Trend 1: Human judgment is still the scarce advantage

AI increases output volume; humans increase output quality.

• AI is less likely to replace jobs that are built on trust, care and judgment.
• AI cites people, not brands: around 75% of LinkedIn citations came from individuals.
• Without a clear human voice, AI only scales the noise.

Call to action: Use AI to clear administrative load; keep humans responsible for decisions that need trust. To do this, identify work that is suitable for AI (routine, rote, repetitive), and work that requires human touch (judgment, nuance, accountability).

1. Map where human judgment must stay non-negotiable in each workflow
2. Coach leaders to publish practical, evidence-backed insight rather than brand-only content
3. Train staff to verify AI output against primary evidence before acting on it
4. Pair AI fluency with deliberate reflection on when not to use the tool

Relevant courses:

GenAI Foundation Course | Data, Privacy and AI: Practical Decisions Every Day

Read more:

Safe from AI: which jobs will help you thrive in the future? | How AI is redefining executive thought leadership | AI Human Amplification: Compounding Value of the Human Voice

Trend 2: Adoption is not impact: organisations must reinvent work

Value shows up when firms redesign workflows, talent and operating models, not when they hand out tools.

• 70% of people feel ready for AI; only 27% of leaders say their organisations are (McKinsey).
• Firms reinventing work report real value nearly four times more often than those stuck on tools (48% vs 13%) (McKinsey).
• Only 6% of companies qualify as AI leaders, yet they outperform peers by nine percentage points (BCG).

Call to action: Stop treating licences and courses as proof of progress. Move from enablement theatre to operating-model change: redesign the work, then measure outcomes.

1. Close the readiness gap: Morgan Stanley built governance and evaluation into its adviser assistant before scaling it to 98% of adviser teams. (Morgan Stanley)
2. Integrate AI into workflows: BBVA embedded an internal assistant into banking operations, cutting query handling from 7.5 minutes to around one minute. (OpenAI)
3. Reinvest gains into growth: Moderna uses AI to accelerate clinical work while pursuing up to 15 new product launches over five years. (Moderna)
4. Make learning practical and shared: BBVA's 90,000-person AI community runs bootcamps, meetups and mentoring around employees' real work. (BBVA)

Relevant courses:

Leading Teams Through New AI Workflows | AI Steering Committee and Governance

Read more:

From adoption to impact: Three horizons of AI transformation | How AI Leaders Create Competitive Advantage | Morgan Stanley uses AI evals to shape the future of financial services | How BBVA is scaling AI from pilot to practice across the org | BBVA drives AI adoption through talent

Trend 3: "I want to break free." — Frontier AI models

Confident AI output and autonomous agents raise the bar for verification, governance and defence.

• OpenAI models escaped a sandbox and compromised Hugging Face during a cyber test.
• Anthropic reports three instances of models gaining unauthorised access to private systems.
• Researchers say they've found the first documented case of AI agentic ransomware.

Call to action: As AI agents become more capable, verification and containment become core operating disciplines. Assume advanced models can chain novel exploits, and respond to AI output with evidence, not opinion.

1. Harden evaluation and production environments; treat sandboxes as incomplete
2. Actively seek and close potential AI attack vectors, including agentic ransomware patterns
3. Give boards and executives hands-on literacy plus candid cyber-risk assessments

Relevant courses:

Data, Privacy and AI: Practical Decisions Every Day | AI Steering Committee and Governance | Advanced Prompting and Agentic AI

Read more:

OpenAI and Hugging Face partner to address security incident during model evaluation | Investigating incidents in cybersecurity evals | JadePuffer: agentic ransomware for automated database extortion

Using an AI workspace like Cursor

We've found that using an AI IDE workspace like Cursor gives us much more flexibility than relying on a single AI tool. Instead of being restricted to one model, we can choose from a wide range depending on the task, compare the outputs, and take advantage of new models as they become available.

What is it? Cursor is a software development tool that is similar to Claude Cowork because it can work on local files, but different because it can use multiple AI models.

Value: We've learned that no single AI model is the best at everything. Some are better at writing, others at reasoning, coding, research or creating images. Having access to multiple models means we can choose the right tool for the job.

Combining different AI tools

While creating images, we found that using different AI tools for different parts of the process produced much better results. We use Claude first to refine and perfect the prompt, then give that prompt to ChatGPT to generate the image. We've consistently found that this approach creates better outputs than trying to build and refine the prompt entirely within ChatGPT.

Value: Rather than expecting one AI to excel at every stage of a task, we've learned to use different tools for their individual strengths.

ChatGPT is back in the race

For the past six months, we have increasingly relied on Claude models as our primary workhorses because their quality was consistently superior. However, in the last month, with the release of GPT 5.6, the OpenAI models have leapt forward. This seems to have coincided with something of a regression in Anthropic model quality: overly verbose, drifting off task, losing threads.

ChatGPT is more focused, more task-oriented, and generally more helpful across the majority of our daily tasks.

Value: Where possible, try to position your business to avoid vendor lock-in. Models will keep evolving, and it is valuable to have options when using them.