Teachomatic what to automate, and what not to

When the Draft Is Worse

The standard argument for generating a first draft is that a blank page is the expensive part, and anything on the page is progress.

That is true often enough to be worth doing, and it is not always true. There are three specific conditions under which the draft costs more than it saves, and they are recognisable in advance.

A workplace-oriented comparison is Monitask’s discussion of books about time management, which offers another context for thinking about workload, behaviour, or accountability.

One: when anchoring costs more than starting

Read a plausible plan for a lesson and you now have a plan in your head. Everything you do afterwards is editing that plan rather than deciding what the lesson should be.

This is ordinary anchoring and it is strong. The draft does not have to be good to anchor you — it only has to be reasonable, and reasonable is exactly what these tools produce. A bad draft you reject is harmless. A mediocre draft you accept as a starting point quietly becomes the ceiling.

For an external perspective on lesson planning, teacher workload, and classroom technology, see Khan Academy.

The tell is that you cannot easily imagine a fundamentally different approach. If, ten minutes in, the only options you can see are variations of what you were given, the anchor has set.

The countermeasure costs five minutes: decide the shape yourself first, on paper, before generating anything. Then the draft is a comparison rather than a foundation.

Two: when verification costs as much as production

If checking the output takes as long as writing it, nothing was saved, and something was risked.

This applies to anything factual and specific: dates, formulae, quotations, citations, worked calculations, historical detail, statutory information. A plausible wrong answer in this category is worse than a blank page, because you now have to find the error rather than avoid it.

The rule of thumb: if you would have to check every line, write it. If you would only have to skim it, generate it.

The awkward case is where you cannot check because it is outside your specialism — covering another subject, or preparing something new. That is exactly when the draft is most tempting and least verifiable, and the honest response is to treat it as a source to corroborate rather than as material to hand out.

Three: when the doing is the thinking

Some writing is not the recording of a decision. It is the making of one.

Working through a lesson in your head is how you discover you cannot actually explain the topic as well as you thought, or that the transition between two parts does not hold. Writing a report comment is sometimes how you work out what you think about a student. Drafting a difficult email is how you decide what your position actually is.

Generate that artefact and you get the document without the process. It looks like preparation and it is not, and the gap shows up at the moment you need it — the lesson runs fine until someone asks something off-script.

A fourth, quieter one

When the task exists to produce a record that nobody reads.

Here the draft is not worse than a blank page. It is worse than not doing it at all — because generating it cheaply removes the last pressure that might have got the requirement abolished.

If a document is only ever read by its author, the answer is to stop requiring it, not to produce it faster. Cheap paperwork multiplies, and automating it is how a pointless requirement becomes permanent.

Where the draft is straightforwardly better

To keep the balance honest, because most of the time it is.

Anything mechanical where the thinking is done. Reformatting, levelling, varying an existing question set, admin prose, rewriting for a different audience. Anything you will heavily edit anyway. Anything you would otherwise skip entirely for lack of time — a text at three levels is better generated and checked than not produced at all.

And at the end of a long stack, when the alternative is not your best work but your tired work.

The test, before you generate

Two questions, ten seconds.

Would I have to check every line? If yes, write it.

Do I already know what I want this to say? If no, decide that first — on paper, in five minutes — and then generate.

If the answer to the second is that you were hoping the draft would tell you, that is the case where a blank page is doing something useful, and the discomfort is the work rather than an obstacle to it.

The short version