is also a personal thread running through it. ABBA were one of the few bands that worked across generations when I was growing up. They were a constant on car journeys, and I remember saving Christmas money to buy their albums. I was too young to see them live.

And yet, somehow, I now have.

Sort of.

Where this goes next

ABBA Voyage is unlikely to remain unique for long. The underlying technologies are already well understood. What is new is how they have been combined, refined, and delivered at scale.

It is easy to imagine what comes next. Legacy artists returning without touring. Performances running in multiple locations. New acts designing shows that blend digital and physical elements from the outset.

This does not replace traditional live music. It does something different. It trades spontaneity for precision, and physical presence for creative control.

Closing thoughts

What ABBA Voyage demonstrates is not just that this kind of show is possible, but that it works as a format.

It shows that audiences are willing to accept a performance that is not physically live, as long as it feels live. That distinction matters because it opens the door to new ways of creating, distributing, and experiencing performance.

For decades, technology has supported live entertainment from the sidelines. This feels like a shift. Not technology enhancing the performance, but technology becoming the performance.

Credits

ABBAtars and performance image © Aniara Ltd, 2022 used under the fair dealing provisions of UK Copyright law.
ABBA curtain call image by Raph_PH via Wikimedia licensed under the Creative Commons Attribution 2.0 Generic license.

The true cost of complexity in modern cars

markw@markwilson.co.uk (Mark Wilson) — Fri, 20 Mar 2026 00:00:00 +0000

Today, I got my car back after a month-long repair. It started with a slightly unsettling experience when the steering failed. Not in a “that feels a bit heavy” sort of way, but completely. There was no ability to steer the car at all. The AA’s patrolman initially suggested I follow him to the garage until I pointed out that I literally couldn’t turn the wheels. (In the process of recovering my car, he also broke a connector on the electronic parking brake, but that’s another story.)

It got me thinking about how modern cars are put together, and what that means when something goes wrong.

From mechanical to electronic

Modern vehicles hide a lot of complexity beneath what still looks like a simple interface. In older cars, the steering wheel is connected to the wheels. It might be power-assisted, but the linkage is direct. If something wears out, you replace that part and move on.

On newer vehicles, the steering wheel is a controller - similar to one you might use with a driving game on a console. Turning the steering wheel is an input that is interpreted by a control unit, which then tells another system what to do. It works perfectly well until it doesn’t. The interface feels connected, but it really isn’t.

That pattern repeats across the car. Mechanical systems are now mediated by software and sensors. The added features improve the driving experience, but the complexity is easy to overlook until something fails.

Why small faults become big bills

When the steering failed on my car, it wasn’t a worn part. It was a complete unit — a sealed assembly. In this case, a £1600 component that wasn’t even in stock at Volvo’s central distribution centre in Gothenburg and took a few weeks to arrive.

The same applies elsewhere. What might once have been a minor repair can now mean replacing an entire assembly. On older cars, a worn suspension component might mean replacing a rubber boot. Now it is more likely to be a full suspension arm.

The effect is simple: faults that used to be small and local are now larger and more expensive to fix.

Complexity is not an accident

This is not poor design — it’s a deliberate consequence of how cars are built today.

Modern vehicles rely heavily on software and interconnected systems, managing everything from braking and stability to driver assistance features. Many of these features are expected or required, and they do make cars safer.

But each additional system brings more hardware, more software, and more potential failure points. When something breaks, it is rarely isolated, and it is usually expensive.

The cost of a middle-aged car

All of this adds up. The steering repair, labour, wheel alignment (tracking), and parking brake work have cost a sizeable sum on a seven-year-old car. Add in some tyres and brake pads (which are expected service items) and I had a significant bill shock.

All of this was on top of a timing belt replacement, front brakes and a major service a few thousand miles ago. In total, I’ve spent around £5000 on the car in the last 12 months. That’s roughly twice my annual budget for car maintenance.

At some point, I have to think about the sunk cost fallacy. But, for now, the plan is to keep it going, ideally long enough for a summer road trip across France, and then reassess.

Why the UK car market has shifted

For some time now, the UK car market has been built around finance, and servicing. Most new cars are acquired on PCP, HP or lease agreements — finance is used for the vast majority of new car purchases in the UK — and many are simply handed back at the end of the term rather than owned.

I bought my Volvo V60 nearly-new using a PCP and made the balloon payment to take ownership at the end of the agreement. Because the original list price was over £40,000, it still attracted the additional rate of vehicle excise duty for several years, even as a used car.

But £40,000 is no longer a luxury car. It means that running costs are increased for nearly-new cars, which is one of the reasons an all-in monthly charge looks more attractive.

