The Last Mile Is Human

A personal essay on why educational infrastructure only starts to matter once it actually reaches teachers and learners — at the example of AiS.Chat.

In telecommunications, the “last mile” refers to the stretch of line that runs from the distribution box to the living room — and, depending on the technology, it is the most demanding part of the entire route. First the cable, whether copper or fiber, has to make it into the basement of the house at all. From there it continues into the room where the Wi-Fi router sits. And only from there does it become simple and wireless: to the laptop, tablet, smartphone, or the washing machine that can then send a push notification once the laundry is done and wants to be hung up.

When we talk about digital educational infrastructure, it is exactly the same: it’s the last few meters, the last mile, without which no infrastructure, no service, and no educational medium actually reaches use.

From connectivity to actual use #

Years ago, I once looked into what the internet connections of schools actually looked like in practice. These were schools that ran their own local school server — schools that were, by any measure, digitally engaged. And yet their internet connection was, on average, no better than that of an ordinary multi-person household.

Today that picture has likely improved a great deal. In Berlin, for instance, all state schools now have fiber connections. But even once we assume that schools generally have fast internet, working Wi-Fi, and enough devices, that still doesn’t mean that the offerings provided by, say, the state ministries of education actually arrive. Because in the end, technology only does one thing: it makes an offer available to potential users. Whether it gets used is decided elsewhere.

Using AiS.Chat, the AI chatbot, as an example, I want to show what it takes for offerings to actually be adopted sustainably and at scale.

Breadth and depth of features #

If an AI chatbot can’t functionally keep up with other chatbots — or if its existing features aren’t as good or as comprehensive — it loses out in comparison. That’s why the feature set of AiS.Chat keeps being expanded. Recently it gained built-in web search along with the ability to use tools and hold more agentic conversations. That matters, because users who are also familiar with other chatbots will otherwise quickly drift back to the one they know — regardless of the fact that it may actually be less attractive when it comes to data protection and terms of use. The same is true for design and usability: once a solution feels less smooth to use than the competition, switching is not far off.

Especially in the fast-moving AI space, it becomes clear how hard it is to offer a “competitive” feature set, because the world doesn’t stand still. Other providers increasingly offer functionality for local work as well (Cowork-like capabilities that reach directly into files and working environments, for instance). So it always has to be checked which of these innovations actually make sense in a school context — and which don’t.

Onboarding, training, and communities #

They exist: the technology pioneers, the early adopters. But every solution has to hold itself to the standard of working for everyone — of addressing the so-called long tail.

That applies before use, during use, and in between. Especially with new topics like AI, training is not just a nice-to-have; in many cases it is mandatory.

When someone starts using the chatbot, they should, on request, be guided through a built-in onboarding flow. Returning users, too, should be kept informed about new features and significant changes.

Around AiS.Chat, a first online community has already grown up on Signal. The way many engaged teachers and staff from different institutes share content and make material available there shows how shared work and community can emerge across state lines.

But it’s just as important that such communities are actively maintained, and that their suggestions and needs actually feed back into further development. The emergence of local communities — whether in the staff room or beyond it — can also be supported, for example through multiplier concepts.

Thinking in more open structures #

As discussed, among other places, in “Openness Needs an Operating System”, looking at a single offering in isolation falls short. Thinking through the last mile also means looking at where users actually are, and how they need to be met there. That includes, for example, prior knowledge in using AI tools, but also the teaching and learning environments in which users work — and are at home.

If teachers, for instance, spend a lot of their time in Moodle, then generative AI functionality should be available right there — with the context that already exists in that environment, but also with all the advantages of the AiS.Chat world, such as the planned access to media libraries and curriculum frameworks. The same applies to the technical last mile on the device itself, as I described in “From Platform to Classroom”: even the most capable offering doesn’t count for much if it doesn’t reach the place where teaching actually happens.

The last mile stays an ongoing task #

Fiber in the basement, Wi-Fi in the classroom, and enough devices are necessary. But they are only the distribution box. The actual last mile is decided where an offering can keep up with the competition, where people are guided through onboarding instead of being left on their own, where communities are carried and heard, and where solutions show up wherever teachers and learners already are.

This last mile can’t be laid once and then checked off. It has to be rethought, maintained, and adapted to new needs again and again. That is exactly what makes it an ongoing task — and, in fact, the decisive task of educational infrastructure.