From Prison Classrooms to AI Tutors: What Actually Matters in EdTech Now
Why teachers’ influence, focused platforms, and equal access matter more than another “engaging” feature
There’s a moment in Greg Margas’s story that won’t leave me.
Years ago, after leaving corporate life in Ireland and retraining as an educator, Greg found himself teaching creative writing in one of the largest prisons in the country. One of his students, Martin, was deeply depressed, waiting for his next court date and stuck in his own worst narrative.
Greg did something small and very human: he handed Martin a simple piece of jewelry as a tangible reminder that “it’s going to be okay.” Months later, Martin wrote to say how profoundly that gesture—and the learning space around it—had affected him. He wanted to return the jewelry.
Greg didn’t actually want it back. What stayed with him was the realization that education, at its best, is quiet but deep influence. It reaches people who’ve lost jobs, people in prison, students who have no idea where to go next. It can literally change a life trajectory.
That’s the standard against which we should be measuring our platforms, our AI tutors, our entire EdTech ecosystem.
Are we increasing that kind of influence—or just making nicer dashboards?
Engagement vs. overstimulation: when “cool” quietly kills learning
For Greg, the early problem he wanted to solve in e-learning was obvious: engagement. Completion rates were low, compliance training was a checkbox exercise, and online courses were browser tabs to be ignored until the deadline.
Today we’re “solving” this with a particular instinct: make learning platforms feel like social media.
Infinite feeds. Badges and streaks. Constant notifications. Chat everywhere. Micro-content designed to be swiped through in seconds.
On the surface, it works. These products look good in a sales demo. They feel familiar, especially to younger learners. But Greg’s worry is simple: we’re confusing stimulation with learning.
We’re taking learners out of one overstimulating environment—their phones—and placing them into another that is ostensibly “educational” but driven by the same dopamine loops. We might keep them clicking, but do we help them focus? Do we help them go deep enough to actually change competence, identity, or opportunity?
Greg isn’t against microlearning or smart UX. Short, well-designed units clearly have a place. His line in the sand is different: fewer “thousands of notifications, chats and games,” more tools that help learners cut themselves off from noise long enough to enter a flow state.
In other words: engagement that respects attention, not exploits it.
AI tutors are here. Teachers are not obsolete—but lazy teaching might be.
Greg has spent the last years implementing LMS platforms like Moodle, building AI-based assistants for universities, and closely watching tools like Khan Academy’s Khanmigo or Google’s new tutoring features.
His verdict on today’s AI tutors? “Maybe a six out of ten.”
When they’re well-trained on good data, they can do some genuinely valuable things:
Act as 24/7 mentors, available when the human teacher isn’t.
Guide rather than simply give answers, nudging students in the right direction.
Offer limitless patience and encouragement—even when a human would be exhausted.
He’s seen his own children benefit from AI-guided explanations that push them to think rather than copy. He’s tested new tutoring models that refuse to just spit out the solution.
But he’s blunt about the illusions we’re buying:
We don’t know whether AI will replace teachers—or which kinds of teaching it might replace. Repeating the slogan “AI will never replace teachers” is as naive as declaring “AI will replace everyone.”
Our expectations are often unrealistic. We want AI to be perfect, and it isn’t. Even internal models, carefully trained on an organization’s own data with strict policies, still hallucinate. Trusting them blindly, Greg learned the hard way, is a mistake.
We underestimate what human passion and presence actually do. Most people can name a teacher who shaped their lives—not because of a perfect curriculum, but because of a contagious love for a subject and an authentic relationship.
Greg’s baseline remains old-fashioned but true: “If a student doesn’t want to learn, he will not learn.” AI can personalize, assess competencies, and recommend paths. It can’t reliably create the inner desire to learn. That work is still profoundly human.
The future Greg sees is hybrid: modern teachers who know where to use AI—and where not to. Let AI handle some of the routine work, the constant availability, the gentle nudging. Let humans focus on creativity, values, and the kind of influence that can follow someone far beyond a course.
Is Moodle dead? Only if we treat it as a corpse.
Ask around and you’ll hear it: “Moodle is legacy. We need something shiny.”
Greg pushes back—but in an interesting way. He compares Moodle’s bad reputation to virtual reality in education. Many people’s first VR experience was dull, low-fidelity “educational” content. Their takeaway isn’t that VR is weak. It’s that their first encounter was weak.