That starts to make sense when you look at how cars behave in practice. When a single failure can cost four figures and take weeks to resolve, predictable monthly costs (e.g. leasing) and a warranty begin to look less like convenience and more like risk management.

Rethinking car ownership

Cars today are better than they have ever been. They are safer, more efficient and more capable. But they are no longer simple machines, and that has consequences.

Ownership now carries more uncertainty than it used to. Not because cars are worse, but because they are more complex, less repairable, and more expensive to fix when something goes wrong.

If the practical way to manage that risk is to hand the car back before anything expensive breaks, then it is worth asking a simple question:

Do we really want to own our vehicles, or just run someone else’s and give it back when the warranty ends?

Archiving my Twitter/X history

markw@markwilson.co.uk (Mark Wilson) — Tue, 17 Mar 2026 00:00:00 +0000

For many years I was an active user of Twitter/X @markwilsonit. I shared links, ideas, photos from events, commentary on technology and the industry, and some personal chit-chat too. Over time that added up to more than 70,000 tweets.

But the platform changed. Eventually it reached the point where continuing to use it no longer felt compatible with my values, so I stopped posting there, and made the tweets private.

Historical records

That left me with a problem though. Buried in those tweets is a lot of history – links to blog posts, conversations with people I respect, and snapshots of what was happening in the industry at the time. Twitter/X lets you download an archive of your data, but it also contains items that should remain personal to the user, like direct messages - so it can’t be published directly.

I wanted to build a small static website that turns the archive into something searchable and browsable that I could share more widely. There were some scripts around for converting Twitter archives to websites, but they didn’t seem to work for me when I tried a couple of years ago – maybe the archive format changed since they were written.

Vibe coding a script to create my own archive site

I ended up “vibe coding” (with ChatGPT) a small Python script that converts the Twitter data export into a static Twitter archive site. It’s intentionally simple: a static page; and a lightweight search index using JavaScript and JSON. All the tweets remain exactly as they were written (although with expanded URLs where t.co short links were in place), but now they’re accessible outside Twitter/X.

Rather than keep the script to myself, I’ve published it on GitHub so anyone else can do the same with their own data. If you’re curious about how it works – or you’d like to build your own archive – the code and documentation are available in this GitHub repository.

Just be aware that I downloaded my archive in August 2024 – Twitter/X archive formats have changed over time so, if it doesn’t work for you, you might need to tweak it.

Importing Eufy smart scale data into Garmin Connect

markw@markwilson.co.uk (Mark Wilson) — Sun, 15 Mar 2026 00:00:00 +0000

For years, most of my health and fitness data has ended up in one place: Garmin Connect. My watch syncs automatically. Steps, heart rate, sleep, runs and rides all flow into the same timeline. Activities are also synced from Garmin to Strava. There are a few bits and pieces in Apple Health (medications), but it’s pretty complete.

So when I bought some smart scales, I wanted the same thing. One place for all the data.

“Man maths”

The obvious option would have been scales from Garmin or Withings. But those are not cheap. Instead I bought a set from Eufy. Even my man maths couldn’t justify the extra expense.

The theory is that investing in technology to help manage my weight is worthwhile. The reality, of course, is that compulsive eating and weight management are much more complicated than that. But if a gadget nudges me in the right direction, I’m willing to try it.

The problem is that the Eufy app does not integrate with Garmin. So the data sits in its own ecosystem.

Fortunately, Garmin Connect allows body metrics to be imported from a CSV file. With a bit of spreadsheet wrangling, the Eufy export can be reshaped into something Garmin understands.

The format Garmin expects

Garmin’s import format is very simple. A CSV file like this:

Body
Date,Time,Weight,BMI,Fat
11/01/2026,07:22:16,75,25,22

Exactly like that – nothing extra, just those five columns. And the Body line seems to be essential – without it the import fails.

Preparing the Eufy export in Excel

Once you know that structure, the challenge becomes transforming the Eufy export into the same layout.

Start by exporting your data from the EufyLife app and opening the CSV file in Excel.
Delete the extra columns so you are left with Time,WEIGHT (kg),BMI,BODY FAT %.

The Eufy export stores date and time together in one field. Garmin expects them separately so we have to split the date and time.

Insert two new columns (B and C) after Time.
In cell B2, enter =INT(A2). Then format the column as Short Date (dd/mm/yyyy).
In cell C2, enter =MOD(A2,1). Then format the column as Time (hh:mm:ss).

Excel stores date and time as one number. The integer part represents the date. The decimal part represents the time. These formulas simply separate them.

Fill both formulas down the column for as many rows of data as you have.

Reshaping the file for Garmin

Rename the columns so they read DateTime,Date,Time,Weight,BMI,Fat.

