Building for Meaning with AI

Scarce Things in Abundant Systems: Part 3

By now, most of us have seen some version of the Khanmigo headline. Sal Khan predicted an AI tutor revolution; three years in, he’s telling Chalkbeat that for many students it was “a non‑event”. They had access. The bot worked. They mostly didn’t use it.

You can treat that as a product story, or a personality story. But if you’re a system leader, there’s a more uncomfortable reading: we’ve hit a different scarcity. It’s not content anymore, and it’s not just attention.

It’s meaning.

1. When capability is abundant but meaning isn’t

In the Chalkbeat piece, Khan compares Khanmigo to a very capable person sitting quietly at the back of the classroom, waiting for students to walk over and ask for help. Some will; most won’t. Passive availability doesn’t generate motivation or even the right questions.

For system leaders, Khanmigo is almost a natural experiment:

• It had early access to powerful models.
• It sat inside a platform students were already using.
• Districts partnered, philanthropy flowed, and eventually the bot was left on by default because students weren’t voluntarily toggling it open.

And still, for a lot of students, “it was a non‑event”. I mean, if a state‑of‑the‑art, well‑integrated AI tutor with that much backing can’t clear the bar of “students want to use this”, then we probably need a different story about what’s binding.

In Part 1 we traced the arc:

• content scarcity eased with videos, MOOCs, open resources;
• attention scarcity emerged as phones, feeds, and platforms competed for focus;
• now, with AI, many systems have something like capability abundance: explanation and feedback on tap.

Khanmigo suggests that even when content and capability are abundant, engagement does not automatically follow. The missing piece is whether the work feels meaningful enough to bother.

2. What “meaning scarcity” actually is

“Meaning” can sound hand‑wavy, so let me be concrete. At system level, I’m talking about three very ordinary questions that students ask themselves every day:

• Why this? Does this task connect to anything I care about, or is it just another hoop?
• Why now? Does this fit into a story about my own progress, or is it a random demand?
• Who cares? Does my effort matter to anyone beyond this grade and this moment?

When those questions have weak answers, it is rational for students to conserve effort, even when powerful support tools are sitting right there. They might tap the AI occasionally. They’re unlikely to build a habit around it.

There’s a decent evidence base behind this intuition, even if nobody calls it “meaning scarcity”. Syntheses on school connectedness show that students who feel valued and part of the school community attend more, achieve more, and have better long‑term outcomes. Longitudinal work suggests that day‑to‑day feelings of connection to classmates and school predict academic engagement partly because they satisfy needs for autonomy, competence, and relatedness.

On the task side, studies of project‑based and authentic learning find that when work has some real‑world relevance, involves choice, and culminates in a product or performance that someone actually sees, intrinsic motivation increases — even when test‑score effects are mixed. Students talk about those experiences differently; they remember them.

Layer on top the literature on “second” and “third‑level” digital divides: once access to devices is in place, gaps open up in how students use technology and in the offline benefits they get from it. In other words, the divide shifts from hardware to meaning — who finds the tools significant enough to use in ways that change their trajectory.

So meaning scarcity, for me, is not mystical. It’s a system‑level way of saying: we haven’t done enough to make sure that the effort we ask from students feels connected, purposeful, and noticed.

3. Four levers for building meaning in a system

If meaning is the binding constraint, the interesting question for system leaders is not “Which AI tools should we buy?” but “What kind of system would make any of this feel worth using to a 14‑year‑old on a Thursday afternoon?”

I keep coming back to four levers. None of them is new. The slightly provocative claim is that if we don’t move these, the rest is theatre.

3.1 Relational infrastructure: who is structurally allowed to matter

Most writing on “relationships” in schools stays at the level of sentiment. From a systems perspective the sharper question is: who is structurally allowed to matter to whom, and with what continuity?

The connectedness literature is blunt: students who feel accepted, respected, and part of their school community attend more, achieve more, and are less likely to engage in risk behaviours. One recent study even found that connections with classmates and a sense of belonging predicted engagement more strongly than teacher–student connection, via basic psychological needs.

At system level, relational infrastructure is all the boring design work that decides:

• whether a student keeps the same form teacher long enough for that relationship to become a reference point;
• whether you have mechanisms (sociograms, PSR frameworks, case conferences) that make invisible students and fragile peer networks visible, and then do something about it;
• whether time with students is protected in schedules, or constantly cannibalised by “more urgent” business.

