March 2025

The artificial intelligence revolution is no longer creeping forward—it’s sprinting. In my recent conversation with Brett Roscoe, Senior Vice President and General Manager of Data Governance and Cloud Operations at Informatica, we explored the findings from their CDO Insights 2025 report that reveal both the explosive growth in AI investment and the obstacles companies are encountering on their journey.

What struck me most during our discussion wasn’t just the sheer scale of financial commitment—with 87% of companies increasing their generative AI investments in 2025—but the widening competitive chasm between early adopters and those hesitating to embrace AI’s transformative potential.

The Investment Stampede Is Real

The enthusiasm for generative AI isn’t just talk—it’s backed by serious capital. As Roscoe explained during our conversation, “87 percent of our customers… are increasing their investments in gen AI in 2025. Now that was already on top of what they’ve done to increase it in 2024.”

This investment surge isn’t happening in isolation. Alongside their AI spending, 86% of organizations are simultaneously boosting their investment in data management practices to support these initiatives. This dual-track approach signals a growing recognition that successful AI implementation requires robust data infrastructure.

What are companies hoping to achieve? According to Roscoe, “the biggest areas were operational efficiency… and enhancing customer experiences or employee experiences.” These goals are already materializing in companies like CN Rail, which has slashed data preparation time from months to just two weeks, creating the foundation for accelerated AI project development.

The Early Adopter Advantage Is Widening

Perhaps the most compelling insight from our conversation was the growing evidence that AI early adopters are pulling ahead—and fast. As Roscoe noted, “Early adopters of AI had a big advantage. In fact, to the point where these early adopters saw something like 35 percent cost decreases and 58 percent growth attributed to their Gen AI projects.”

This isn’t incremental improvement—it’s transformative change. When McKinsey surveyed organizations already deploying generative AI, they found that, on average, these companies attributed 20% of their EBITDA to their AI initiatives. Numbers like that aren’t just competitive advantages—they’re existential threats to laggards.

I’ve observed this dynamic firsthand across industries: the gap between AI leaders and followers isn’t stable—it’s expanding daily. Every organization that doesn’t see itself as AI-enabled is falling further behind those that do. This creates a powerful incentive for companies to push ahead despite challenges.

The Hidden Obstacles To AI Success

While investment enthusiasm runs high, implementation reality is more sobering. The CDO Insights 2025 report revealed that a staggering 97% of organizations struggle to demonstrate business value from their generative AI investments—a significant roadblock to securing continued funding and executive buy-in.

Technical challenges are equally prevalent, with 92% reporting that issues like data quality, responsible AI use, and compliance concerns are delaying the progression from proof-of-concept to production. As Roscoe emphasized, companies are grappling with questions like: “How will this AI model behave? What risks are associated with putting this into production?”

In my experience working with organizations across sectors, I’ve observed additional barriers: executive teams not thinking ambitiously enough, scattered pilot projects without strategic cohesion, leadership knowledge gaps, skills shortages, cultural resistance, and data silos that prevent AI from reaching its potential.

According to recent research, 67% of business leaders believe that generative AI will bring significant change to their organizations over the next two years.

But in the rush to adopt and deploy this world-changing technology, it’s pretty likely that mistakes will be made.

The downside of this enormous potential is that when things go wrong, the damage can be quite serious too, from reputational harm to harsh fines and, perhaps worst of all, loss of customer trust.

So here’s my overview of five of the most common mistakes that I believe many businesses and business leaders will make in the coming year so you can plan to avoid them.

Omitting Human Oversight

Powerful and transformative as it undoubtedly is, we can’t ignore the fact that generative AI isn’t always entirely accurate. In fact, some sources state that factual errors can be found in as many as 46 percent of AI-generated texts. And in 2093, the tech news website CNET paused the publication of AI-generated news stories after having to issue corrections for 41 out of 77 stories. What this means for businesses is that proofreading, fact-checking and keeping a human-in-the-loop is essential if you don’t want to run the risk of making yourself look silly.

