When AI Does the Thinking First, Students Stop Doing It at All

We are solving the wrong problem too efficiently.

Give a student a complex text today, and within seconds AI can:

• summarize it
• simplify it
• translate it
• even generate a response

Everything becomes faster. Cleaner. Easier.

And that’s exactly the problem.

Multilingual learners don’t struggle because they lack tools.
They struggle because language development is hard work.

It requires:

• holding meaning in your head while decoding structure
• trying words that don’t quite fit yet
• building sentences slowly, sometimes awkwardly

That process is not a barrier to learning.

It is the learning.

But now we’re inserting AI at the very beginning of that process.

Before the student reads—there’s a summary.
Before the student wrestles—there’s a simplified version.
Before the student forms a response—there’s a generated one.

So what exactly is left?

Completion without construction.

Participation without depth.

Research in second language acquisition has warned us about this for decades.

Jim Cummins emphasized that academic language develops through sustained engagement with complex ideas—not through constant simplification.

Merrill Swain showed that learners grow when they are pushed to produce language, not just consume it.

AI, when misused, interrupts both.

Not loudly. Not obviously.

But consistently.

In Texas classrooms, this tension is sharper.

We are accountable for access.
We are accountable for outcomes.
We are accountable for growth in both language and content.

So we scaffold.

We support.

We adjust.

All necessary.

But now we’re facing a different question:

What happens when support arrives before thinking even begins?

Because that’s where AI quietly changes the game.

Not by replacing teachers.
Not by disrupting systems overnight.

But by shifting when the work happens.

Or more accurately—

By shifting who does the work.

This is where instructional clarity matters more than ever.

Support should:

• position students to think
• sustain cognitive demand
• build independence over time

But when support:

• simplifies too early
• structures too tightly
• answers too quickly

…it starts doing the learning instead of the student.

And the hardest part?

It looks effective.

Students finish faster.
Teachers feel supported.
Lessons move smoothly.

But underneath, something important is thinning out.

The stretch.
The effort.
The formation of language.

This isn’t an argument against AI.

It’s an argument for precision.

Use it:

• after students attempt the work
• to extend thinking, not replace it
• to refine language, not pre-build it

Because in the end, the question is not:

“Did the student complete the task?”

It’s:

“Did the student actually do the thinking required to grow?”

And that’s a question no tool can answer for us.

References

• Texas Education Agency. (n.d.). Guidance on instructional programs for emergent bilingual students and technology use.
• Jim Cummins. (2000). Language, Power, and Pedagogy.
• Merrill Swain. (1985). Comprehensible Output Theory.
• Digital Promise. (2023–2024). AI in Education Guidance.
• EdReports. (2023). Instructional Quality in the Age of AI.