I had written about how AI can help here, but only if it serves those decisions. A system that lets a form teacher see, at a glance, how a student’s attendance, participation, and peer nominations have shifted over six months is using AI to strengthen an existing relational contract. A system that uses AI to blast generic wellbeing messages while tutors are timetabled to the minute is doing the opposite.

3.2 Purposeful work and products: where effort goes to live

One of the quiet tragedies of school life is how much student effort evaporates. Work is done, marked, and essentially buried. From a meaning point of view, that’s brutal. If nothing you make seems to matter to anyone, it is rational to down‑regulate your effort.

Project‑based and authentic learning studies line up with what many of us have seen: when tasks have some real‑world relevance, involve choice, and culminate in a product or performance that someone actually sees, intrinsic motivation goes up. In one PBL maths study, self‑reported intrinsic motivation jumped from 25% to 67% over a year for the project group.

System‑level levers here are blunt but useful. How much of your assessment diet is made up of thin, disposable tasks, versus work that has a life beyond the grade? Do you ever ask, at curriculum meetings, “Where in this year does a student produce something they could plausibly be proud of?” That single question is often more useful than another rubric.

AI is interesting here not because it can dream up “engaging tasks”, but because it can lower the friction of running thicker experiences. It can draft briefs and rubrics, generate examples, help students research and revise, and cut some of the logistical overhead that often kills ambition. Used that way, AI is a scaffold for more purposeful work, not a generator of more busywork.

3.3 Coherent narratives: making school add up to something

Most students don’t experience school as a curriculum map. They experience it as “this week’s stuff”. Part of meaning is being able to say, in some rough way, “This is what I’m getting better at,” and “This is where this is going.”

Research on motivation and school connectedness points to the importance of coherent stories: students engage more when they can see progress and understand how current demands fit into a longer journey. System leaders shape that story every time they make what looks like a technical decision:

• changing a syllabus without thinking about what that signals about what is valued;
• adding a new initiative without retiring an old one, so the visible story becomes “we always add, never let go”;
• designing reporting so that all a student ever sees is a list of marks, not a trajectory.

AI is not going to invent a better story for you, but it can help you see where your current story is incoherent. Curriculum teams can use it to trace where concepts recur, where assessments bunch up, and where students are likely to experience sudden difficulty spikes. That’s not glamorous work. But it’s the kind of work that quietly shifts school from “a sequence of unrelated tasks” towards “a place where my effort is going somewhere”.

3.4 Noticing and follow‑through: whether signals go anywhere

The last lever is almost insultingly simple to name and perversely hard to fix: what happens when something changes? A student’s attendance drops. A usually engaged student starts handing in half‑empty work. A quiet student becomes disruptive. Does the system hold that signal long enough for anything to happen?

Early warning systems and learning analytics show we can predict dropout and disengagement reasonably well using attendance, behaviour, and grades. The problem is not the math. The problem is that in many systems, the signal goes into a void.

AI can make it cheaper to notice and remember. It can aggregate small changes across subjects, flag patterns a single teacher might miss, and condense long case histories into something humans can digest. The real opportunity is not “better prediction”. It’s the chance to move from a culture where follow‑through depends on a few heroic adults to one where the system itself expects to do something when the data shifts.

But — and this is the caveat we keep coming back to — if you don’t change anything about who meets, how often, with what authority, then more alerts just mean more things to ignore.

4. What AI is actually good for in a meaning‑scarce system

So if meaning is the actual bottleneck, what is AI really good for at system level?

I don’t think the most important answer is “more personalised explanations” or “less teacher admin”, even though those are useful. The deeper benefit, if we choose to take it, is that:

AI can buy back attention for the parts of the work that only humans can do: making judgements, building trust, telling better stories about why learning matters.

This is a frightfully clear take — for the work that only humans can do.

4.1 From doing more to choosing what not to do

When AI genuinely takes a chunk out of the grunt work around planning, marking, or reporting — early pilots talk about saving several hours a week per teacher — you can either:

• use that capacity to generate more tasks, more interventions, more data, or
• deliberately leave some space unfilled, and spend it on work that deepens meaning.

