Trend 1: Leaders are "adopting" AI without a clear business goal

Organisations are buying AI tools and commissioning training before they have decided what they want AI to achieve.

The fear of missing out drives more than half of technology investment, while only a small minority of enterprises have documented AI goals. Staff are also less confident in the training they receive than leaders are in the training they provide. The result is activity without a common goal.

Businesses already have more AI capability than they use. Agent adoption remains rare, and much of the technology has yet to be connected properly to company data and day-to-day work. Another leap in model capability will not resolve those gaps.

Call to action: Define the business result before buying another tool or launching another course.

1. Choose a small number of business outcomes that AI should improve.
2. Set a baseline so progress can be measured against cost, quality, speed or customer experience.
3. Fund tools, training and workflow changes only when they support those outcomes.

Relevant courses:

What Problem Are You Trying to Solve? | AI Steering Committee and Governance

Read more:

Leaders Are Going All In on AI Without a Strategy | An AI Slowdown Could Be Exactly What Businesses Need

Trend 2: Stricter governance is needed when AI Agents do real work

AI governance changes when a system can send, update, approve or trigger work instead of simply recommending an answer.

An assistant can produce a poor draft for someone to review. An agent can carry that error into another system before anyone sees it. Existing policies often explain how employees may use AI, but leave open what an agent may change, when approval is required and who can stop it.

Agents need the organisational equivalent of onboarding, defined authority and an audit trail. Each one also needs a named human owner, even when the system completes most of the work by itself.

Call to action: Set the boundaries of an agent's authority before giving it access to live work.

1. List the actions each agent may take and reserve consequential decisions for human approval.
2. Record its activity, corrections and overrides so failures can be investigated.
3. Name an owner with the authority to pause, restrict or withdraw the agent.

Relevant courses:

When AI Agents Join Your Team: Governance for Leaders | AI Steering Committee and Governance

Read more:

When AI Moves From Advising to Acting, Strategy Must Follow | How Chief Legal Officers Must Architect Enterprise AI Governance | How to Govern AI Through Human and AI Symbiosis

Trend 3: Many employees have not started because the rules are unclear

AI adoption is often stalled by confusion, not resistance. People may be unsure whether the tools apply to their work or whether they are allowed to use them.

About half of the people starting a large public-sector AI course had never tried generative AI. Common reasons included limited skills, uncertain relevance and unclear permission. Managers reported the same uncertainty, leaving employees without anyone who could confidently explain the boundaries.

The key is clear communication and staff training. This is what allows beginners to make the largest gains in knowledge and confidence. Confidence also helps them recognise more useful applications. Clear permission and practical support therefore need to arrive together.

Call to action: Give employees a safe and specific starting point instead of waiting for adoption to spread by itself.

1. Publish which tools are approved, what information may be used and when human review is required.
2. Show each team a small number of relevant uses drawn from its actual work.
3. Equip managers to answer questions and judge whether use is improving the service or outcome.

Relevant courses:

GenAI Foundation Course | Leading Teams Through New AI Workflows

Read more:

What 250,000 Public Workers Reveal About Adopting AI Well

Google Meet's built-in tools

We have been using automated meeting and transcription tools since they emerged a few years ago. It's only when someone in a meeting pulls out a notebook that you realise some people have not embraced this small but powerful approach to modern meetings. Having a transcript is like gold: you don't divert attention to take manual notes, you know exactly who said what, and you have a formal record of decisions and insights.

Be careful of these tools, though: some defaults can see the whole group automatically emailed a copy of the notes, which can be awkward if you are not expecting it.

Frontier models can seem dumb if you underestimate them

If you ask a powerful model like Fable to do trivial work, you won't be very impressed because it will take longer to respond and might even answer worse than a simple model. You really have to push Fable in terms of what problem you want to solve to unlock its true capabilities. We used it with a client, giving Fable a high-level strategic goal - something that needed to happen over the next few years. Then we asked the model to identify the tactical steps that'll help execute that strategy, and it blew the client away.

Ambition doesn't have to be that grand. If you're putting together a presentation, don't just prompt it for slides; share the core message you want to communicate, the impression you want to leave, and what you actually need to achieve. It might even suggest a format that works far better than a slide deck.

Move work from ideas to action using Linear

Modern work is usually made up of strategy (the goal) and tactics (how you achieve it). AI tools are excellent at helping you refine both, but the work can get lost in a mountain of chat history unless you create actionable tasks. Tools like Linear are fantastic for this: they work really well with AI tools, and as you work on a project, you can simply let Claude create tasks for you in Linear. Or, ideally, for someone else. Linear is a task management platform where you create tasks, assign them to people, set due dates and more. A simple and effective way to move work from ideas to action.