AI in 2025: the year experimentation ended

Over the past year, something subtle changed in how organisations use AI, not so much in the tools themselves, but in how they became part of everyday work. By the end of 2025, AI was no longer impressive or novel, it was simply present, embedded in routines and decisions, and for many organisations, that was the moment experimentation quietly stopped being enough.

As AI moved out of pilots and innovation labs and into daily activities such as emails, documents, search, customer interactions, and internal analysis, its adoption often happened without a formal decision and without clear ownership. In many organisations, AI spread through convenience rather than intent, creating real gains in speed and output, but also introducing new forms of friction that leadership teams were not yet set up to manage.

That shift matters more than any individual model release.

What actually changed

Three shifts reshaped how AI shows up inside organisations.

1. AI crossed a delegation threshold

The biggest change was not speed, it was scope. New systems can now handle multi-step work: gathering information, comparing options, drafting outputs, revising based on feedback. And this changed how teams use AI day to day.

That shift shows up clearly in adoption data: according to McKinsey’s 2025 global survey, nearly 9 in 10 organisations now use AI in at least one business function, up from around three quarters just a year earlier. The increase did not come from new pilots, but from expanded use inside existing roles.

In practice, teams began delegating:

• First drafts of reports

• Research synthesis

• Internal briefs

• Customer response preparation

AI moved from “support tool” to work participant. And once that happens, leadership questions must change.

2. AI became infrastructure, not a tool

AI is no longer something employees “go and use”, it is now embedded in email and document tools, in search, operating systems, and mobile devices.

This is why adoption accelerated so quickly. Microsoft reported that over 90% of Fortune 500 companies are already using Copilot features inside everyday productivity software, google now serves AI-generated responses to billions of search queries each month.

AI became part of how work happens, not an optional add-on.

As a result, usage spreads bottom-up, not through strategy decks, but through convenience. And a real gap opened in 2025, not between users and non-users, but visible use and invisible use.

In addition, a new way of working took shape: vibe working

As AI became part of everyday tools, work became conversational. Instead of writing detailed briefs, step-by-step plans, or finished drafts, many employees began working with AI through direction, examples, and continuous adjustment. They would describe what they were aiming for, look at what the system produced, react, and refine.

This shift is known as vibe working (see my previous article on Vibe Working here), and comes from the extension of vibe coding to everyday knowledge work.

3. Friction surfaced

As AI spread, consequences followed: regulation moved from discussion to enforcement, the EU AI Act entered its first implementation phase, copyright disputes moved into courtrooms, and errors and hallucinations became visible in real business decisions.

At the same time, skills gaps widened. For example, the World Economic Forum estimates that almost 60% of workers will need some form of retraining by 2030, with AI being a primary driver of change. Many organisations felt this already in 2025: a small group of employees became highly effective with AI, while others struggled to keep up.

AI stopped being a technical topic and became governance, legal, reputation and workforce issues. And that was the moment AI entered the leadership agenda for real.

Why experimentation reached its limit

For the past two years, experimentation made sense. Leaders needed to understand what AI could and could not do, and low-risk pilots were the right response to uncertainty.

In 2025, that logic broke down, not because experimentation failed, but because AI outgrew it.

McKinsey’s data shows the tension clearly: while adoption is high, only a minority of organisations report seeing material, organisation-wide impact from AI. Value appears unevenly, risk accumulates quietly, and decisions drift without ownership.

At that point, experimentation without structure creates noise, not progress.

What leadership looks like after experimentation

The organisations that made progress in 2025 did not “use more AI”, they did different things. They:

• Made AI usage visible.

• Defined boundaries instead of blanket rules.

• Redesigned specific workflows rather than adding tools.

• Measured impact instead of activity.

• Treated AI as a recurring leadership topic, not a one-off initiative.

In other words, they moved from exploration to management.

A practical reset for leaders going into 2026

If experimentation ended in 2025, leadership begins now. A useful reset looks like this:

1. Map where AI is already used.
   Not perfectly. Just honestly.

2. Classify usage.
   What is acceptable, what needs review, what is off-limits.

3. Redesign one workflow end-to-end.
   Choose something repetitive and cross-functional.

4. Define a small set of metrics.
   Time saved, cycle time, rework rate.

5. Build role-specific fluency.
   Not general training. Practical use tied to real responsibilities.

6. Review AI regularly at leadership level.
   As you would finance, risk, or cybersecurity.

None of this requires new technology. All of it requires leadership attention.

The real question

AI did not replace leadership in 2025, but it exposed where leadership was missing. If AI is already shaping how work gets done inside your organisation, the relevant question is no longer whether to experiment, it is whether leadership is ready to take responsibility for what is already happening.