The second option is harder to defend on a dashboard, but it’s the one that gives teachers time for conversations, calls, and adjustments to tasks — all the small, human moves that make effort feel seen. AI’s value here is not speed for its own sake; it’s creating the conditions for restraint.

4.2 From individual heroics to system noticing

Right now, a lot of “care” in schools relies on individual memory and conscience. The form teacher who happens to spot a pattern. The year head who just knows twelve different backstories. It’s fragile and unfair.

AI can’t make us care, but it can make caring less dependent on luck. Used well, it can keep a quiet eye on basic indicators over time, nudge us when a student’s pattern changes in ways that are easy to miss, and make sure the story is on the table when adults meet. The point isn’t that “AI will fix pastoral care”. The point is that it can help us move towards a system where noticing isn’t just the job of whoever happens to be most conscientious this year.

4.3 From thin tasks to thicker experiences

Finally, AI is more interesting when it makes it easier for schools to sustain richer learning experiences, instead of just making thin tasks faster.

If you want more students to experience projects, inquiries, and performances that actually have an audience, someone has to pay the setup cost: writing briefs, managing logistics, supporting students through the messy middle. AI can’t remove that entirely, but it can lower the activation energy — drafting materials, suggesting timelines, helping students research and revise.

That doesn’t magically make those experiences meaningful. But it nudges the feasibility curve in the right direction. And in a system that has been living on thin tasks for too long, that’s not a trivial shift.

5. A different starting point for system strategy

It’s very easy, especially in a policy outfit or HQ, to get drawn back into the old pattern: find the next impressive tool, run a pilot, publish some before‑and‑after numbers, repeat. I don’t think that’s evil. I just don’t think it’s where the real constraint is anymore.

The Khanmigo moment is a useful pause button. It suggests that:

• content and explanation are still necessary, but rarely the limiting factor in well‑resourced systems;
• attention is fragile, and needs deliberate architecture;
• and meaning — belonging, purpose, coherent narratives, follow‑through — is now doing a lot of the heavy lifting.

If that’s roughly right, then the starting point for AI strategy is not “What can this model do?” but “Where, in our system, are attention and meaning currently failing, and what are we willing to change to address that?” Only then, I think, is it worth asking what kind of AI would help.

AI can remove excuses. It can speed up some work. It can give us new kinds of feedback. It cannot, on its own, manufacture meaning. That is still our job, as system leaders and educators. The tools either help us do that work, or they distract us from it.

Further reading

1. Meaning, belonging, and engagement

• Strengthening School Connectedness to Increase Student Success — EdResearch for Action (2024) https://edresearchforaction.org/research-briefs/strengthening-school-connectedness-to-increase-student-success/
• Longitudinal Effects of School Connectedness on Adolescents’ Academic Engagement (2025) https://www.sciencedirect.com/science/article/abs/pii/S0190740925001288

2. Purposeful work and where effort goes

• Action Research on Impact of Project‑Based Learning on Intrinsic Motivation (2021) http://www.clausiuspress.com/article/1259.html

3. Digital divides beyond access

• The Digital Divide in Online Education: Inequality in Digital Readiness and Outcomes (2022) (Pretty interesting read explaining first‑, second‑, and third‑level digital divides and shows why access alone doesn’t guarantee benefit) https://pmc.ncbi.nlm.nih.gov/articles/PMC9435462/
• Bridging the Third‑Level Digital Divide (2024) http://sbbwu.edu.pk/journal/Winter%202024,%20Vol.%2018,%20No.%204/2%20Bridging%20the%20Third-Level%20Digital%20Divide.pdf

4. AI tutors and human‑centred design

• Human‑Centred AI for Student Engagement and Academic Efficiency — IACIS (2025) https://iacis.org/iis/2025/1_iis_2025_324-337.pdf

5. AI, workload, and “buying back” attention

• AI Reducing Teacher Workload: Case Studies — UNOWA (2026) https://www.unowa.eu/blog/ai-for-teacher-workload-reduction-case-studies
• AI and Teacher Workload in Schools: What Evidence Says — Karimi (2026) https://www.linkedin.com/pulse/ai-teacher-workload-schools-what-current-evidence-means-arafeh-karimi-x5g6c