Issue 10
What Gartner’s 2026 Hype Cycles say about technology, work and accountability
In my 34 years of global technology industry experience, I have not seen companies spend more on technology than they are doing right now. The spend goes up quarter on quarter, and they have been for the last 3 years or so. The first reining in was observed recently when Accenture capped what it centrally funds for each employee’s Claude use at $29 per month.
Most companies still cannot quantify what exactly they gained from the spend.
Gartner publishes reports on AI, and they published 3 Hype Cycle reports between July and August of this year. The first one was about AI in general, the second was about the future of work as we know it, and the third was the most important one, in my opinion: AI governance. When I examined all of them, I noticed a very interesting thing. All of them are moving at different speeds.
I think of them as three clocks. Technology runs fastest. Work runs slower. Accountability runs slowest.
Let’s discuss the Three Clocks as I see them.
Illustrative editorial graphic. Original synthesis, not Gartner data.
The Technology Clock
Since GenAI was released for us to use, the models keep getting better. They write code, read images, call other software. Agents are starting to run pieces of business processes on their own. The IT team used to battle with shadow IT in the past, but now they are trying to cope with shadow AI in their organisation.
We have absolutely no question about whether AI can be used to build software. That was proven conclusively by the plethora of chest-thumping social media posts declaring that a non-technical, non-software-engineer has built their own applications for their own use. And over a weekend.
The question, at least in my mind, is who can take accountability for what the AI is doing and what it costs.
Gartner’s survey (as seen in the October article on the AI Hype Cycle) covered 1,303 responses. All of the respondents work for a company with at least $50 million in revenue. The survey ran between January and April 2026. The results are very interesting.
Gartner survey · 1,303 respondents · revenue of $50M or more · January to April 2026
Scaled AI beyond their own team or business unit, or went AI-first
Of functional business leaders plan to spend more this year
Did not know what their function spent on AI last year
Source: Gartner, 1 September 2026.
Only 22% were able to scale AI beyond their own team or business units, or had decided to go what is now known as AI-first. 85% of the functional business leaders expressed their desire and plan to spend more this year. A whole one-ninth did not even know what their team/business unit spent on AI last year.
The last one is quite shocking to me.
That is more than 10% of the responses, with no idea of what was spent across all of last year.
What this tells me, and you probably started getting where I am going with this, is that the money is being spent without any measurement.
I have seen many of my clients and prospects say that they are using AI, or that they have found it useful. My first question to them is what was measured when they said it was useful. The second question is whether there were any governing principles on how it was used.
Having spent so long across various countries and organisations, I can tell you one thing for sure.
A demo only proves that it exists, not that it will add value. A pilot, likewise, only proves that it worked once, in controlled conditions. Value and ROI show up much later.
What each stage proves
It exists.
It worked once, in controlled conditions.
Value and ROI show up much later.
The Organisational Clock
Gartner’s Future of Work Hype Cycle looks at people only. When AI starts taking over some of those tasks, a lot of things will change. Roles will be altered because some decisions and skills will change.
In the old world, before AI, that is, the initial work was to gather information. Once that is done as best as possible, exceptions were identified and dealt with. Those are the easy tasks. It takes experience to decide how it has to be done. The judgment of doing the right thing took years of context, of the organisation, of the customers, of the culture and various other variables in the equation.
This is the value the veteran employees bring to the organisation.
More importantly, guiding and mentoring the younger team meant that the knowledge of the judgment was passed down.
| The work | What AI does to it |
|---|---|
| Gathering information | AI can speed it up |
| Identifying and dealing with exceptions | AI can speed it up |
| Judgment | The real challenge. It took years of context: the organisation, the customers, the culture. If the AI system does not have good data, or is not trained well enough, the calls made could be very wrong |
These days, when AI is asked to speed up the first two tasks, judgment is the real challenge. If the AI system does not have good data, or the system is not trained well enough, the calls made could be very wrong. At the cost of all the stakeholders of that decision. Most reports we see today talk about a company saving costs by reducing hours or eliminating workforce.
Those companies are going to learn this lesson the hard way.
Routine work may shrink, but the work that is left over is the hardest part.
Let’s not forget the lessons veterans learned over their tenure. They know customer history, odd ducks and what the reason was for making a different decision a decade ago. Gartner predicts that by 2029, 30% of employees laid off due to replacement by AI will need to be rehired. This will most likely come at a significantly higher cost. Of course, one can argue that it’s a prediction and not a fact. But the question remains nevertheless.
Gartner prediction · by 2029
30%Of employees laid off due to replacement by AI will need to be rehired, most likely at a significantly higher cost.
A prediction, not a fact. Source: Gartner, 9 September 2026.
When AI changes a job, who decides which knowledge stays?