The DateTime column is the original combined value exported by Eufy. We keep it temporarily while splitting the date and time.

Save the file (still in .CSV format) and close it.
Reopen the file. Excel will now have replaced the formulas with calculated values.
Delete the DateTime column.
Insert a new row at the top and enter Body in cell A1.
Delete any rows that already exist in Garmin Connect so the file only contains new entries.
Save the final CSV file. Your file should now look similar to this:

Body
Date,Time,Weight,BMI,Fat
14/03/2026,12:55:51,101.9,32.8,34.7
13/03/2026,10:12:27,103.4,33.3,35.4
12/03/2026,06:54:42,104.3,33.6,35.5

Importing into Garmin Connect

Open the Garmin import data page and upload the CSV file you prepared earlier.
The Garmin Connect site will recognise the data as Fitbit-style body metrics and ask how the values should be interpreted. Make sure you choose Kilograms for weight units and a date format of dd/mm/yyyy (31/12/2026), then click Continue.

The import should complete successfully.

The new entries should appear alongside the rest of your health data (Weight).

That all seems a bit of a faff, Mark!

It’s not quite the seamless integration I would have liked. But it works, for now. As I was writing this up, I thought I should probably script the CSV transformation. Maybe that’s the next step.

For now, at least I have my weight data in the same system as the rest of my fitness data.

Which feels like a small victory – even if the real challenge still happens in the kitchen, at the supermarket, at the petrol station, in the office and just about everywhere else.

It’s time to end “manels” at tech events

markw@markwilson.co.uk (Mark Wilson) — Fri, 13 Mar 2026 00:00:00 +0000

With International Women’s Day taking place last weekend, I’ve been reminded of a couple of industry events I attended late last year. I thought I’d posted on LinkedIn at the time about the poor female representation, but I can’t find those posts now. So while we’re celebrating women, it’s worth making the point again – because it’s still an issue.

The content at the events was… okay. One was a major IT channel/distributor event – and it was probably a bit too panel-heavy for me. The other event was run by a storage vendor and I was soon bored of tech product-and-feature talk (I only stayed for the excellent afternoon keynote). Both brought together people from a multitude of IT companies. Unfortunately, almost everyone on stage was male.

At the first event, it was well into the afternoon before the first female presenter took the stage. And the day was almost over before we saw a panel that wasn’t an all-male line-up.

The number of “manels” (all-male panels) was so noticeable that I started to keep a tally. At the end of the day the final speaker count was 30 men and 4 women. Thirty to four. In 2025. In a sector that claims to value progress and innovation.

At the second event, the only woman on stage had opened the event and quickly handed over to male colleagues. We didn’t hear a female voice again until after the lunchtime break.

To be fair, I’ve attended some really good events recently that have been far more balanced. That just proves it can be done. But those two stood out – for all the wrong reasons.

Yes, IT is still a very male-dominated industry. But that’s not an excuse for putting on events that simply mirror the imbalance. If anything, it’s a reason to work harder to widen the pool and to champion the diverse voices that are out there.

And to be clear: I’m quite happy to have my share of the limelight. But I’m a middle-aged, white, straight man. If someone else can bring a different perspective to the table, I’ll happily stand aside. We don’t move our industry forward by hearing the same voices over and over again.

There are organisations trying to change things. For example, techUK committed to ending “manels” back in 2017 and reaffirmed that commitment in 2020. For March 2026, they have a TechTogether campaign that is:

“Celebrating what organisations are doing to support diversity, equity, and inclusion, while acknowledging how much more needs to be done.”

And Thoughtworks has written about the importance of bringing diverse voices to tech conferences. But that was back in 2020 - and the events I described earlier make it clear that progress is still too slow.

So a simple ask for event organisers: try to make your events diverse. I know it’s not always easy. But if your speakers are mostly white men, and your panels are all “manels”, it might be worth reconsidering your line-up.

Events shape narratives, and narratives shape culture. If we want a more inclusive industry, we need to start by changing what – and who – we put on stage.

What exactly is a Frontier Firm?

markw@markwilson.co.uk (Mark Wilson) — Thu, 12 Mar 2026 00:00:00 +0000

Back in October 2025 I spent three days at Microsoft’s SME&C EMEA Sales and Partner Summit in Dublin. More than 3000 people were there — a mix of Microsoft staff and partners working across the Small, Medium Enterprise and Channel (SME&C) sector.

At the time I wasn’t entirely sure how much I was supposed to be sharing publicly, and I certainly didn’t get the chance to download any slides. Since then, Microsoft’s Ignite conference, Microsoft AI Tour, and a steady stream of other announcements have brought much of that thinking into the public domain.

