I recently had a long conversation with Ben Hoyle that left me thinking about how scientific thinking, particularly from fields like astrophysics, becomes one of the best crash courses in translating AI for real people. Scientists are used to making complicated systems understandable, and that skill is suddenly priceless when your company has 45,000 people who need to know, plainly, how this tech helps them, not how model training works.
Ben’s journey reminded me that astrophysicists were doing big data before “big data” was a thing. That background matters, because he didn’t just show up with algorithmic skills, he showed up with a habit, honed over years, of explaining complex things to non-experts. He put it bluntly, when people ask him how to use generative AI his first response is often, “what are your tasks?” That question is gold. If you don’t know the problem, you can’t match the right tool to it.
We often forget the obvious, which is that most people don’t need a primer on attention layers, they need a concrete example of a tool solving the exact task they care about. Ben’s playbook is simple, and effective. “What we’ve done is we take people or use cases where people have been using generative AI to help solve the task in some way, and then we use them as a figurehead,” he said. Make it visible, make it relatable, let people see a peer using it, and the lightbulb follows.
If you run training alone, you get compliance. If you run showcases and pair them with training, you get adoption. Zeiss pairs both. They collect internal examples, make short videos or demos, and publish them so people can see someone in a similar role actually using the tool. That acts as an “inspiration trigger,” Ben says, and it works. A practical example trumps abstract evangelism every time.
There’s also a political reality. Ben explained that getting buy-in from the top and simultaneoulsy building grassroots communities is the only realistic way to reach a company of tens of thousands. That’s where the executive demo plays out as propaganda in the best sense, “if the boss of the boss of the boss is using this technology, then maybe it’s okay if I use it too,” Ben observed. People need to feel permission, and they need to see the tool producing value in real contexts.
Trust is where things get philosophical. “How do you trust a system?” Ben asked. Companies that were burned by early data leaks, where employees uploaded secrets into public models, now sit in a deep, stubborn skepticism. That matters, because you can’t roll out a tool people are afraid to use.
Zeiss responded by building a cross functional team that covered security, cloud, legal, and compliance. When the company blocked external access, the pressure from people who had already been experimenting forced an internal solution. The result, Ben explained, was a controlled rollout where people got access to safe tools, plus guidelines on what data they could or could not give the models. Sanctioning usage from the top, and providing secure alternatives, transformed clandestine, risky usage into legitimate, supported practice.
This is where some blunt practical advice is useful. A few years ago, dumping a big Excel file into a large language model and asking for a financial report was a bad idea. Use Excel functions, they are reliable and tested. “Don’t use it to solve problems which you can already solve with existing technologies,” Ben warned. That’s still true, but the nuance has shifted.
The real switch is using language models as orchestrators, the brain that decides when to call deterministic tools. So instead of asking the model to add up columns, you ask it to write or call the right Excel or Python function. The model becomes a translator and coordinator, which is where it really shines. This hybrid approach gives you the best of both worlds, human explainability and automated muscle, while keeping deterministic parts actually deterministic.
The single biggest lever for impact, Ben argued, is integration. Give people the power inside the apps they already use. If your company can chat to internal docs within Teams, in the flow of work, adoption accelerates. The technology to do this mostly exists, the barrier is gluing systems together securely and thoughtfully. He painted a neat picture, “I could tell Alexa that I’m going to Berlin and the whole transaction could happen without me needing to pay any money, like physically pay,” meaning, we can orchestrate an entire journey with APIs and agents if we just stitch them together.
I like his timeline, because it removes the hysteria. We don’t need to wait for some mythical perfect model to change how we work. We need better UX, clearer guidelines, and the translation layer that connects tech to tasks. That’s less sexy than a new model announcement, but it’s where value lives.
If you run any team, hire the translators. These are the people who can distill the tech into tasks, who ask “what are your tasks?” and then map tools to outcomes. Build showcases that are specific, pair them with training and clear data rules, and get the top leaders to publicly use the tech. Treat LLMs as copilots and orchestrators, not magic ovens for every problem. And finally, invest in integration, because putting AI in the places people already work is the real multiplier.
We’re not going to lose the human taste for human things. People will still want handmade furniture, crafted meals, and other experiences that scream human provenance. AI will free us to do more of what we actually enjoy, or at least give us fewer hours stuck in repetitive spreadsheets. That sounds worth building toward, slowly and intentionally, with a scientist’s care for evidence and a translator’s knack for plain English.
Commenting Rules: Being critical is fine, if you are being rude, we’ll delete your stuff. Please do not put your URL in the comment text and please use your PERSONAL name or initials and not your business name, as the latter comes off like spam. Have fun and thanks for your input.
Join a growing community. Every Friday I share the most recent insights from what I have been up to, directly to your inbox.