The Accountability Clock
Last year, in Dec 2025, I released a framework called AI Traceability and Accountability Ledger. You can review it at atal.elytrasecurity.com. The third report released by Gartner is partly dealing with the same kind of question. The paper covers what they call AI governance technologies. The paper talks about the tools required to turn policy documents into real controls that can be deployed in the company. We already see quite a few tools in the market that allow enterprise-wide AI control.
I have a very strong interest in this subject. We have started building the applications without proper guardrails.
It is akin to building a home without a proper foundation.
I have argued elsewhere, in my other articles, about the dangers of letting AI make decisions without a mechanism to properly attribute them. If AI is taking a decision today, how many companies can say with 100% confidence that they can attribute it to AI? Do they have proper traceability controls to get to the exact sequence that led to the decision? Who takes accountability for the decision if that decision was wrong and accidentally affected a person or many people, for that matter? AI is being used in many industries today. And all of them touch humans in one way or the other. If a wrong decision destroys a person, family, or a town, who will be held responsible?
Three questions on accountability
Can we say with 100% confidence that a decision was taken by AI?
Can we trace the exact sequence that led to the decision?
Who is accountable if the decision was wrong and affected people?
Governance has been important whenever power over a people or processes is granted to another person or process. AI acts at machine speed, and some of those decisions could be irreversible.
Annual reviews and policies just don’t cut it.
Strong technical controls, traceability when an important decision is taken, and a named human being held accountable are the only way to make this real.
When the Clocks Disagree
The three clocks often don’t see eye to eye. I know companies that are using AI, but they are all experimenting with it. They don’t know what it is capable of. The CTO is worried that someone might upload a sensitive company document, but has no means of controlling it today.
Having an AI policy or set of policies is pointless. Pointless because they cannot be enforced.
The executives are now realising that if a tool saves 2 hours of work, but takes 1 hour or more to check that it was done correctly, the savings are not as high as they thought they were. If an AI agent is right 95% of the time, what is the guarantee that the mistake made (in the 5% cases) is not the one that matters more than the others?
The arithmetic executives are now doing
A tool saves 2 hours, but takes 1 hour or more to check. The saving is not what they thought.
Right 95% of the time. Is the 5% the mistake that matters more than the others?
Five Questions for the Next AI Budget
What business result actually improved?
Include checking and rework in the count.How has the work changed?
Roles, decision rights and skills, not only tools.Who is accountable when an AI-assisted decision goes wrong?
Can we show what the system did?
A request, an authorisation and a completed outcome are three different things.What does it cost to run at scale?
Include integration, upkeep, assurance and errors.Nothing on that list is exotic. AI makes it urgent because it runs faster than our processes were built for.
What I Would Like to See Change
I have worked in enterprise technology for 34 years. A lot of today’s AI excitement feels familiar. A company buys the technology expecting a step change, and questions about process, people and results get asked late, if at all.
AI raises the stakes. These systems now take part in decisions and take actions.
I’m not asking anyone to slow down. I’d like to see more experiments, as long as we look at the results honestly and admit when something doesn’t work. Five things I’d do:
Five things I would do
Start with the work.
Understand the process before picking the platform. How does it perform today? Where does the time go? What would an improvement be worth?
Hand over authority carefully.
An agent that helps with research and an agent that approves a payment are different things. Match the authority to the cost of a mistake, whether you can undo it, and whether a person can step in.
Ask for proof of what happened.
“Task completed” from a system is not evidence of an outcome. Keep records that separate what was requested, authorised and executed, without hoarding sensitive data.
Put experienced people in the room.
Don’t judge AI by headcount removed. The people who know the exceptions and the history are the ones who should decide what AI does.
Count everything.
Subscriptions are the small part. Integration, supervision, security, training and cleaning up mistakes are the rest. Measure over a real operating period.
None of this has to wait for better models. I worry we are going from “technically possible” to “company commitment” faster than we are building the ability to manage what we deploy. I’m still optimistic about what AI can do.
That optimism holds up better when we check the results.
The closing signal
The better question for executives is how well they run what they already have.
Sources and Further Reading
- Gartner, Hype Cycle for the Future of Work, 2026, 20 July 2026.
- Gartner, Hype Cycle for AI Governance Technologies, 2026, 7 August 2026.
- Gartner, Hype Cycle for Artificial Intelligence, 2026, 28 August 2026.
- Gartner, Gartner Identifies 4 Shifts Shaping the Future of Work, 9 September 2026.
- Gartner, The Latest Gartner Hype Cycle for Artificial Intelligence Puts Control Ahead of Capability, 1 October 2026.
- Gartner, Gartner Survey Finds Only 22% of Organizations Have Successfully Scaled AI Across Multiple Business Units, 1 September 2026.
- Forbes, Accenture Clarifies Claude Spending Limits As AI Cost Questions Grow, 16 October 2026.
- Elytra Security, AI Traceability and Accountability Ledger.
An independent interpretation of Gartner’s public research. The full Hype Cycle reports require Gartner client access.