One phrase I heard repeatedly during the event was clearly designed to stick in the mind:

“Frontier firm.”

It’s one of those terms that feels destined to appear in slide decks and conference presentations for the next few years. But behind the branding there is an interesting idea about how organisations might evolve as AI becomes embedded in everyday work.

Where the term comes from

The phrase didn’t originate at the summit. Microsoft introduced it earlier in its 2025 Work Trend Index report, which was titled “The Year the Frontier Firm Is Born”.

The idea is that organisations are moving towards a new operating model built around hybrid human-AI teams, where people increasingly direct AI assistants and agents to complete work.

In a blog post titled “Becoming a Frontier Firm: Unlocking the business value of AI”, Microsoft described the concept like this:

“A Frontier Firm is defined not by its size or industry, but by its mindset and execution. These organizations lead with AI-first differentiation, embedding intelligence across every layer of the business — from employee experiences to customer engagements to core processes. ‘Becoming Frontier’ means moving beyond experimentation to enterprise-scale transformation…”

In other words, it’s less about a particular technology and more about how organisations operate when AI becomes part of the normal flow of work.

AI as the defining technology of our time

The summit’s opening keynote set the tone clearly. AI, we were told, is the defining technology of our time.

That may sound like the usual technology conference rhetoric, but I think most people will agree that the scale of investment and innovation is staggering. Something significant is happening.

One message that comes through repeatedly is that AI is only as good as the data and the platform behind it. That’s not unique to Microsoft. It’s a point we hear across the industry, and it’s probably one of the biggest barriers to meaningful AI adoption. Many organisations are discovering that their data is fragmented, poorly governed, or simply not ready to support large-scale AI use.

Another challenge is choosing the right starting point. Successful AI projects tend to begin with a clearly defined business problem — something that can be measured and improved — rather than with the technology itself.

Unsurprisingly, Microsoft’s framing focuses heavily on the platform and the tools. After all, that’s what Microsoft sells.

But the underlying message still holds: without strong foundations in data, platforms, and clear use cases, AI initiatives rarely move beyond experimentation.

The path to becoming a Frontier Firm

At the summit, Microsoft described the journey toward a Frontier Firm as a progression — from humans with assistants, to human-agent teams, and ultimately to human-led, agent-operated organisations.

Stage	Description
Human with assistant	Employees use AI assistants embedded in everyday tools. Copilot (other AI assistants are available) helps draft documents, summarise meetings, analyse information and automate routine tasks. Humans remain firmly in control.
Human-agent teams	Specialised AI agents begin handling parts of workflows. Humans still provide direction and oversight, but agents gather information, analyse data and trigger actions across systems.
Human-led, agent-operated	Humans set strategy and intent while AI agents execute much of the operational work needed to deliver outcomes.

Whether organisations will actually operate this way remains to be seen. But it’s a useful way of describing the direction Microsoft believes AI-powered organisations are heading.

Microsoft also outlined a success framework for organisations becoming Frontier Firms. Rather than focusing purely on technology, the framework highlights four areas where AI is expected to reshape how organisations operate:

Enrich employee experiences
Reinvent customer engagement
Reshape business processes
Bend the curve on innovation

These themes reflect where Microsoft sees AI having the greatest impact across organisations — not just in productivity tools, but in how businesses interact with customers, design processes, and innovate.

Data foundations

We’ve already established that good data is essential for successful AI adoption, so it’s no surprise that this is one of the areas Microsoft spent a lot of time talking about.

The problem is easy to understand. Most organisations didn’t design their data environments for AI. Systems have grown organically over many years, often across multiple platforms, acquisitions, and generations of technology. Data ends up scattered across applications, databases, file stores, and analytics tools — often with inconsistent governance and visibility.

Microsoft’s answer to that challenge centres on Microsoft Fabric. Fabric is the platform designed to bring together analytics, engineering, integration and governance capabilities in a single environment.

Underneath Fabric sits OneLake — a SaaS-based data lake intended to act as a single, unified data store for the organisation.

The goal is to reduce fragmentation and make data accessible across analytics, applications and AI workloads. In Microsoft’s view, that kind of unified data estate is what allows organisations to move from isolated AI experiments toward something that can operate at scale.

The platform underneath the AI story

Behind all the talk of AI sits something more familiar: the cloud platform.

Microsoft currently describes its portfolio in three broad areas — AI business solutions, cloud and AI platforms, and security — built on what it calls a trusted and secure foundation.

Microsoft Azure has matured significantly since the early days of Windows Azure more than fifteen years ago. Today it provides the infrastructure, development platforms, data services and AI capabilities that underpin much of Microsoft’s technology stack.

