Recently I kept coming back to the same uncomfortable truth, scaling AI in a company is not primarily an engineering problem. You can throw cloud instances at it, spin up more GPUs with five clicks, and call that “scaling.” But people won’t suddenly hand over their calendars to a bot because a load balancer got happier. The real work happens in the human layer, the bits between inspiration and adoption, where trust, culture, and plain curiosity live.
We all live in a little AI bubble sometimes, where everything smells like model architectures and hyperparameters, and then you speak to someone in a restaurant and they say, oh yeah, I know about ChatGPT. As Ben put it, “it’s very easy for people who are not in this bubble” to feel left behind. The remedy is not more slides about embeddings, it’s showing tangible examples that matter to people in their day to day.
That’s why internal showcases work. You don’t convince everyone by lecturing about possibilities, you show someone like them using the tool to solve something concrete. Ben explained the tactic plainly, we “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.” Seeing a peer use a chatbot on the exact documents you wrestle with lowers the barrier to try it yourself. Inspiration seeds curiosity, curiosity grows skills.
You can be an AI nerd and still be useless to a business if you cannot translate. The most common question Ben gets when he talks to teams is, “what are your tasks?” He points out, “I don’t know what your tasks are.” Translation is about asking the right questions, digging past the symptom, and mapping problems to sensible solutions. If the business owner says they need a report, the translator asks why, checks whether Excel already does it, and only then suggests a GenAI path if it genuinely adds value.
This is where science backgrounds shine, when used as translators rather than ivory tower monologues. If your technical people are stuck explaining models in complete abstraction, adoption stalls. If they learn to say, in plain language, here is how this will change your week, things move.
Culture shifts happen fastest when you combine pressure from above with momentum from below. Ben recommends convincing your board first, “you need to convince your board of directors that this is a technology that you need to be ready for and we need to prepare the workforce for,” while simultaneously empowering grassroots users with tools and training. And if you get a high-ranking manager to use the tech in a live demo, adoption explodes, because, as Ben put it, “if the boss of the boss of the boss is using this technology, then maybe it’s okay if I use it too.”
Legitimacy matters more than policy theater. If people feel sanctioned, they will try the tools, ask questions, and iterate. If you ban everything and do nothing constructive, you just create a secret economy where people use unsafe tools under the table, and that is where real risk lives.
Trust is fragile, and once broken it is surprisingly hard to rebuild. Ben used the Samsung leak as a turning point, when companies were suddenly blocked from using public models, and employees “were now told they’re not allowed to use it.” That created a demand for safe, internal alternatives. Rebuilding trust requires more than PR, it needs transparent, cross-functional engineering that satisfies security, legal and privacy teams while still enabling real use.
This is why sensible guidelines are crucial. People want to know what they can put into an LLM without setting off alarm bells. Give them that, plus hands-on training, and you reduce risky behavior and increase productive experimentation.
Yes, sometimes you should not use GenAI. Ben’s blunt practical rule is useful, “don’t use it to solve problems which you can already solve with existing technologies.” If your task is deterministic, and Excel or a small script already does it reliably, throw GenAI at it and you invite chaos. A large model is a sledgehammer, great for certain jobs, absurd for others.
That said, GenAI is evolving. It can now orchestrate calls to deterministic tools, write and run Python snippets, and act as a high level coordinator. Treat it as a smart foreman who decides whether to use the saw or bring in the CNC machine.
Scaling a product technically is easy compared to scaling adoption. You can auto-scale infrastructure tomorrow, but you cannot auto-scale trust. Ben describes the solution as two parallel tracks. One team builds secure, scalable infrastructure. Another focuses on human adoption, with guidelines, training, showcases, and structured events like internal datathons where business problems meet developer curiosity. Those datathons act like an internal laboratory, they turn vague needs into concrete, funded projects.
Also, remember curiosity beats FOMO every time. If your people are only reacting out of fear, results will be patchy. Better to spark genuine curiosity by showing meaningful wins, than to herd everyone into panic-driven pilots.
The single biggest multiplier is integration into existing workflows. People rarely want to redirect their lives into new apps. Embed the intelligence where they already work, whether that is Microsoft Teams, Excel with Copilot, or the CRM they open every morning. Ben’s point was simple, when these tools are accessible in the software people already use, adoption follows. And add permissioning that mirrors what a person would allow an assistant to do, then you go from copy-paste into actual delegation.
We are close to a future where you might not need to stare at screens as much. That’s not a fantasy, it’s an integration problem. Solve the UX, solve the policy, and people will adopt because the tools are useful, not because a manager told them to.
Scaling AI is about more than models and GPUs, it’s about humility, curiosity, and human systems that actually fit into people’s work lives. If you want real adoption, focus first on the story, the translation, and the trust. The rest is plumbing.
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.