Of course, humans make mistakes, too, and any business involved in information exchange should have robust procedures for verification regardless of whether they use generative AI or not.

Substituting GenAI For Human Creativity And Authenticity

Another mistake I am worried we will see far too frequently is becoming over-reliant on genAI as a substitute for human creativity. This is likely to have negative consequences on the authenticity of a business or a brand voice. While it’s easy to use ChatGPT or similar tools to churn out huge volumes of emails, blogs, social media posts and suchlike super-fast, this frequently leads to overly generic, uninspiring content that leaves audiences feeling disconnected or even cheated. Video game publisher Activision Blizzard, for example, was recently criticized by fans for using “AI slop” in place of human-created artwork. It’s important to remember that generative AI should be used as a tool to augment human creativity, not to replace it.

Failing To Protect Personal Data

Unless a generative AI application is run securely on-premises on your own servers, there’s often no real knowing what will happen with the data entered into it. OpenAI and Google, for example, both state in their EULAs that data uploaded to their generative chatbots can be reviewed by humans or used to further train their algorithms. This has already caused problems for some organizations – Samsung stated that its employees had inadvertently leaked confidential company information by entering it into ChatGPT without being aware of the consequences. Incidents like this create a risk for companies that they will end up in breach of data protection regulations, which can lead to severe penalties. This is likely to be an increasingly common occurrence as more and more companies start using generative AI tools, and organizations – particularly those that handle personal customer data at scale – should ensure staff are thoroughly educated about these dangers.

Overlooking Intellectual Property Risks

Many commonly used generative AI tools, including ChatGPT, are trained on vast datasets scraped from the internet, and in many cases, this includes copyrighted data. Due to the lack of maturity in AI regulations, the jury is still out on whether this constitutes a breach of IP rights on the part of AI developers, with several cases currently going through the courts. The buck might not stop there, however. It’s been suggested that businesses using genAI tools could also find themselves liable at some point in the future if copyright holders manage to convince courts that their rights have been infringed. As of now, failing to assess whether AI-generated output could contain copyright or trademark-infringing materials is likely to land businesses in hot water in 2025, if they aren’t taking proactive measures to make sure it doesn’t.

Not Having A Generative AI Policy In Place

If you want to minimize the chances that anyone working for your organization makes any of these mistakes, then probably the best thing to do is to tell them not to. The potential use cases for genAI are so varied, and the opportunities it creates are so vast that it’s almost certainly going to be misused at some point. Perhaps the most important single step you can take to reduce the chance of that happening is to have a clear, defined framework in place setting out how it can – and can’t – be used.

As far as I’m concerned, this is a no-brainer for every organization that stops short of a blanket ban on generative AI, which would be a pretty big mistake, given the opportunities it creates. Without such a policy in place, you can almost guarantee it will be used without appropriate oversight, overused to the detriment of human creativity, and lead to unauthorized disclosure of personal data, IP infringement, and all the other mistakes covered here.

To wrap up – in 2025, we will see organizations take huge steps forward as they become increasingly confident, creative and innovative in the way they use generative AI. We will also see mistakes. Being fearful of the transformative potential of generative AI will most likely hand the lead to our competition, but adopting a careful and cautious approach can save us from costly mistakes.

Quantum computing is shaping up to be among the most transformative technologies of our era.

While still in their infancy, these powerful machines are expected to help us solve many problems by accelerating the speed at which we can process certain types of data by a factor of hundreds of millions.

But not all quantum computers are the same. Researchers are working on many different ways to apply principles of quantum mechanics to computing technology. This has led to a variety of methods, architectures and paradigms, all suited for different use cases or tasks.

So here I’ll overview some of the different categories, giving a brief explanation of what makes each one unique and what it’s hoped they will achieve.

First, What Is Quantum Computing?

Just in case you’re completely new to the topic – quantum computing refers to a new approach to computing that harnesses some of the strange and powerful properties of quantum mechanics, such as entanglement and superposition. Instead of using traditional “bits” (ones and zeros) like a classical computer, quantum computers use “qubits” that are spookily able to exist in more than one state simultaneously. This means they can potentially solve some very complex mathematical problems, such as those involving optimization problems or simulating complex real-world systems like molecular physics – far faster than existing computers.