Developers also play a central role in this story. Tools like Visual Studio and GitHub sit at the heart of how organisations build applications, integrate systems and increasingly create AI-powered solutions.

In other words, while the conversation may focus on Copilot and agents, the underlying platform — cloud, data and developer tooling — is what makes those capabilities possible.

Trust becomes critical

If AI systems are going to operate across business processes, trust becomes even more important.

AI can surface information that previously sat quietly in documents, emails and collaboration platforms. That creates new challenges around governance, security and compliance.

Identity, data protection and threat detection therefore move from the background to the centre of the conversation.

Security is no longer something that sits alongside AI. It becomes part of the foundation that makes AI possible.

Microsoft’s answer to that challenge sits in its broader security and identity platform — services like Entra for identity, Purview for data governance and compliance, and Defender and Sentinel for threat protection and monitoring — alongside endpoint management through Intune.

From experimentation to transformation

The shift Microsoft is describing with the idea of a Frontier Firm isn’t about a single product or capability. It’s about how organisations operate when AI becomes part of everyday work.

That means embedding AI into real business processes, supported by strong foundations in data, cloud platforms and security.

So what exactly is a Frontier Firm?

In simple terms, a Frontier Firm is Microsoft’s way of describing organisations that embed AI deeply into how they operate.

Employees work alongside AI assistants. Agents begin to handle parts of workflows. Human judgement focuses more on direction and decision-making, while systems take on more of the operational work.

Whether the phrase itself will stand the test of time is another question. Technology marketing has a habit of introducing new labels every few years.

But the underlying idea is straightforward enough: organisations using AI not just as a tool, but as part of how work gets done.

“AI Eats the World”

markw@markwilson.co.uk (Mark Wilson) — Wed, 11 Mar 2026 00:00:00 +0000

At the recent NashTech Connect event, Benedict Evans delivered a keynote titled “AI Eats the World”.

Benedict (Ben) regularly publishes large presentations exploring long-term trends in the technology industry. As he describes it on his website: “Twice a year, I produce a big presentation exploring macro and strategic trends in the tech industry.”

“AI Eats the World” is Ben’s current presentation.

What I liked about this talk is that it combines several familiar themes into a clear narrative, illustrated with historical reference points.

Platform shifts

The talk opened with the idea that the technology industry moves in platform shifts.

We know the pattern by now. Mainframes gave way to PCs, then the web, then smartphones. Each time, the old world didn’t disappear, but innovation, investment and company creation moved to the new platform. New gatekeepers emerged. Old gatekeepers lost their grip. New markets appeared, old ones were reshaped, and some things were simply destroyed.

That matters both inside and outside the technology industry. Inside tech, platform shifts change which companies matter. Outside tech, every business ends up asking the same questions: is this just a new tool, a new source of revenue, or an existential threat? How much attention does it deserve?

Microsoft is a good example. When the PC sat at the centre of the industry, Microsoft dominated. When the centre shifted to smartphones, Microsoft became much less relevant for a period because it was no longer central to the platform that set the agenda. After missing the smartphone revolution, Microsoft rebuilt its position through cloud, but with far less dominance than it once had in the PC era.

[I personally witnessed Microsoft’s drive to push Azure as the company attempted to prevent Amazon Web Services (AWS) from gaining market dominance.]

Ben also pointed out that the first companies into a new market are rarely the ones that capture most of the long-term value. That was true for PCs, browsers, search, social media and smartphones. It may well prove true for generative AI too.

One consequence of this is that a lot of what surrounds a platform shift will turn out not to matter. There will be noise, hype, dead ends and ideas that look important for a while before fading away. That is normal.

Bubbles, hype and uncertainty

Platform shifts often bring bubbles with them.

People draw straight lines on charts, talk about exponential growth and insist that this time everything is different. In one sense they are right. Every major technology wave is different from the last one. But that doesn’t mean it can’t still be a bubble.

After the noise clears, the world has usually changed. The technology becomes woven into everyday life. It stops feeling novel and starts feeling normal.

That is why platform shifts matter even when the hype gets ahead of reality.

There is one important difference this time, though. With previous platform shifts we broadly understood the physical limits. We knew what phones couldn’t do. We knew what networks couldn’t support. With AI, we don’t really know why these models work as well as they do, which makes it harder to judge how far they might go.

That means the range of possible outcomes is unusually wide. This may turn out to be “only” as significant as the Internet, the PC or the smartphone. Or it may be something more fundamental. We simply don’t know yet.

