Question for the room, and I want specifics rather than opinions.

A study of 76,485 students found that AI use followed their existing study schedule instead of creating a new one. Peak hours were weekday late morning to afternoon. Heaviest day was Tuesday. Students with jobs shifted to evenings.

So: what is your block?

Name the day and the hours where you reliably do focused work. Not the block you wish you had. The one that held last week.

Then tell us whether any AI practice actually lives inside it, or whether every tool you have tried is sitting outside your week waiting to be invited in.

I will go first. Mine is early morning before anyone needs anything, and the practice that stuck there is drafting, because that is what I was already doing at that hour.

Today's post has the numbers: https://www.harvestkernel.com/ai-joined-their-schedule-76000-student-study/

Where did your AI line get drawn, and who drew it?

Most AI policies are one page and cover every task at once. Manchester replaced theirs with a four value scale set per assessment. Stellenbosch would not write one rule at all.

Here is today's question, and it applies whether you teach or not.

Take one task you own. Not the whole job, one task. Where does it land?

Prohibited. No AI on this at all.
Minimal. Narrow uses only, copy editing for example.
Permitted. Named uses, spelled out.
Required. Using AI is part of the work.

Now the part that costs something: write the reason in one sentence, and make the reason connect to what the task is supposed to prove about the person doing it.

Post your task and your value below. I will argue with a few of them, in the useful way.

Today's piece: https://www.harvestkernel.com/ai-policy-four-settings-not-one-rule/

Adoption is finished. Fluency is not.

Two numbers from the same survey this month, 18,114 faculty across 35 countries:

77 percent of faculty use AI in their teaching, up 16 points in a year.
17 percent describe themselves as advanced or expert users.

Sixty points of daylight between holding the tool and being good with it.

And a third number that should sting a little: only 29 percent of students believe their instructor is equipped to guide them on AI. In the US and Canada, 17 percent.

Here is what I think is actually going on. That gap is not an access problem, because everyone already has access. It is not a policy problem, because most institutions have three of those. It is a practice problem, and we keep trying to fix it with the one format that research says does not work.

Darling Hammond, Hyler and Gardner reviewed effective professional development for the Learning Policy Institute in 2017. Seven elements. The seventh is sustained duration, and their language is blunt: strong programs run over weeks, months, or academic years, rather than short one off workshops.

We knew that. Then AI showed up and we booked the August in service day anyway.

So today I taught the alternative. I call it the Standing Row, because a farmer who tries to fix the whole field in one season fixes nothing. He picks one row.

Seed, week one: pick one task you repeat every single week. The boring recurring one.
Sprout, week two: run it twice on real student work. The first rep teaches you the tool, the second teaches you your judgment.
Grow, week three: write down where the reclaimed time goes before it arrives, and point it at a student.
Harvest, week four: show one colleague and record their objection.

About 105 minutes across four weeks. It runs on a free chatbot, on purpose.

The full kit with all 15 working fields is here: https://www.harvestkernel.com/standing-row-faculty-kit/

Tell me in the replies: which row would you pick, and what is currently eating the hour you would get back? I will answer every one.

Quick experiment from today's post. Think of the last ten things AI told you that you actually used. A number, a source, a summary, a recommendation.

How many did you verify against anything? And of the ones you checked, how many held?

Most of us cannot answer either question, and that second number is the one that decides whether AI is making you faster or just more confident. A randomized trial in JAMA found physicians with GPT-4 scored two points better than physicians without it. The model alone beat both groups by sixteen. The gap was judgment, not access.

Drop your honest guess at your own miss rate below. I will post mine after I run today's log.

https://www.harvestkernel.com/ai-foundational-knowledge-core-grammar/

Teaching Day question, and I want your actual number.

Open your syllabus. Control F. Count the sentences that begin with "You." Count the ones that begin with "We."

Mine was 31 to 4 the first time I ran it.

That ratio is measurable. Gurung and Galardi tested it on 257 students in 2021: same course, same policies, warm tone versus cold tone. The warm version raised help seeking in three of five scenarios. An explicit "please reach out to me" sentence moved only one of five. The tone of the document beat the sentence that invited help.

The frame from today's video. Your syllabus gets read four times, by four different people:

Compliance asks is it legal.
The nervous student asks is it survivable.
The skeptical student asks is it fair.
The hiring manager, six months later, asks was it worth it.