So What Are The Different “Types” Of Quantum Computers?

Several distinct quantum computing methodologies have emerged, each leveraging quantum properties in different ways, making them suitable for carrying out different types of computation. Here’s an overview of some of the most popular:

Quantum Annealing

This is a quantum computing methodology that’s particularly well-suited to solving optimization problems. These are computations that require finding the best combination of a large number of variables. It can be of use in real-world scenarios ranging from planning the most efficient route for multi-drop delivery drivers to optimizing stock portfolios. D-Wave is recognized as a leader in this field of quantum computing and has worked with companies, including Volkswagen, to create systems that use annealing methodology to optimize assembly line packaging operations and delivery logistics.

Superconducting Quantum Computers

One of the most mature quantum computing methods involves building circuits from superconductive materials such as niobium or aluminum, cooled to near absolute zero temperatures. This allows qubits to exist in superposition states of both one and zero simultaneously, where they can be manipulated by microwaves. In simple terms, this lets them carry out computational logic operations (and/or/not etc) in a way that lets them explore multiple possible solutions to a problem in parallel, rather than one at a time. Superconductive quantum computing is being pioneered by companies such as IBM and Google and has real-world applications in drug discovery, artificial intelligence, and encryption.

Trapped Ion Quantum Computers

This involves using positively charged atoms (ions) trapped and held within a 3D space in a way that entirely isolates it from the outside world. This means that it can be held in its superposition state for a very long time rather than decohering into one or zero. Lasers are used to switch the ions between different states as required for calculations, as well as to retrieve the information that forms the “answer” to the question that needs to be solved. Leaders in this field of quantum computing include IonQ, which has worked with the United States Air Force to create secure quantum networking technology for communicating between drones and ground stations.

Photonic Quantum Computers

This involves harnessing photons, which are light waves, and manipulating them using optical components like beam splitters, lenses and mirrors. Having no mass, light waves are not affected by temperature. This means that photonic quantum computing doesn’t require super-low temperatures and a specially configured environment. Another benefit of being light beams is that the qubits encoded in photons can maintain their coherence over relatively long distances. Real-world applications for it have been found in quantum cryptography and communications, and leaders in the field include Xanadu.

Where Next For Quantum?

Although real-world use cases for quantum computing are increasing, much of the work in the field is still purely hypothetical, and various other methods are under development in labs and academic institutions.

Other research is focused on reducing the error rate of quantum computing caused by the delicate nature of qubits held in a quantum state.

It’s also worth noting that most quantum computing taking place today involves a hybrid model of quantum and classical methodologies.

As research and development continue, there’s no doubt we’ll start to see more breakthroughs in the journey towards practical, scalable and useful quantum computing.

AI promises to make businesses of all shapes and sizes more efficient, more innovative, and better prepared to deliver whatever their customers want.

However, it will also inevitably expose them to new dangers and risks.

In fact, the risks of implementing new technology often scale in line with the potential for positive transformation – and with AI, that potential is huge.

So here are some of the biggest mistakes that I believe businesses will make this year as they race to implement it:

1. Failing To Align AI Strategy With Business Strategy

One of the mistakes I see many businesses making is rushing to adopt AI simply because they’ve been told they should do so without understanding how it fits with their overall goals.

This approach – putting technology before business strategy – is probably the number one driver of failed AI initiatives and – worse still – disillusionment and giving up on AI altogether.

It’s easy for us to get so excited by the hype around AI that we rush out and start spending money on tools, platforms and projects without aligning them with strategic goals and priorities. This inevitably leads to fragmented initiatives that fail to deliver meaningful results or ROI. To avoid this, always “start with strategy” – implementing a strategic plan that clearly shows how any project or initiative will progress your organization towards improving the metrics and hitting the targets that will define your success.