Capital is flooding in

From there, the keynote moved inside the technology industry to look at capital, starting with two quotes from industry leaders:

“The risk of under-investing is significantly greater than the risk of over-investing.”
– Sundar Pichai

“The very worst case would be that we have just pre-built for a couple of years.”
– Mark Zuckerberg

That belief is driving extraordinary spending, apparently based on a fear of missing out (FOMO). The biggest platform companies are pouring vast sums into infrastructure. NVIDIA can’t keep up with demand. Infrastructure suppliers often benefit first in platform shifts – Ben compared the moment to earlier cycles where companies like Sun Microsystems rode the wave of new computing platforms.

The power industry is struggling to keep pace too. In some cases, access to electricity is becoming a bigger constraint than access to chips, raising questions about power grids, infrastructure planning and the wider political implications of AI’s energy demands.

“It’s been almost impossible to build capacity fast enough since ChatGPT launched.”
– Kevin Scott, Microsoft CTO

The scale is enormous. Hundreds of billions of dollars are being committed to data centres and associated infrastructure, with more to come. Some of that is funded from the enormous cash flows of highly profitable companies. Some increasingly depends on leasing, borrowing or other financial engineering as firms try to build faster than their balance sheets might naturally allow.

Ben illustrated how different this is from earlier eras of software economics. Microsoft’s historic model was effectively a one-dollar CD in a ten-dollar box sold for hundreds of dollars. Software used to be extraordinarily capital-light, and often benefited from powerful network effects that created dominant platforms.

AI infrastructure looks very different. The frontier keeps moving forward, but each step is more expensive than the last. That raises a difficult question: how long do you keep chasing the frontier?

If models continue to converge in capability, the risk is that they become commoditised.

More models, but not much separation between them

The investment boom has produced rapid progress. Models keep improving. New ones appear constantly. There are proprietary models, open-source models, sovereign models and a growing alphabet soup of acronyms.

But the leading models are increasingly clustered together in capability. On many benchmarks they sit within a relatively narrow range of one another. There may be a new leader every few weeks, but there is not much long-term separation.

Where there is separation is in distribution. OpenAI has broken through into public awareness in a way that others have not. Some competitors may perform just as well on benchmarks, but they don’t have the same reach.

The ambition for companies like OpenAI is to become the universal platform that others build on – the equivalent of Windows for the AI era. But the market may end up looking more like cloud computing, where organisations mix and match components from several providers.

That raises the possibility that AI models evolve more like the semiconductor industry than traditional software – capital intensive, expensive to build, and dominated by a small number of players rather than a single winner.

Models are improving quickly. What remains much less clear is where the long-term value will sit.

Most organisations are still in the “absorb” phase

Outside the technology industry, the pattern tends to be different. The difficulty is that no-one really knows what will work yet.

New technologies usually follow a sequence: first organisations absorb them. They turn them into features and automate the obvious tasks they already understand. Only later do new products, new revenue streams and real disruption appear.

The absorption phase is where most generative AI activity still sits today.

The obvious early use cases are software development, marketing, customer support and point solutions inside large organisations. In many of these cases the change is incremental but still useful: software becomes faster to build, new capabilities appear, and there are tasks where AI simply performs better than previous tools.

Part of the challenge is that generative AI behaves very differently from traditional software. It is good at things computers have historically been bad at – language, summarisation and pattern recognition – and bad at things computers have traditionally done well, such as precise calculation.

Even usage numbers can be misleading. Many people who have access to these tools don’t use them constantly. They use them occasionally, perhaps weekly or monthly, and often only for a narrow range of tasks. Even among those using them, relatively few are paying customers, and usage is often light.

Ben compared this to early web analytics. Page “hits” once looked impressive until people realised that every image on a page counted as another hit. The numbers sounded large, but they didn’t necessarily mean very much.

Many organisations are still running pilots, experimenting with use cases and trying to work out what this technology should actually do for them. The hard part is rarely writing the code. It is working out what the actual problem is, and who would use the solution.

“People don’t know what they want until you show it to them. You’ve got to start with the experience and work backwards to the technology” – Steve Jobs

Automation comes first. Change follows later.

Ben used spreadsheets as an analogy. Before VisiCalc, spreadsheets were literally sheets of paper. When VisiCalc appeared on the Apple II in 1978, it cost the equivalent of around $10,000 in today’s money.

But it could turn a project that took a week on paper into something that could be recalculated in minutes. For accountants, that was transformative. For lawyers, who rarely build financial models, it was far less interesting.

The first step was automation. The deeper changes came later.

Barcodes offer another example. Initially they were introduced to reduce the effort involved in stock-taking. But, once deployed across supermarkets, they enabled far more stock-keeping units (SKUs) to be managed and eventually reshaped the entire supply chain, with automated inventory systems and electronic reordering.

That leads to the next question: once we have automated the obvious things, what comes next?

