At BETT 2026 last week, the conversation finally shifted.
For the last two years, the sector has been in a "honeymoon phase," dazzled by the parlor tricks of Generative AI. Walking the floor at the ExCeL London amidst the noise of 30,000 educators, the initial hype of "look what this can write!" has settled. It has been replaced by a more mature, critical, and urgent question: What is this technology actually for?
We attended two "mighty sessions" that perfectly framed this new state of play: a keynote featuring the Education Secretary and mathematician Professor Hannah Fry, and a deep-dive report launched by the Education Policy Institute (EPI).
The consensus? The era of "Generalist" AI is over. We need to move away from generic "junk food" tools and towards specialist infrastructure that solves real problems without bypassing the learning process.
1. The "Junk Food" Trap
In a fascinating discussion on the future of learning, Professor Hannah Fry offered a stark, sticky analogy that became the talk of the conference. She warned against the "junk food version of every relationship"—interactions that feel easy, instant, and frictionless, but ask "nothing of you."
This analogy hits hard in education.
- Nutritious Learning: Requires "productive struggle." The neural pathways are built during the moment a student wrestles with a complex equation or debates a thesis statement.
- Junk Food Learning: A student asks a chatbot to "write an essay on Macbeth." The output is instant and technically correct, but the student’s brain has done zero work.
As Fry pointed out, "the struggle is the bit where you learn." If AI removes that struggle, it removes the education.
The Strategic Shift: The goal for 2026 isn't to use AI to replace thinking (the "Junk Food" model). It is to use AI to support the humans in the loop. We need tools that handle the low-value administrative friction so that teachers and students have more energy for the high-value productive struggle.
Summary slide from the EPI talk, focusing on the takeaways
2. The "Enthusiastic Amateur" Dilemma
While the philosophy was being debated on the main stage, the Education Policy Institute (EPI) released a report highlighting the messy practical reality in schools.
They found that high-stakes decisions about technology often fall to "enthusiastic middle leaders". These are passionate individuals—often Heads of Department or Digital Leads—who are "AI champions" but lack the strategic support, budget, or time to properly vet every new tool.
The Consequence:
- Pilot Fatigue: Schools are drowning in free trials of "cool" tools that don't talk to each other.
- The Compliance Gap: "Enthusiastic amateurs" rarely have the legal expertise to stress-test a tool for full data compliance or bias, leaving the Trust exposed.
- No "Value for Money": Without a clear strategy, schools buy tools that look flashy but fail to deliver measurable time savings.
The EPI report is a wake-up call: We cannot build the future of education on the backs of tired volunteers. We need professional infrastructure.
3. Why "Vertical AI" is the Answer
This is exactly why we built ExamGPT.
The problems identified at BETT—the fear of "junk food" learning and the burden on "enthusiastic amateurs"—stem from relying on broad, Horizontal AI (like ChatGPT) that tries to be "quite good" at everything.
At Excelas, we champion Vertical AI: a specialist infrastructure trained on specific pedagogical data for a single high-stakes purpose—like accurate assessment.
How Vertical AI Solves the BETT Paradox:
- Preserving the Struggle: We don't write the essay for the student. We don't solve the math problem for them. Instead, we mark the result of their struggle. By automating the feedback loop, we validate their effort instantly. This encourages them to try again (reinforcing the struggle) rather than giving up because they have to wait 3 weeks for a grade.
- Empowering the Middle Leader: We solve the "enthusiastic amateur" dilemma by providing robust, evidence-based infrastructure. We don't offer a toy that requires constant prompting and tweaking. We offer a compliance-first engine with 90-98% accuracy. We give that "enthusiastic HoD" the rigorous data they need to walk into a SLT meeting and prove impact, turning them from an "amateur enthusiast" into a "strategic leader."
- Real, Invisible Inclusion: The sessions also touched on "invisible inclusion"—using tech to support SEND students discreetly. Vertical AI supports this by flagging learning gaps early and accurately, allowing for personalised intervention without the stigma of a student raising their hand to say "I don't get it."
The Bottom Line
As the Education Secretary noted in her closing remarks, the ultimate goal of any EdTech implementation is simple: "Free up teachers to spend more time teaching."
That is our mission. We aren't here to feed students "junk food" answers. We are here to clear the table so the real feast—teaching and learning—can begin.
We handle the data. You handle the inspiration.