You write for the first one because it is the only read with a deadline attached. Your chair emails you in August. The nervous student never does.

The kit is live: https://www.harvestkernel.com/four-reads-syllabus-kit/

Four phases, about 95 minutes, works on one real syllabus you already have open.

Drop your You to We count below. Nobody is getting judged, I posted mine first for a reason.

Question for the room today.

43.5 percent of what people ask AI to do at work belongs to somebody else's job. That is from OpenAI reading 800,000 real messages, published this morning.

So: what is the one task you still hand off that you suspect you could now do yourself?

Not the one you want to do. The one you keep waiting four days for.

I will go first. For years I handed off anything that needed a real chart. Not a spreadsheet graph, a chart somebody would put on a slide. I can do it now, and the thing that changed was not the tool. It was finding out what questions a designer asks before they open the file.

Your turn. And if you have crossed a line and it went badly, post that too. That is the more useful story.

Context: https://www.harvestkernel.com/ai-task-crossover-job-titles/

Vanderbilt did the math on its AI detector and shut it off. At a one percent false positive rate, their submission volume meant about 750 students a year wrongly accused by a tool that cannot explain its own verdict. And the students it flags most are the ones who learned English second.

Here is the reframe I keep coming back to: integrity in the AI era is not a policing problem you buy software for. It is a design problem you solve once, at the assignment level. Name where AI is welcome, ask students to show their prompts and what they changed, grade the judgment they added on top, and let them defend one choice out loud. No detector required, and nothing to appeal.

So I want the real stories, not the headline version. Have you ever had a detector flag a student you knew was honest? What did you do, and what did it cost the trust in that room? And if you have already dropped detection, what did you put in its place?

Full post: https://www.harvestkernel.com/ai-detection-tools-failing-what-replaces-them/

Teaching Day question for the educators here.

Your students already use AI. Employers already grade them on it. The 2026 numbers are stark: about 88 percent of students use AI in their studies, only around a quarter of this year's graduates got any formal instruction on it, and AI skills now show up in more than one in ten job postings.

That gap does not close with a policy memo. It closes in one place: your course.

So here is the honest question. What is the one assignment you would redesign first so students show their AI work, the prompt, the output, and what they changed, instead of hiding it? Drop your discipline and the assignment below. I will build on the best ones.

Quick gut check for the educators here.

The new Instructure numbers say most of us already use AI in our work, but almost half of us never got a minute of formal training on it. When I read that, my first reaction was not guilt. It was recognition. We were handed the tool and told the training would come later, and later never showed up.

Here is what I keep coming back to: the confidence gap is not a character flaw. It is reps we have not done yet. A teacher who runs one assignment through an AI tool each week, watches where it helps and where it lies, and names that line for students is quietly building the exact judgment the research says is missing.

So I am curious about your real experience, not the headline version. What was the first moment you realized a student knew something about an AI tool that you did not? And what did you do with it, lean in, shut it down, or quietly go learn it that night?

No wrong answers. The honest stories are where the real training lives. Full breakdown in today's post: https://www.harvestkernel.com/ai-training-gap-students-educators/

AI did not break your course. It exposed a handful of assignments.

Teaching Day thought for the faculty in here. The most useful reframe I can offer about AI and your classroom is this: it is not a discipline problem, it is a design problem. A discipline problem means you police every student all semester. A design problem means you fix a few assignments once. The second one is a much better life.

Here is the four move redesign, SeedStacking style.

Seed, audit. List every graded task and rate each one green, yellow, or red against one question: could a student get a passing version from AI in under ten minutes, undetected? Most of your course is green. The reds cluster.

Sprout, reframe. Lift the outcome above what the model fakes. Turn summarize into critique and defend. Turn describe into diagnose and fix. Same content, higher altitude.

Grow, rebuild. Make the reds AI resilient (oral defense, live application, show the process) or AI integrated (grade students on out reasoning the AI). Rebuild one this week, not all of them.

Harvest, share. Pilot in one unit, measure whether it got harder to fake and whether students actually learned more, then hand your one page to a colleague.

The full kit with the audit grid, reframe verbs, and rebuild templates is here: https://www.harvestkernel.com/the-ai-ready-course-redesign-kit/

Tell me in the comments: what is your reddest assignment, and which rebuild path fits it?