What changes when you have “infinite interns”?

Ben framed the next phase with a thought experiment: what happens if AI gives you the equivalent of “infinite interns”?

[At this point I was reminded of Chris Weston’s 100 Homer Simpsons analogy.]

Do you do the same work with fewer people? Or do you do far more work with the same number of people?

This is similar to what happened in earlier industrial revolutions. Steam engines effectively gave nineteenth-century Britain the equivalent of millions of additional labour units. The question now is what generative AI might do as a productivity multiplier for knowledge work.

The deeper question is whether the tasks we are automating were ever the real job in the first place. Spreadsheets automated calculations, but the goal was never the spreadsheet itself – it was better financial understanding and decision-making.

That leads to a broader question: were we really trying to automate tasks, or were we trying to achieve something else entirely?

Recommendations, discovery and agentic commerce

This question becomes especially interesting when we look at recommendation and discovery online.

Before the Internet, human editors decided much of what people saw – in shops, newspapers and broadcasters. Internet platforms scaled that process through algorithms.

Platforms like Amazon often operate through correlation rather than deep understanding. They know that people who bought one SKU often buy another, even if the system doesn’t really understand the products themselves.

LLMs may create a different kind of recommendation layer. They may operate at a higher level of abstraction, interpreting intent rather than simply correlating behaviour.

There’s also a long-standing usability principle, often attributed to Bruce Tognazzini: a computer should never ask a user a question if it can work out the answer itself. Systems that understand intent more deeply may reduce the amount of searching, filtering and form-filling users currently have to do.

That could lead to very different buying journeys – asking an assistant what to buy rather than browsing a catalogue or search engine.

The closing perspective

There’s an old line that history doesn’t repeat, but it often rhymes. Technology waves tend to follow familiar patterns.

All the technologies people were excited about before generative AI are still there. E-commerce continues to grow. Autonomous vehicles are becoming more real. Smart glasses, robotics and other long-running ideas haven’t gone away.

That broader perspective matters because technology change is cumulative. New waves arrive before previous ones have fully played out.

Looking further back reinforces the point:

An IBM advert from the 1950s promised that an electronic calculator would deliver the equivalent of 150 engineers.

A US government report from 1955 discussed the coming age of automation, including lifts (elevators). At the time many lifts required attendants to operate them. Then they were automated. Eventually people forgot that they had ever worked any other way. When automation works well, we stop noticing it.

Ben noted that image recognition once felt like something close to magic. Fifteen years ago it seemed almost impossible. Today it is a routine feature inside everyday software.

That brings us to Larry Tesler’s famous line:

“AI is whatever machines can’t do yet.” – Larry Tesler, 1970

Once it works, we stop calling it AI. It simply becomes software. Infrastructure. Just another ordinary part of how things work.

When design forgets the real world

markw@markwilson.co.uk (Mark Wilson) — Tue, 10 Mar 2026 00:00:00 +0000

Design trends come and go, a bit like fashion.

At the moment, one of the prevailing styles seems to be light text on dark backgrounds. On screen it can look clean and modern. Used well, it is perfectly readable too.

But sometimes design choices are made with the screen in mind and not much thought for what happens next.

The dark background problem

A good example appeared in my inbox recently. I was getting ready for attending the ServiceNow AI Summit and, as with many events, the organisers published the agenda as a graphic on their website.

That is already not ideal from an accessibility perspective. Text embedded in an image is not searchable, scalable, or friendly for screen readers.

There is also a practical problem because the agenda uses light text on a dark background. That looks fine on screen, but becomes less practical if printed.

Suddenly you are printing what is essentially a large black rectangle with white text on top. Toner disappears quickly. And if you are like me, the printer available is a perfectly serviceable black and white laser device. Great for documents. Less good for printing a mostly black page.

A slightly odd workaround

So, I took the agenda graphic and inverted it, creating a negative version which had dark text on a light background. On screen it looks dreadful, but on paper it works perfectly well.

More importantly, it doesn’t try to empty the toner cartridge in one go.

It is a small example of what happens when something is designed purely for screens without thinking about what people might do next.

Why print it at all?

You might reasonably ask why I printed the agenda.

Because I wanted to plan my day and, without an event app, I took the analogue approach. I printed the schedule and started circling sessions, and scribbling notes.

It is simple, quick, and surprisingly effective because paper is still a very good interface for thinking.

Of course, I could probably solve this with an iPad and a stylus. But that feels like rather a lot of electronics just to draw circles around a agenda items. A pen and a sheet of paper works perfectly well.

Design for how things are used

None of this is really about dark backgrounds. They can work very well in the right context.

The issue is when design is optimised for appearance rather than use.