Moodle suffers the same fate. A lot of people met it in poorly configured, outdated deployments. Of course it felt clunky.
Greg’s framing is different: think of Moodle as a core, a skeleton, surrounded by thousands of plugins and a huge open-source community. In his work with universities—from Sorbonne and ESSEC to leading institutions in Poland—he’s seen Moodles that look ancient and Moodles that feel modern and engaging. The difference isn’t the core code. It’s vision, updates, manpower, design.
For an individual lecturer, that’s a mixed message. You can’t single‑handedly redesign the platform; that sits with IT and leadership. You can, however, refuse to use it as a PDF graveyard. You can:
Move beyond file uploads into quizzes and interactive activities
Use tools like H5P or interactive video
Learn the course-format capabilities well enough to create structure and variety
It does require time, creativity, and what Greg calls “a bit of passion.” Teachers who’ve done things the same way for 20 years need to be willing to learn again—because, as he notes, “what I did seven years ago might not work nowadays.”
Moodle isn’t automatically legacy. Treating it as finished, instead of a flexible skeleton that needs life breathed into it, is what makes it obsolete.
Too many tools, not enough outcomes: a simple 12‑month playbook
Founders are building AI features into everything. Institutions are drowning in demos. Teachers are overworked and undertrained. The risk is clear: we replay the “interactive whiteboard” tragedy—huge spend, marginal learning impact.
Greg’s advice to education leaders is refreshingly unglamorous:
Stop buying for three months. Pause the hype cycle. “Don’t buy anything” is his first rule.
Define 2–3 measurable outcomes. Not generic “innovation,” but concrete targets: reduce dropout in year one, shorten time-to-competency, increase completion with real evidence of learning.
Audit what you already have. Many institutions are underusing tools they’re already paying for.
Run small, focused pilots. With one team or department, test specific tools against specific outcomes.
Scale what works, kill what doesn’t. Don’t keep tools because of sunk cost, emotional attachment, or slideware.
For EdTech founders, his message is parallel: pick one problem that truly itches you and solve it deeply. He cites examples like gamification or accessibility where companies have built solid businesses around a single, sharply defined wedge. Trying to “do everything AI + education” just fuels noise.
Our 10‑year test: fewer lost geniuses, more second chances
Behind all of this is a looming macro challenge: by 2030, a massive share of the workforce will need to be reskilled or upskilled. Greg cites estimates in the 39–54% range. Job losses linked to automation are already visible in some countries; Europe will feel it more soon.
In that context, Greg offers a simple way to judge ourselves in ten years:
Equal access. Does a kid in rural Poland have access to learning of the same quality as a student at Sorbonne? Not the same branding—but comparable depth, structure, and support.
Fewer “lost geniuses.” Greg loves the idea of the “lost genius”: people born with potential that never surfaced because they lacked access to formal education. He’s worked with platforms that bring free learning to underserved populations, especially in Africa. Are we shrinking that pool of undiscovered talent?
Dignified upskilling for the displaced. When someone loses their job to technological change, can they find a path to learn something new, discover they love it, and rebuild their confidence? Or do they just get nudged into generic content while their data is sold for ad targeting?
If, in a decade, we can honestly say that EdTech and AI helped us move those needles—more equal access, fewer lost geniuses, more people pulled “out of the darkness of losing a job” into new paths—then we’ll have used this moment well.
If not, all we’ll have done is recreate social media inside our learning systems and call it progress.
The prison classroom, the overworked lecturer, the laid‑off worker, the first‑generation student in a small town—they’re the real benchmark. Our tools, our AI agents, our LMS configurations either extend meaningful human influence into their lives or they don’t.
The choice is not between humans and AI, or between “legacy” Moodle and glossy new stacks. The real choice is whether we design for attention, belonging, and opportunity—or for metrics that look good while lives quietly stay the same.
Strong closing insight Design for influence, not impressions. If our tools don’t create attention, belonging, and a path to new opportunity, they’re just prettier tabs in a crowded browser.
Great advice for education leaders, and edtech founders shouldn't confuse obsession with the problem for obsession with the product.
Just because you can build another feature, doesn't mean you should!