2. Underestimating The Impact Of AI On The Workforce

Implementing AI is likely to mean big and possibly frightening changes for the people who make your business tick.

While it won’t mean they all become redundant, it may mean they need to learn new skills or adapt to new ways of working. Assessing the skills and possibilities of training or reskilling, ensuring there is buy-in across the board, and addressing concerns people might have about job security are all critical.

Many businesses will make the mistake of thinking solely about the technical steps they have to take while forgetting that humans will still ultimately be responsible for success or failure – and this is a serious error.

3. Giving Up Too Quickly – Or Too Slowly

Not every AI project is going to work – in fact, recent Gartner research puts the current failure rate of AI initiatives at around 85%. But getting it wrong the first (or second, or third) time isn’t necessarily a reason to give up. I firmly believe that just about any business can benefit from AI, but that isn’t the same as believing that they will benefit right away or that all they have to do is launch an AI project and they’ll immediately be successful.

On the other hand, being slow to pull the plug on projects that aren’t working out can also be a recipe for disaster – potentially turning what should simply be a short, sharp lesson into a long-term waste of time and resources. There’s a reason that “fail fast” has become a mantra in tech circles. Projects should be designed so that their effectiveness can be quickly assessed, and if they aren’t working out, chalk it up to experience and move on to the next one.

4. Failing To Properly Assess The Cost

Like the previous trap, this one is also two-pronged – it can be easy to both overestimate and underestimate the cost of developing AI initiatives, and both can be problematic.

Make no mistake, going full-throttle on AI is expensive – hardware, software, specialist consulting expertise, compute resources, reskilling and upskilling a workforce and scaling projects from pilot to production – none of this comes cheap. The question of how much it costs to get an enterprise AI initiative off the ground is similar to “How long is a piece of string?” but most estimates go into the millions.

At the same time, lots of smaller businesses will be put off because they think AI is only for big companies with huge IT budgets, ignoring the fact that there are ways to implement AI cost-effectively while still providing real benefits.

Again, the answer is thorough preparation and being as accurate and diligent as possible in your costing before jumping headfirst into AI.

5. Letting The Competition Pip You To The Post

Of course, despite all of the precautions covered here, perhaps the biggest single mistake would be deciding it’s too difficult, risky and expensive and choosing to sit out the AI revolution entirely.

Make no mistake: AI is set to transform every industry, and those who do stick their heads in the sand and pretend it isn’t happening are going to be left behind.

AI will make businesses more efficient, meaning those that don’t implement it at all are effectively throwing away money. But it will also drive innovation, meaning competitors that do adopt it will create new products and services, redefining customer expectations and leaving laggards looking decidedly old-hat.

Don’t Fall Into These Traps

I know from my experience of working with companies of all shapes and sizes on digital innovation strategies that these hazards and pitfalls will cause many businesses to experience big problems in 2025.

Understanding the risks and having the foresight to see them before falling into them headfirst is the key to success, as well as the secret to delivering AI-driven growth and transformation.

By aligning AI projects with business goals, ensuring your people are included in everything you do, accurately assessing costs, and knowing when to press ahead and when to quit, you give yourself the best chance of avoiding making these mistakes.

This will result in AI projects that deliver better returns and, perhaps more importantly, prepare your organization for the even bigger opportunities that are yet to come.

ChatGPT’s new Operator mode is its first step towards becoming an AI agent – a new type of AI tool that carries out far more complex tasks without the need for human intervention.

But what exactly can it do?

Currently, in preview mode, the latest upgrade to ChatGPT works by combining the GPT4o reasoning model with computer vision capabilities. This lets it “see” and interact with anything on a screen.

It does this with the help of a built-in web browser, and after telling it what we want it to do, you can just sit back and watch as it moves the mouse, presses buttons and inputs text.

So what can it do? Well, just from playing around with it, it’s evident that it is still at an early stage of development. Nevertheless, it has some impressive tricks up its sleeve.

Here’s an overview of some of the tasks I’ve seen it perform so far, as well as a look ahead to what this could mean for the future of “agentic” AI.

