Trend 1: AI Is Reshaping Who Gets Hired and Who Does the Hiring

AI has moved into the recruitment process itself. Algorithmic screening now filters candidates before a human reads a CV, hiring criteria have shifted toward AI fluency as a baseline expectation, and the entry-level pipeline is narrowing. There are two risks: the first is that the wrong people get filtered out, and the second is that organisations are hiring for AI skills they don't reinforce once people are through the door.

• LLM-based screening tools may already favour CVs written by the same model family doing the evaluating, creating a self-reinforcing loop that rewards AI-polished applications over real capability.
• Employers increasingly list AI literacy as a requirement, yet most still offer thin formal training, which leaves a gap between what they ask for and what they invest in.

Call to action: The organisations that win on talent will audit hiring models to assess candidate qualifications rather than AI-tool stylistic patterns, and close the gap between the AI skills they advertise for and those they actually develop.

1. Audit what your job postings ask for versus what your onboarding delivers
2. Test screening tools for bias toward AI-generated text over human substance
3. Train recruiters to assess judgment, not just tool familiarity

Relevant course:

GenAI Foundation Course

Read more:

Nvidia exec: AI likes to use AI and algorithmic hiring may favour AI-written CVs | Employers want AI skills. What is the best way to learn them? | Airbnb CEO: two kinds of people who risk falling behind in the AI era | A reality check on the AI jobs hysteria

Trend 2: When Fluency Masks a Lack of Understanding

Last month we covered hallucinations: AI confidently making things up. This month the risk is subtler. AI assistants now produce work that looks rigorous because it's well-structured, plausible, professionally worded, but it can bypass the human reasoning it was supposed to support. The finished page looks fine while the thinking behind it may never have happened.

• AI writing tools smooth over gaps in reasoning rather than exposing them, producing persuasive structure even when the underlying question is unclear or flawed.
• Research shows that sustained AI-assisted writing weakens brain connectivity, reduces ownership of the work produced, and leaves users unable to recall or defend what they wrote.
• Business schools are now moving past tool training to teach when to trust, question or override AI outputs.

Call to action: Fluency is not the same as understanding. If your review processes can't tell the difference, the gap will show up in work, decisions and the credibility of your people.

1. Design reviews that test reasoning, not just the finished product
2. Alternate AI-assisted and unassisted practice to keep judgment sharp
3. Train people to challenge confident-sounding output, not just check for errors

Relevant course:

AI Fluency: Pushing Back on Polished Output

Read more:

Eager to please AI assistants smooth over the gaps in our thinking | Your Brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing | Business schools move beyond the basics to teach collaboration with AI

Trend 3: AI Is Disrupting the Business Models That Businesses Run On

The disruption conversation has focused mostly on jobs and workflows. The bigger structural question is now surfacing: which businesses survive as AI absorbs the value proposition they were built on?

• AI is now being packaged directly into vertical tools that compete with design, legal and small-business software products.
• The most defensible competitive positions belong not to whoever controls the AI layer, but to organisations controlling the data AI needs to function. This makes data ownership a business model question, not just a technology one.
• First-mover advantage is real but closing: in real estate, only 25% of firms qualify as AI leaders against 40% across industries, and those treating AI as an add-on to legacy models are already falling behind on margins and timelines.

Call to action: The question is no longer whether AI will affect your business model but whether you are building the moats that survive what AI commoditises next.

1. Stress-test whether your core product is a scaffold or a business
2. Identify where proprietary data creates durable advantage and invest in it
3. Stay model-agnostic by treating AI providers as infrastructure, not strategy

Relevant course:

AI Steering Committee and Governance

Read more:

What did Claude just kill? Six moats to protect yourself | The private capital opportunity in AI-enabled climate and sustainability sectors | Four ways to accelerate growth with AI and analytics | The AI-first real estate company: an opportunity for structural advantage

Claude Design

Our Learning Director took to Claude Design to see how it compared with how Google's NotebookLM, which she currently uses to test UX options for our new pop-up courses. This is how she used Claude Design:

• Gave it our brand: fonts, colours, the full palette
• Shared one of our longer courses as a style reference
• Wrote prompts focused on UX (flow, navigation, readability for a shorter format)
• Reviewed what came back on the canvas, then refined through chat and inline comments

Why we love it: By the end, she had a couple of strong prototypes ready to take to the design team, which were way better than NotebookLM. Her favourite part was being able to leave inline comments on specific items that she wanted changed, which sped things up massively.

First draft in the canvas for Copilot in Outlook

Launched in March

Our programme co-ordinator handles many emails from multiple addresses. They're enjoying using an Outlook update, where Copilot writes a first draft directly in the canvas and can then iterate with the user to continue improving it. Instead of generating a one-off draft, Copilot writes in place and asks clarifying questions about the goal, audience or tone, which is very useful for different inboxes that deal with distinctly different issues. The email is then updated as the user responds.

Why we love it: It lets users refine a message in a few quick turns until it's ready to send and keeps every change visible in Outlook with no copy/paste or formatting surprises.

Google AI Studio

If you're working with AI models and haven't tried Google AI Studio, it's worth ten minutes of your time. It's a free browser-based tool that lets you test prompts against Google's latest Gemini models (currently 3.5 Pro, 3.5 Flash, and 3.1 Flash-lite) without writing a line of code.

Why we love it: The killer feature for us is the ability to re-run a single step in a conversation in isolation: rather than replaying an entire chat to test one change, you just tweak and re-fire that turn, which makes prompt refinement much faster.

We also use it to run the same prompt across different model sizes back-to-back, which quickly tells you whether you need the full power of 3.5 Pro or whether the lighter Flash variant does the job just as well.

Value: For more technical use cases, it's excellent for stress-testing agentic workflows: you can probe function calls and structured outputs with awkward edge-case inputs before anything goes near production.