Two regrettable rules for junior scholars: AI supercharges my long standing advice (post 4 of 4)

AI doesn’t change the pre‑tenure advice. It makes it harder to rationalize ignoring it.

In the first three posts, I argued:

• Student collaboration is often negative expected value for pre‑tenure output (i.e. submitted papers) unless alignment is unusually strong.

• Trying to be an excellent teacher in years 1–3 can be a career risk; aim for “solid and improving.”

• Protect research days like your job depends on it—because it often does.

Now the add‑on:

AI tools don’t fundamentally change the incentive structure.
They change the alternatives, which makes the old advice harder to ignore.

1) AI dramatically lowers the cost of “solid teaching”

A big reason junior faculty over‑invest in teaching is time: the first draft is expensive.

AI is becoming unusually good at first drafts:

• reorganizing slides and lecture notes

• generating examples and alternative explanations

• drafting discussion questions and in-class activities

• proposing rubrics, practice problems, and checklists

For a deeper and very interesting dive: Let me send you over to scott cunningham’s great post on making beautiful decks and related posts here:

Claude Code Series (part 7): Making Beautiful Decks For My Future Self

Back to my points: If your goal is “solid and improving,” AI can often give you a workable starting point quickly.

That means the common justification—“I need 12 more hours this week to make lecture 4 perfect”—becomes less compelling when you can:

• generate 10 candidate explanations,

• keep the best one,

• edit and verify,

• and stop at “good.”

Important guardrail: you still own accuracy. AI can be confidently wrong. The right workflow is “draft fast, verify carefully,” not “copy-paste.”

If you really want to expand your thinking about using AI in the classroom: take a gander at Justin Wolfers :

2) AI is a substitute for many common RA tasks (and avoids the coordination tax)

Look at the tasks many people hope an RA will handle:

• cleaning data pipelines

• writing reproducible scripts

• producing exploratory plots and tables

• refactoring and debugging code

• drafting documentation

• generating test cases

• summarizing papers for triage

For a lot of these execution-heavy tasks, AI can be:

• faster (no waiting for calendar time)

• cheaper (often ludicrously cheaper than student support)

• lower coordination (little onboarding, no misaligned priorities)

This doesn’t mean AI replaces students as intellectual collaborators. It doesn’t.
But it does mean:

If you’re hiring a student mainly to get execution bandwidth, you now have a serious alternative.

And that alternative has the exact properties junior faculty need most: speed, availability, and predictable turnaround.

3) The “obvious” pre‑tenure operating system in the AI era

Here’s the simple synthesis:

For research:

• Use AI for execution support where it’s strong (with tests, replication checks, and sanity checks).

• Collaborate with students when alignment is high and deliverables are short and verifiable. I’ll go into this in more detail in a future post—but one idea is to only hire hourly RAs to reproduce (i.e. ‘by a human’) the findings that AI has already completed and you want to pursue

• Don’t build your tenure plan around “students will scale my output.”

For teaching:

• Use AI to compress prep time and reduce perfectionism.

• Aim for “solid and improving,” not “award-winning,” until your research pipeline is safe.

• Protect research-only days.

The point isn’t cynicism—it’s sequencing

AI makes it easier to do the responsible version of this strategy:

• teach competently without drowning in prep,

• make research progress without relying on high-variance collaborations,

• preserve your energy for the parts of academia that actually require human judgment.

Which leads back to the core message of this whole series:

Survive the gate first. Build the ideal academic life after.