How To Access ChatGPT Operator?

First off, ChatGPT Operator is only available in the US right now and only to users who’ve taken out the $200-per-month Pro subscription.

This probably won’t always be the case – ChatGPT creators OpenAI tend to roll out new features to small groups such as Pro users first before opening them up to a broader audience.

But if you are lucky enough to meet those criteria, then you should be good to go. Simply head to operator.chatgpt.com to get started.

Things To Do With Operator

Operator is designed to carry out more complex, multi-step tasks than is possible using standard ChatGPT. It’s capable of carrying out up to three of these tasks at the same time.

While it’s designed to be autonomous, there are times when it will have to hand control back to you – for example, to log into websites or to solve CAPTCHA challenges.

One really useful feature is integrations. These are instructions on using specific sites or services, such as Airbnb or OpenTable, so Operator doesn’t have to learn how to use them from scratch every time it comes across them. Since these integrations are created using natural language prompts, businesses can easily develop and share their own with customers.

Here are some of the things it can already do:

Find And Book Accommodation Through Airbnb

Tell it to head to Airbnb to find a room based on your preferences, and it will search options, check reviews, and ensure you’re happy with its choice before going ahead with the booking.

Sample prompt: “Find an Airbnb room in [destination] from [check-in date] to [check-out date] for [number of guests]. Prioritize properties with good reviews, Wi-Fi, and amenities like [list preferred amenities, e.g., kitchen, balcony, pet-friendly]. Ensure the location is close to [landmarks or areas, if important]. Confirm availability and details before booking.”

Make Restaurant Reservations

Operator integrates with OpenTable and can scan services like TripAdvisor to research and book restaurant reservations.

Sample prompt: “Find a highly-rated restaurant near [location or city name] that serves [preferred cuisine or dietary preferences, e.g., Italian, vegan-friendly]. Book a table for [number of people] on [date] at [time]. Ensure the restaurant has good reviews and confirm the reservation before finalizing.”

Book Tickets To An Event

Want to watch live music or sports or take in a show? Operator will browse events directories and integrate directly with StubHub to find the best seats at the best prices.

Sample prompt: “Search for events happening near [your location] on [specific date or date range, e.g., February 10th or next weekend]. Include concerts, theater shows, festivals, or anything unique and interesting. List options with event details, times, and ticket prices. Once I choose one, book the tickets for [number of people] and confirm the booking.”

Plan And Shop For Meals

Not only will Operator plan your menu, but it will integrate with Instacart to order the ingredients and have them delivered to your door.

Sample Prompt: “Plan meals for a family of [number of people] for a week (breakfast, lunch, dinner). Take into account the following dietary requirements: [list requirements, e.g., gluten-free, vegetarian, nut-free, low-carb]. Include a mix of healthy and family-friendly recipes. Once planned, order the ingredients online and ensure delivery is scheduled for [preferred delivery date and time]. I already have [insert staples, e.g., milk, sugar, flour], so you don’t need to order them.”

Update Or Make Changes To A Website

How about something more challenging? From uploading blog pages to changing design elements and generating entirely new content, Operator can integrate with no-code building platforms like Wix to get the job done. While building an entire site from the ground up might be a little taxing (at the moment), it can carry out routine maintenance, design tweaks and updates with relative ease.

Sample Prompt: “Edit the website [website URL or description of the page] to update the following: [describe what you want to be changed]. Log into [insert no-code service, e.g., Wix] to make the changes. Ensure the design remains consistent and user-friendly. Once edits are made, show me the final result for review.”

The Start Of Something Big?

Although it’s early days, I believe that Operator could mark the start of a new chapter in the history of AI. Possibly even a change in the fundamental relationship between humans and machines.

Agentic AI is the first step towards the creation of truly autonomous thinking machines that can act as assistants to us in all fields of life.

Some believe it’s a step towards artificial general intelligence (AGI) – the “holy grail” of AI where machines can learn just like we do in order to perform tasks beyond those they’ve been programmed for.

