Along with many companies in mid-2025, we are exploring and embracing the possibilities of AI within our every day work. We are looking for ways to automate the things that used to require human time, effort and, yes, expertise. This naturally leads to the question: do human skills still matter in an AI world? Clearly, the answer (at least in mid-2025!) is *yes!* - human skills still matter. But what is also clear is that the nature of the skills that we need to bring to the workplace is shifting - we need to adapt and develop our uniquely human skills. According to [McKinsey & Copmany Superagency in the workplace](https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/superagency%20in%20the%20workplace%20empowering%20people%20to%20unlock%20ais%20full%20potential%20at%20work/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-v4.pdf?shouldIndex=false), Agentic automation could eliminate 40-60% of repetitive cognitive tasks in enterprise workflows by 2027, but as this report frames it, AI could augment, not replace human workers and should be seen as a cognitive automation tool to which humans can delegate reasoning, decision making and information processing. [["The Future of Work is AI with humans, not AI versus humans" - Calvin Chu|We need a mindset shift to use AI as a strategic collaborator]]. This is particularly important given growing research on [[The impact of using AI on the brain]]. We need to get clear about [[What AI Can Do (and What It Can’t)]] - a position that will undoubtedly change but is key. And from that we can assess [[Thinking models that still matter]]. In particular, it is currently vital to check the accuracy of results, particularly regarding linked sources: https://www.theregister.com/2025/05/15/anthopics_law_firm_blames_claude_hallucinations/ In order to understand the limitations of LLMs, it is useful to get a grasp on [[Comparison of human learning and LLM 'learning']] and [[Tokens in Large Language Models]]. More detail can be found in [[More on how ChatGPT processes tokens]].