Strategy, Leadership and HR Insights for Tech Companies | SOTA

The leadership skills that gain value as AI spreads

Cover: The leadership skills that gain value as AI spreads
AI has moved from pilot to default. According to ICONIQ's 2026 State of Scaling report, more than half of US businesses now pay for at least one AI product, up from 7.5% in January 2023. Inside ICONIQ's portfolio of high-growth companies, over 70% of employees use AI tools every day, and access is close to universal: 99% of people in sales and marketing, 98% in R&D and 97% in G&A.
Spending tells a less even story. Ramp's AI Index shows how differently companies invest. The most AI-intensive 1% of US firms spend about $7,500 per employee per month on it. The top 10% spend about $611. The median firm spends $11.38, roughly the price of one seat on an enterprise ChatGPT or Claude plan.
That gap will close. Spend at the top grew 14.1% in a single month, and the median will follow as AI is built into the software teams already use. So the useful question for a leadership team is not how much to spend. It is what the people deciding how to use AI need to be good at, once everyone has it.
We reviewed research from Harvard Business School and BCG, alongside decades of leadership effectiveness studies, to answer that. The short version: the capability that matters most is agency, and the skill inside it that gains value fastest is judgment about where AI's competence ends.

Agency: deciding what matters before the picture is clear

Agency is the ability to decide what matters, choose a direction under uncertainty, and take responsibility for moving forward until there is a result. AI makes options fast. It drafts the strategy memo, models five pricing scenarios and summarises forty customer interviews in a minute. It does not choose between them or own the consequences. That part stays with a person, and the more options the tools produce, the more weight the choice carries.
In our work with leadership teams of fast-growing tech companies, agency rests on three capacities. We call them muscles because they are built through repetition, not acquired in a single workshop.

1. Emotional regulation

The ability to manage stress and setbacks without abandoning the goal. A leader who shuts down when a launch slips, or swerves to a new plan every time a metric dips, gives the team noise instead of direction. AI compresses cycle times, so setbacks also arrive faster. The leaders who hold steady are the ones whose teams keep executing.

2. Rapid learning and adaptive thinking

The capacity to transfer knowledge from one context to another and adjust strategy as new information arrives. Leaders who aren't attached to a specific methodology are more likely to thrive when the rules change. They experiment, learn, adapt, and keep moving, rather than waiting for certainty or the "right" moment.

3. Human judgment

As AI generates more options, choosing well becomes more valuable. Judgment means reading the room and building trust, practising empathy, weighing ethics and social impact, and understanding context well enough to put long-term consequences ahead of short-term optimisation.
Which of the three muscles does your leadership team need to develop most?

The skill that gains value: knowing where AI's competence ends

In a field experiment run by Harvard Business School and BCG, 758 consultants worked on realistic tasks with and without AI. On tasks inside the tool's capability range, consultants using AI were about 25% faster and produced higher-quality work. On a task deliberately designed to sit just outside that range, the same people were 19 percentage points more likely to reach a wrong answer than colleagues working without AI.
The researchers described the boundary as a "jagged frontier": tasks that look equally hard can fall on opposite sides of it. And the tool gives no signal which side you are on. It writes with the same confidence either way.
So the most valuable question a leader can ask of AI output is not "is this well argued?" It is "is this the kind of problem the tool is reliably good at, and how would we know if it isn't?"

This is less new than it sounds

Leadership research has measured effectiveness since the 1950s, through every technology wave, and the predictors barely move: set direction, develop people, decide before the picture is clear.
Knowing when to trust a source is the same skill that governed whether you believed the forecast, the consultant or the analyst's model. What has changed is that this source is fluent enough to be persuasive when it is wrong.

The uncomfortable part

That judgment is built through exactly the hands-on work AI absorbs fastest: first drafts, first models, first analyses that a senior person takes apart. Junior people learned where the edges were by doing the work and getting it wrong in front of someone who knew better.
Organisations that stop developing junior people because the tools can now do the junior work will find that judgment scarce in 5-10 years. For founders and HR leaders, that points to two decisions worth making now:
  • Keep a share of foundational work human on purpose, and treat the review conversation as the training.
  • Make "Where did you not trust the AI model, and why?" a standing question in reviews at every level.
Which roles to keep, grow, and hire for in the age of AI is a talent strategy question, not only a cost one. We looked at how scale-ups can plan for it in Why technology scale-ups need talent market strategists.

Is your leadership team building these muscles?

Your team may be under-developing agency if:
  • Decisions that need a call get another round of analysis instead.
  • A missed target triggers a new plan.
  • AI-generated analysis reaches the board deck without anyone saying where it might be wrong.
  • People are burning out: the tools made work faster, but nobody decided what to stop doing, so the extra speed became extra load.
If the second point sounds familiar, the fix is usually a review rhythm, not a new plan. We describe how that works in Make strategy part of everyday work.

About SOTA: how we develop leaders for the AI era

SOTA works with founders and leadership teams of Series A–C technology companies across Cyprus, the EU, the UK and the UAE, on strategy and leadership development. Clients include Gcore, Fundraise Up, SplitMetrics and AdTech Holding.
Our leadership programmes are built around the capacities in this article and the business goals your leadership layer has to deliver:
  • We assess before we design. We measure where your leaders actually stand on decision-making and the three muscles, so the programme targets real gaps rather than a catalogue of topics.
  • We train on your decisions, not generic cases. Sessions work through the live calls your managers are making, including how they use AI and where they should stop trusting it.
  • We protect the learning loop. We help you design how judgment gets built in first-line and junior roles once AI absorbs the routine work.
  • We measure what changed on the job. After the programme we track behaviour and results, not what people thought of the workshop.
Your managers do not need another framework. They need to run the quarter without you in the room. Book a 30-minute call to talk through where your leadership team is today.

Sources

Leadership Development