And the problem is not limited to conference agendas. I’ve seen similar with airline boarding passes surrounded by dark-coloured ads on an A4 PDF. I’ve also experienced event tickets made available in PDF format but using dark designed graphics.

If something might reasonably be printed, a print-friendly version helps. If something contains structured information like a conference agenda, publishing it as text instead of a graphic helps even more.

And if a design choice accidentally turns a simple print job into a toner-consuming black rectangle, that is probably a small sign that the design process has drifted a little too far from real-world use.

Sometimes the most useful design choice is simply remembering how people actually use things.

AI outcomes, not AI hype: reflections from a PyData panel

markw@markwilson.co.uk (Mark Wilson) — Thu, 5 Mar 2026 00:00:00 +0000

A few weeks ago I joined a panel at the PyData Milton Keynes meetup.

PyData is a global community of developers, data scientists and engineers working with open-source data tools. The Milton Keynes chapter only began running events in 2025, and this was its first in-person gathering.

The panel was hosted by Grace Farayola. I was joined by Samantha Roberts from SDG Group, Madara Premawardhana from the University of Buckingham, and James Graham, former CEO of BsoftB.

The evening’s theme was FutureTech 2026: innovations that will shape tomorrow.

Ahead of the event Grace shared a set of questions that she intended to ask the panel. The discussion covered a lot of ground, but I’d prepared a few thoughts in advance.

Sadly, I don’t have a record of the great insights that my fellow panellists shared (it’s pretty hard to take notes when you’re on the panel). Instead, this is a short reflection on the topics I was asked to cover, based on the answers I prepared beforehand.

The future of tech is not just about “the shiny stuff”

When asked what the future of technology means from my perspective, my answer was fairly simple.

The future of tech is outcome-led: secure platforms, trusted data and modern applications solving real problems, with AI built in so value can be delivered safely and at scale.

It’s not particularly glamorous.

There are plenty of exciting innovations and bold ideas in the industry right now. But most organisations are still trying to get the fundamentals right while staying responsive to their customers, clients or citizens.

That may not make headlines, but it’s where progress really happens.

AI is becoming a core operating layer

The more interesting question is what organisations should prioritise in 2026.

The answer isn’t “more technology”. It’s the technology that turns AI into measurable outcomes.

AI has moved beyond experimentation. It is becoming part of the operating fabric of organisations: a productivity multiplier, a margin lever and a differentiator for customer experience.

That puts CTOs firmly in the hot seat.

Technology leaders are increasingly expected to turn AI into real business impact, not innovation theatre.

The foundations that make this possible are not new:

Trusted, unified data.
Modern, composable platforms.
Secure and resilient infrastructure.
Experience-led technology that people actually adopt.
Sustainable architectural choices.
Cross-functional teams that can deliver outcomes.

Put simply: AI only works when the foundations are solid.

Turning experiments into outcomes

Many organisations are still experimenting with AI and cloud technologies.

The challenge is moving from experiments to measurable value. The simplest way to do that is to start with the outcome: what does good look like? (Remember Stephen Covey’s “Seven Habits of Highly Effective People”, where the second habit is “begin with the end in mind”.) In other words, decide which metric you want to change, then work backwards.

A practical playbook looks something like this:

Start with the outcome – which metric will move?
Prepare the data – quality, lineage and access.
Prepare the platform – APIs, automation and observability.
Put guardrails in place – security, governance and responsible AI.
Deliver a thin slice – prove the value quickly.
Scale what works.

It’s often said that a very high percentage of AI projects fail (usually quoting various analysts and academic institutions). It’s not just AI projects either. Many initiatives struggle because they start with the technology rather than the outcome.

If you can’t explain the impact in a sentence, the project probably needs more thinking.

Innovation and compliance are not opposites

One of the debates during the panel was whether organisations can realistically be both innovative and compliant.

I think they can.

Innovation and compliance are often presented as competing priorities (“while America innovates, Europe regulates”), but that’s the wrong way to think about it.

It’s more like driving a car. The brakes and steering don’t slow you down. They allow you to move faster, safely.

Technology governance works the same way. When guardrails are built into the design from the start, they enable innovation rather than restrict it.

The real shift for technology leaders

The biggest change we are seeing is not technical. It’s organisational.

AI outcomes are becoming career-critical for technology leaders.

In many organisations, AI is no longer a side experiment. It is becoming a core part of how the business operates, influencing productivity, margins and customer experience.

That means the expectations placed on technology leaders are changing.

Just as CFOs are held accountable for financial performance, CTOs will increasingly be judged on whether AI initiatives move the business forward.

Not demos. Not pilots. Not slideware. Real business outcomes.