Why might workers rather pick up AI tricks quietly inside their everyday tools than sign up for a formal training course?

This explores why workers might rather pick up AI skills inside the tools they already use than in separate training programs. To be upfront, none of these notes directly compares the two. What the corpus does have is a set of findings from different angles, and together they suggest the preference is less about convenience and more about social risk, recognition, and how skills actually take hold.

Start with social risk. Across four experiments with more than 4,400 people, AI users expected colleagues to see them as less competent and less diligent, and they were less willing to tell managers they used AI Do people fear judgment when they use AI at work?. A formal training program makes AI use public: you enroll, you show up, your name goes on a list. Guidance inside the tool lets people learn privately. If using AI carries a stigma that people see coming, learning quietly is the sensible choice.

Second, workers often don't know what to ask for. Evans argues that making tools easier to build doesn't help much, because most workers don't see their own tasks as automatable Does easier tool-building actually solve enterprise adoption problems?. A course teaches general capabilities and leaves people to connect them to their own work later. In-tool guidance shows up at the moment the task is in front of them, so it does that connecting for them. This fits a finding about agent 'skills': in more than 8,000 trials, skills worked mainly as procedural anchors that steady action in context (65.7% of cases), and rarely as injections of missing knowledge (4.5%). They failed when they were invoked out of context Do skills teach procedures or inject missing facts?. If procedures stick best when they are anchored to the task at hand, that may hold for people too.

There's a twist, though. Wanting guidance isn't the same as wanting to be taught. In 200,000 Bing Copilot conversations, users mostly wanted information gathered or writing done, but the AI mostly coached, advised, and taught. In 40% of conversations, what users wanted and what the AI did didn't overlap at all Why does AI default to coaching instead of doing?. So 'in-tool guidance' that turns into an unrequested tutorial may be no better than a classroom. The more promising design is the one described in Learning to Guide, where the AI points out the parts of a problem worth paying attention to and the decision stays with the person Can AI guidance reduce anchoring bias better than AI decisions?.

Two more findings weaken the case for formal programs. Company policies (tool mandates, allow-lists, data rules) set how much control engineers have over AI before personal preference comes into play Does personal preference shape how engineers use AI tools?. That means a general course may teach workflows people aren't allowed to use. And recruiters gave AI skills a 8 to 15 point boost in interview invitations, but certificates added only a modest amount over simply listing the skill Do AI skills help candidates get more job interviews?. If a formal credential barely pays off, workers have little reason to sit through the program that grants it.