Part 10 of an ongoing series on vibe coding in EdTech.

When Spearfish School District started using the Danielson walkthrough tool I’d built, the first thing they did was ask for changes. The tool began as a fairly plain Sheet that administrators use to record walkthroughs against the Framework for Teaching. Spearfish wanted their own set of questions instead of mine. They wanted teachers to be able to add their own evidence, so a walkthrough became something closer to a two-sided record than a report written about someone. They also changed some of the wording so it matched how they talk. None of these requests were unreasonable, and all of them were specific to how Spearfish runs observations.

It’s worth asking what else they could have done. A Google Form would have held their questions, but it has no good way to let a teacher answer back with evidence tied to a particular walkthrough. A survey platform like Qualtrics could stretch further, at the cost of a license and a set of options designed for every customer the company has. Neither one would have been built around the way Spearfish already talks about teaching. Some of what they asked for, including the teacher evidence piece, eventually made its way back into the version I share with other districts. I’ll come back to that, because it’s less simple than it sounds.

The second story is shorter, and I had nothing to do with it. A colleague of mine built a behavior tool that Spearfish also picked up. Their IT staff extended it on their own so that it connected with more of the people around a student, including parents and IEP team members, and more of the people involved were aware of what was happening. Nobody from our team had to help. They changed the tool to fit how they already coordinate around kids.

I’ve been reading James C. Scott, the political scientist and anthropologist who died in 2024, and his work gives me a way to say why these two stories feel different from a vendor rollout. I want to be upfront that I’m borrowing a lens here. Scott wrote about states, peasants, and forests, and not about software. But a good part of his argument travels, and he said himself that it applies to markets as well as governments.

What was Scott worried about?

Seeing Like a State (1998) opens with forests. In late eighteenth-century Prussia and Saxony, state foresters wanted to know how much timber the forests could yield and how much revenue that meant. A real forest is hard to count, so they simplified it. They cleared the underbrush, planted single species in straight rows, and managed the forest as a table of figures. The first generation of these forests did well, because the soil was still living off everything the old, messy forest had built up. The second generation declined badly, enough that the Germans coined a word for forest death. The simplification had made the forest easy to see from the center, and it removed the parts nobody had thought to count.

Scott calls this legibility. Institutions that manage from a distance need the world arranged so they can read it, which means standard units, fixed categories, and records that look the same everywhere. The trouble comes from what gets lost in the translation. Scott’s name for that is metis, borrowed from the Greek, meaning the practical knowledge people build up by doing something in a particular place over time. It’s the farmer who knows when to plant by the look of a certain field, or the teacher who knows which student needs a question asked twice. Metis is local and hard to write down, and schemes designed from the center tend to treat it as noise.

Scott is clear that this isn’t only a state problem. He describes large-scale capitalism as a force for homogenization and standard grids every bit as strong as any government, and that’s the part of the book I’m leaning on. A company that sells to thousands of customers needs those customers to be legible to it in much the same way.

I should mark where the comparison stops. Scott’s states could conscript and tax people, and no EdTech vendor can do that. The pressure works through procurement, defaults, and whatever fields a product decides to include. Scott also isn’t arguing against all standardization, and neither am I. Much of my own work is templates. The question he keeps asking is who designs the simplification, for whom, and whether the people living with it can push back, which is close to the question my working ethos for this thread is built around.

Why does a tool built for everyone have to be thin?

A vendor building a walkthrough tool for thousands of districts can’t build it for Spearfish. It has to build for a district that is the average of all its customers, and that district doesn’t exist. What results is what Scott called a thin simplification: categories general enough to fit everyone, which means they fit no one closely. The district-specific parts end up in an open notes field, in a settings menu, or nowhere.

The settings menu is worth a closer look, because it’s usually offered as the answer to this problem. It gives a district choices, but they’re choices the vendor anticipated. If what Spearfish needed wasn’t on the list, the options are to file a feature request and wait on a roadmap built around the largest customers, or to change the practice to fit the tool. In Scott’s terms, metis adapts to the scheme instead of the scheme adapting to metis. Over time, that’s how a platform’s categories become the district’s categories.

This isn’t a vendor behaving badly. It’s what serving everyone requires, and that’s also why it’s hard to fix from inside a product built at that scale.

What changes when the back is left open?

The tools I build for districts are Sheets with a config tab and the script visible to anyone who owns the copy. That started as a practical habit, so one tool could move between districts without my rewriting it. Reading Scott, I’ve come to think it matters for another reason.

Scott spends a good deal of time on measurement. Before states standardized weights and measures, units were often local. A measure of grain might vary by village or by who was paying whom, and those variations carried meaning for the people using them. State measures replaced them with units readable from the capital. A config tab is a small place where a district gets to set its own measures on a shared frame. Spearfish’s questions, their wording, and their evidence field are exactly the kind of local knowledge a thin tool leaves out, and in this case the tool bent around them.

In Two Cheers for Anarchism (2012), Scott makes a related point about judgment. Institutions that make every decision in advance take away people’s chances to use their own judgment, and judgment that doesn’t get used weakens. He tells the story of a Dutch traffic engineer, Hans Monderman, who removed traffic lights and signs from intersections and found that drivers paid closer attention to each other. A locked tool is a bit like the fully signaled intersection. An open one leaves some of the decisions to the people who are actually there, and the behavior tool shows what that looks like. Spearfish’s IT staff didn’t wait for anyone. They read the tool and changed it.

The conclusion of Seeing Like a State also offers a few rules of thumb for anyone who has to plan for other people, and they read almost like a spec for this kind of work: 1) take small steps, 2) favor reversibility, 3) plan on surprises, and 4) plan on human inventiveness. A pilot with one district is a small step. A Sheet the district owns and can copy, change, or walk away from is about as reversible as a tool gets, since there’s no contract to unwind. A config tab assumes I won’t know every local need ahead of time, and leaving the script open assumes someone local will think of something I didn’t. The parent and IEP connections are that inventiveness showing up.

Who actually opens the back?

It would be easy to stop there, with district-owned tools on Scott’s side and vendors on the other. The Spearfish stories are more mixed than that when I look at them honestly.

The behavior tool is the clean case, since local staff changed it with no help. The Danielson tool isn’t. Most of what Spearfish wanted went through me. They knew what they needed, and I made the changes. That’s a much shorter path than a vendor roadmap, and I’m one email away. But it still puts me in the middle. For that tool, I was the small center that local knowledge had to pass through before it became part of the tool, and a district that can only change a tool by asking me is less independent than the open back suggests.

Folding changes back in raises a harder question. When the teacher evidence feature went into the version I share with other districts, Spearfish’s practice improved the tool for everyone. That’s the reverse of the forest story, where the center flattened what was local. It’s also how metis gets turned into a standard: someone collects local practice, tidies it, and hands it out as a template to places that never asked for it. The next district gets teacher evidence as a default, and I’ve spent a lot of words on this site arguing that defaults are never neutral. The difference I’d point to is that the back stays open, so the next district can take the feature out or change it. Whether they will is a separate question, and I don’t have the numbers to answer it.

Scott would also remind me that metis isn’t always right. Local practice can carry local blind spots, and a tool that bends to a district’s habits bends to the bad ones as well as the good. The open back doesn’t decide which is which. People do.

What I’m left with is that the openness is real but uneven. Some districts will open the back and some will call me, and I don’t know yet which is more common or what it would take to shift that. If you’ve inherited a tool like this, or handed one off, I’d like to hear which way it went. You can reach me at licht.education@gmail.com, and there are more tools, articles, and resources at bradylicht.com.