The arrival of ChatGPT just two short years ago marked a watershed moment for AI. Now, the arrival of agentic AI marks another – and I’m just as confident it’s going to change the world in ways that are hard to imagine today.

ChatGPT just launched a new feature that, alongside OpenAI’s groundbreaking Operator agent, signals a future where AI is far more capable than today.

Tasks let users schedule actions to be carried out in the future or at regular intervals. While Operator can actively browse the web and complete complex tasks independently, Scheduled Tasks focuses on automated, time-based interactions.

A simple idea, sure – but one that complements OpenAI’s broader push toward agentic AI – autonomous AI that can fulfill complex, multi-step objectives without needing human input. This vision is already taking shape with OpenAI’s Operator agent, which can independently navigate websites and complete tasks on users’ behalf.

For now, though, ChatGPT users can get used to interacting with the scheduled task feature that can help us with day-to-day activities even when we’re not online.

What Are ChatGPT Scheduled Tasks?

The concept is pretty straightforward – Tasks lets us schedule prompts to be run in the future or at regular intervals, such as once a day, then wait for the results to be relayed back to us by notification or email.

What makes it exciting is that it’s part of OpenAI’s broader initiative to create truly autonomous AI assistants. While Operator handles real-time web interactions, Scheduled Tasks manages time-based automations – both representing different aspects of AI working independently on our behalf.

How To Use Scheduled Tasks

At the moment, Tasks are only available to users with a paid ChatGPT Plus, Pro or Teams subscription.

To get started, just click your profile picture and select Tasks, or select ChatGPT With Scheduled Tasks from the drop-down Model menu. OpenAI has created a page with more detailed instructions, including how to set up push notifications, on its website here.

about:blankClick to accept marketing cookies and enable this content

Five Helpful Ways To Start Using Scheduled Tasks Today

Just to give you an idea of the sort of thing it can do, here are five simple prompts that can be used to get regular ongoing assistance with day-to-day activities.

Personalized Morning Weather And Travel Briefing

This prompt sends helpful daily updates before you leave the house:

Every morning at 7 am, provide me with a weather report for [my location] as well as a travel report covering disruption on any local roads or transport networks, for the current day only.

Create Topical Social Media Posts For A Business

Finding ideas for social media posts can eat into your valuable time, particularly  if you’re running a business by yourself:

At 9 am every day, please generate the text for a social media post for my [insert type of business]. It should relate to something topical connected to my business and should either promote the importance of the products and services we supply or educate or inform the customer about something of interest to them.

Create A Personalized To-Do List, With Reminders

This will help you organize your day and keep track of jobs you need to get done:

Every day at 8 am, ask me what tasks I need to get done that aren’t already on my list. Generate a list of all my ongoing tasks along with tips to help me get them done. Remove tasks from my list as I tell you I’ve completed them.

Keep Up-To-Date On Topics That Matter To You

News briefings on any subject you want:

Every morning at 7 am, send me the latest news and content on the subject of [insert subject].

Meal Planning

Create a customized meal plan for the week:

Every Monday at 9 am, send me a weekly meal plan consisting of seven healthy dinners for a family of [insert family size] and provide a shopping list of all the ingredients I’ll need, broken down into categories.

Towards AI Agents?

We’re already seeing OpenAI’s vision of agentic AI taking shape. While Scheduled Tasks handles time-based automations, Operator demonstrates more advanced capabilities – actively browsing the web, making decisions, and completing complex tasks independently.

The Scheduled Tasks feature may be in beta, but combined with Operator, we can see OpenAI’s roadmap emerging. Future versions might integrate these capabilities – imagine scheduling an Operator task to automatically book your favorite restaurant every Friday or having it monitor prices and purchase items when they go on sale.

This integration would move OpenAI’s technology well beyond simply answering questions and giving advice. With Operator already handling web-based tasks and Scheduled Tasks managing time-based automations, we’re watching the evolution of AI into truly useful everyday assistants that can work independently on our behalf.