In Brief:
- Keeping up with AI tools is a race organisations can’t win: Elemental Concept’s Head of AI notes that tool tips go out of date within months.
- Most failed AI projects fail on unclear business value, not weak models.
- Four tests show whether AI is actually working: quality, reclaimed time, capability and business understanding.
- Automating entry-level work without replacing what it taught creates a mid-level skills gap later.
Drawing on a piece by Alok Ashtikar, Head of AI at Elemental Concept
Why AI value comes from judgement, not the latest tools
Many of the AI questions organisations are asking today are about tools – which model, which coding assistant, which agent framework, and whether to switch now or wait for the next release. They’re fair questions, and someone does need to answer them. But Alok Ashtikar, Head of AI at Elemental Concept, recently made an argument that leaders should take seriously: if keeping up with the tools is the goal, “you have signed up for a race you cannot win, because the track keeps extending under your feet.”
His evidence is hard to argue with. By METR’s measure, the length of task that frontier AI can complete is now doubling roughly every four months. The model releases of a single autumn fill a table that, as Alok points out, starts going out of date the day it’s published, and “if your edge is knowing this list, your edge expires every month.” Prompting tricks and tool preferences lose their value fast, and once everyone has the same tools, using them well stops being an advantage and becomes the minimum.
Alok wrote this for individuals and their careers, but the argument matters just as much for organisations – because organisations are the ones signing the licence agreements.
Why do so many AI projects fail to deliver value?
Tools are not the bottleneck
Most AI projects fail because nobody defined the business value they were meant to deliver, not because the models are too weak.
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing rising costs, inadequate risk controls and unclear business value. The scarce ingredient is business judgement, not better tools.
In practice, that means nobody worked out what the model was meant to change, how anyone would know whether it had, or who owned the decision when it got something wrong.
Alok puts the distinction perfectly: “A person who understands the business can point a fleet of agents at the right problem. A person who only understands the tools can point them at any problem, quickly.” The same is true of organisations. A business that responds to every new release by re-evaluating its stack is busy, but it isn’t making progress, and the difference shows up later as a budget line nobody can justify.
That isn’t to say tool choices tell you nothing: your people have already been running their own experiments, sticking with some tools, dropping others – and those choices show where AI is already creating value. Tool choice is a poor basis for a strategy but a very useful piece of evidence, and many leadership teams aren’t reading it.
How can leaders tell whether AI is actually working?
The better question, at organisational scale
Leaders can tell whether AI is working by looking past usage figures to four tests: whether the quality of work is rising, whether reclaimed time is going to higher-value work, whether people are becoming more capable rather than more dependent, and whether teams understand the business better than they did a year ago.
Adapting the question Alok asks of individuals, “Am I growing?”, to the organisation as a whole:
The four tests of organisational AI growth:
- Is the quality of our work improving, or just the volume? AI makes it easy to produce more. It’s much less clear that it makes anything better, and a team turning out twice as many reports nobody reads hasn’t become more productive.
- Where is the reclaimed time going? If AI saves everyone twenty minutes a day, that time doesn’t turn up on the P&L. It disappears into more emails, another meeting and a slightly more polished slide. Individuals can choose to spend saved time better, but only a leadership team can redesign the work so the gain builds into something the business can bank.
- Are our people getting sharper or more dependent? Research Alok cites links heavy trust in AI with less critical thinking. It also suggests the outcome depends on how the tool is used: one recent experiment found that showing people the learning cost of offloading roughly halved it and improved their unaided performance. In Alok’s words: “Same tool, different relationship, different outcome.” That relationship is something an organisation can shape through how it rolls AI out and what it rewards.
- Do our teams understand the business better than they did a year ago? This kind of understanding becomes more valuable as tools improve, and it’s the one advantage a competitor can’t buy off the shelf.
Individual growth matters. But it isn’t enough on its own; people can sharpen their judgement and deepen their understanding of the business and still find the organisation around them working exactly as it did before. Leaving AI to individuals isn’t a neutral choice either: when expectations rise and the work isn’t redesigned to match, the people in the middle absorb the strain. Growth adds up when leaders decide what AI is for and change the work to suit it.
What happens when AI automates entry-level work?
The apprenticeship problem
When AI automates entry-level work, organisations lose more than tasks: they lose the apprenticeship through which junior staff learn judgement, trade-offs and how the business really works. Entry-level hiring in the most AI-exposed roles has already fallen sharply. Without a deliberate way to replace that learning, organisations could face a shortage of capable mid-level people within a few years.
Alok raises an important point about the role AI plays in our development. Rather than focusing only on which tools to use, we should also be asking whether we're improving the quality of our work, learning more deeply and developing a better understanding of the business.
One point in Alok’s piece deserves more attention. The decline comes mainly from fewer people being hired rather than from layoffs, so it rarely shows up as a visible decision. What disappears is the work through which people used to pick up the “tricks of the trade, street smarts, knack of things, balancing trade-offs” that nobody writes down.
An organisation that automates its first rung may find, in five years or so, that it has a shortage of capable mid-level people and no clear idea why. That needs a plan now, not when the gap starts to show.
How to get real value from AI investment
Getting real value from AI investment starts with deciding what outcome you are buying before choosing a tool. Leadership teams that capture value delegate the labour to AI but keep the decisions about problems, quality and risk with people, treat governance as part of the design rather than an afterthought, and measure results and learning rather than usage.
Alok ends with seven practical habits for individuals, starting with “Think first, then ask.” The organisational version is the same discipline at a larger scale, and just as easy to skip. The tools will keep improving, and they should be used to the full. But, as Alok puts it:
“Tool knowledge depreciates. Judgement, depth and business understanding appreciate.” Learn the latest tricks, by all means. “Just do not mistake them for the goal.”
