For the final 2025 edition of Leading Indicator, we’re doing something a little different: We’re taking a look back at the last 12 months to round up the biggest trends regarding artificial intelligence (AI) in education, as well as setting out our predictions for trends to watch in 2026.

Education Evolution: AI and Learning

2025 was the year we stopped asking whether AI would change education and started asking how — and for whom. The year began with “deep research” models that think step-by-step, produce Ph.D.-level work in minutes, and force us to think about what skills are really critical for students to learn in an AI-native world. By spring, “vibecoding” entered the lexicon to describe entrepreneurs building apps with little more than descriptive prompts, and in turn we coined “vibe-agogy” to describe teachers doing the same with lesson plans and classroom resources. Our warning: Pinterest is already drowning in AI slop — educational resources could be next.

The cheating panic raged all year, but we pushed beyond the fear to bring more nuance to the conversation. In June, Bellwether released Productive Struggle, making the case that students need to struggle to learn, and that well-designed AI tools could amplify that struggle rather than eliminate it. How can we identify well-designed tools? Check out Productive Struggle’s sister report, Measuring AI in Education, for ways to draw connections between tech and impact.

By fall, the conversation darkened. We covered tragic cases of teens developing intense relationships with AI chatbots — including one that may have contributed to a suicide. The family sued OpenAI. Illinois banned AI therapy. And we returned to a warning we’d issued back in August 2024: Extreme caution is warranted when AI becomes part of children’s social and emotional development.

What to Watch in 2026:

In the year ahead, the idea of AI tools amplifying productive struggle rather than eliminating it will face a stress test on two fronts. First, the product problem: Are ed tech companies designing for deeper thinking, or superficial engagement? Will schools evaluate AI tools based on learning outcomes, or time saved? (Pro Tip: Keep an eye out for an upcoming set of Bellwether case studies on tools that do these things well!) Second, the harder underlying pedagogy problem: even if better products exist, students may keep defaulting to ChatGPT on their own — free, frictionless, and just a prompt away. Watch for whether schools move beyond policing AI use and toward redesigning assignments and instruction so that productive struggle feels worth it to students. The tools matter, but the teaching matters more.

Education Evolution: AI and Learning

Adoption numbers tell an interesting K-12 story: Nationwide, teacher AI training for K-12 staff nearly doubled (24% to 48% of districts), student ChatGPT use for schoolwork doubled (13% to 26%), but policy lagged — only 40% of districts had AI policies by spring 2025. In Florida, Miami-Dade County Public Schools captured the whiplash: from banning chatbots two years ago to rolling out Gemini to 100,000 high schoolers in May. 

Higher ed became a battleground for tech giants in 2025, too. Anthropic launched Claude for Education with Socratic-style “Learning Mode.” Google offered free Gemini access to college students through 2026. And OpenAI committed $50 million to university research partnerships. 

The year ended with big money and big promises. As the federal government largely sat on the sidelines, tech companies were eager to fill the vacuum. The American Federation of Teachers partnered with OpenAI, Microsoft, and Anthropic on a $23 million National Academy for AI Instruction for 400,000 educators. Microsoft committed $4 billion to its Elevate program. And Google pledged $1 billion for higher ed.

What to Watch in 2026:

AI use in K-12 expanded in 2025 alongside interest in the downsides of classroom technology and concerns about the decline of reading whole books. Keep an eye on a thermostatic reaction. In public schools, we expect to see more parent advocacy for less Chromebook time and more actual physical books. In the non-public school sector — especially in politically red states with education savings accounts or other private choice programs — watch for growing demand for technology-light (or tech-free) models like classical or nature-based private schools and microschools.

The Latest: AI Sector Updates

2025 AI model releases came fast: Claude 3.7, 4, and 4.5 with web search, GPT-4.5 scaling compute over thinking time, Llama 4, Mistral’s multilingual reasoning, GPT-5’s complete redesign of the ChatGPT ecosystem, Gemini 2.5 and 3 quietly outperforming competitors, and Nano Banana taking image generation to the next level. In October, Sora 2 powered a video-generation social app that hit No. 1 on the App Store within days driven in part by copyright-skirting videos, like Pikachu storming Normandy. We called it “hollow-scrolling:” less outrage than X, more oblivion than TikTok.

Agentic AI went mainstream, too. OpenAI’s Atlas and Perplexity’s Comet web browsers can shop or plan events for you. Claude got a Chrome extension. Manus.im promised fully autonomous work — close your laptop and come back to finished output. Meanwhile, safety concerns grew louder: OpenAI found that smarter models don’t misbehave less, and pressure to stop bad behavior can make them hide it. Similarly, Anthropic warned that reasoning models may “lie” in their chain-of-thought.

Meanwhile, the labor market signals were stark. Stanford University’s “Canaries in the Coal Mine” study found a 13% relative decline in employment for young workers in AI-exposed jobs since 2022, while older workers held steady. Entry-level tech hiring fell nearly 50% since 2019. LinkedIn’s chief economic officer warned that 63% of executives expect AI to take over entry-level tasks. The first rung of the career ladder is disappearing, and students furthest from opportunity will fall first.

What to Watch in 2026:

There’s growing talk of an AI “bubble.” AI firms, chip makers, and data center construction are driving a huge portion of economic growth in 2025 — but will it continue, or are we heading for a bust? The bulls and bears each have compelling arguments. Sooner or later, massive AI investment will need to pay returns. If things go south in 2026, it could have major economic consequences — and place additional pressure on schools already navigating the post-COVID-19 pandemic fiscal cliff.

Pioneering Policy: AI Governance, Regulation, Guidance, and More

The Trump administration repealed Biden’s AI executive order and promised a new approach. A spring executive order on AI in education called for a task force, a Presidential AI Challenge, and apprenticeships — but in a landscape of pandemic recovery and funding uncertainty, it’s unclear how much will break through. July’s America’s AI Action Plan outlined three pillars (accelerate innovation, build infrastructure, lead internationally) but zero education-specific initiatives. The federal vacuum was real, but there are efforts to fill that space: To round out the year, President Trump signed a December executive order aimed at creating “a minimally burdensome” national policy framework for AI regulation.

In the absence of federal action on AI regulation, states moved forward with their own efforts. More than 1,000 AI bills were introduced in 2025 across the country — up from 743 in 2024 and fewer than 200 in 2023 — and 136 became law. Ohio became the first state to require AI policies for all K-12 districts. California’s Leading Ethical AI Development for Kids Act focused on banning companion chatbots for minors. 

The biggest fight was federal preemption: the House reconciliation bill included a provision blocking states from passing AI laws for a decade. More than 260 state lawmakers from all 50 states sent a bipartisan letter opposing it. The Senate countered with a softer approach tying AI regulation to broadband funding. The debate remains unresolved as we head into the new year.

Data centers also became the new economic development prize in 2025. Amazon announced $20 billion for Pennsylvania. President Trump headlined an energy summit celebrating $90 billion in AI infrastructure commitments. Governors pitched data centers as job creators — but the Wall Street Journal found they don’t actually employ many people post-construction, and critics warned of environmental harms that boosters ignored.

What to Watch in 2026:

Although December’s executive order aimed to prevent the growing patchwork of state regulation, its effectiveness ultimately relies on a federal policy framework that will satisfy both sides of the aisle and the tech lobby. Except it’s rarely a bad bet to expect gridlock in Washington, D.C., especially with 2026 midterm elections on the horizon. If Congress lags on legislative action, watch how the Trump administration uses executive power to curtail the growth and scope of state-level AI regulation.