Decide what to delegate.
Every act of delegation costs something evidential, and the interface never itemizes the bill. The Proficiency Arc traces the movement from procuring output to informed refusal and scheduled reconsideration.
For English language teachers, teacher educators, and the people who train them
The English is good.That is where the professional question starts.
Beyond the Fluent Surface: Professional Judgment and AI in English Language Teaching. A working account of what generated fluency can and cannot evidence, and what a teacher does about it on a Tuesday.
Generative systems have not made fluency worthless. They have severed it from its producer. Treat that as a cheating problem and you will spend the year policing; treat it as an evidence problem and it changes how you assess, design, and protect professional judgment.
The professional question is no longer only “Is this good English?” It is also “What does this English warrant me in concluding?”
Chapter 1, When Fluency Is No Longer Enough01 / The argument
MachineoutputTeacher
judgment
The machine can have the sediment.
The scales stay with the teacher.
A working sequence
Four jobs, worked through in order
Behind every chapter is one governing claim: AI can reduce a teacher’s labor while quietly consuming the judgment that makes that labor professionally valuable.
Compression is not abdication, but the two are quite easy to confuse from the inside, and nothing in the interface distinguishes them for you.
Interrogate what comes back.
Fluent, plausible, and adequate are three different findings. The CAGE Protocol gives four checks a working teacher can actually run: Cutoff, Admission, Governance, Evidence.
Protect what crosses the boundary.
Learner data, protected material, institutional exposure, freelance exposure, and the widening gap between pasting text into a chat window and granting a system the ability to act.
Preserve the effort that forms expertise.
The chapters on deskilling and guardrails deal with the part nobody markets: the work that made you good at this was often the work you would now be glad to hand over.
02 / Terms of engagement
Read this before you buy it
Supply prompt recipes with a shelf life measured in months
Supply decision procedures that survive the next model release
Catalog tools
Interrogate categories of tool behavior, and say when the answer is no
Argue that teachers should adopt AI
Treat adoption and refusal as professional decisions requiring evidence
Present classroom trials it has not run
State the evidential status of its claims, including the weak ones
Promise time savings
Ask what the saved time was previously buying you
Where the evidence is thin, contradictory, or unavailable, the book says so. Stylistic confidence disguising evidential uncertainty is the failure it spends eleven chapters describing, so committing it would be a very strange way to finish.
03 / Inside the book
Contents
Four parts. Eleven chapters. A glossary, full references, and a note on how AI was used in the making of the book.
Part ISeeing Clearly+
When Fluency Is No Longer Enough
Why fluent output is insufficient rather than worthless, and why ELT faces a distinctive evidence problem.
Calibrating the Promise
Separating measured findings from extrapolation, historical hype, product claims, and professional inference.
The Proficiency Arc
From output procurement to specification, evaluation, informed refusal, and scheduled reconsideration.
Part IIInterrogating the System+
Two Different Animals
Systems that generate content against systems that can take consequential action.
The CAGE Protocol
Cutoff, Admission, Governance, Evidence. A practitioner screen with explicit limits.
The Perimeter
Contracts, protected material, learner data, credentials, and institutional and freelance exposure.
Part IIIDesigning Around Generated Fluency+
Assessment Beyond the Fluent Product
Process-visible assessment, linked stages, oral defense, calibrated permission, fairness, and repair.
Materials That Earn Their Place
Vetting generated texts, tasks, items, distractors, transcripts, and synthetic audio against ELT criteria.
Judgment in Practice: Three Cases
Three end-to-end reconstructions, including one that ends in a decision to constrain or refuse.
Part IVKeeping the Teacher in Working Order+
The Deskilling Question
Performance gains against learning gains, hidden labor, and why career stage changes the answer.
Guardrails for Judgment
Attempt-first work, acceptance criteria, human gates, visible disagreement, task sorting, and scheduled review.
“Back to the teacher at the kitchen table, and one changed decision.”
04 / The reader
Written for the person who has to decide
The classroom teacher
You are already using these systems, or already refusing to, and in both cases without much to reason from. The decision has to be defensible on Monday. It also has to be cheap enough to survive your actual week.
The teacher educator
You are being asked to teach AI literacy before the field has agreed what it consists of. This book separates operating a system from diagnosing what its output warrants.
The academic manager or DoS
You are writing policy that will be read as permission or prohibition, and neither is what you mean. Start with the chapters on assessment, the perimeter, and guardrails.
06 / Method
The book discloses its own method
Beyond the Fluent Surface carries a full account of how AI was used to research, brief, audit, and draft it, with the boundaries stated and the limits named. A book that asks teachers to interrogate their tools has no standing to be vague about its own.
Beyond the Fluent Surface
Still being revised.
Coming later in 2026.
The manuscript is in its final editorial stage. Confirmed formats, publication timing, and retailer links will appear here when they are real. There is no preorder button because there is nothing to preorder yet.
Return to the surface