The New Digital Divide: Agentic AI
Agentic AI is the future, but right now only a few can afford to learn it
I’m in a panic over the upcoming Fall semester. I’m committed to teaching my students the most current skills, but I can’t. Most of them don’t have access to what they need to learn: Agentic AI. AI agents are semi-autonomous systems that accomplish user-specified goals. They go far beyond chatbots, and only recently have become reliable and effective. Learning how to effectively initiate and manage agents is very clearly the skill of the future. Right now, they are most useful for coding, but that’s changing fast. Recently both Anthropic and OpenAI have released special agentic tools for finance. All the big labs are working on AI tools for medicine. Anthropic just released “Claude for Legal.” The list of domains where agentic AI will become an essential tool for professionals will only grow. Knowing how to use agents is what college students need to learn now, but unfortunately, access is limited to those who can afford it. It’s the latest example in a long history of classroom inequities.
Money has always given students an advantage. There was a time when only those with money could afford an education at all. Wealthy families can often choose to live in better performing school districts. Only those with means can afford private tutoring. Technology has famously created new income-based classroom inequities, dubbed the “Digital Divide.” Only some students can afford laptops, tablets, etc, while other students live in areas with limited internet access.
AI is the latest source of inequity. Both model providers and school districts are aware of this, and from the beginning have tried to mitigate it. All LLM providers have made a free tier available from the outset. Some have made higher levels of intelligence available to students for a relatively low cost. Some colleges have also given students access to more advanced AI through institutional subscriptions such as Google Workspace for Education, ChatGPT Edu, Claude for Education, etc. However, no provider (and no school I am aware of) is offering students the level of AI required for agentic uses. This is still out of reach to all who aren’t willing or able to pay hefty subscription fees.
At the time of this writing, the most effective agents are Anthropic’s Claude Code/Cowork and OpenAI’s Codex. Both are only really usable for agentic workflows at the $20/month subscription level, which is already a price tag that is inaccessible for many students. At that level they provide a limited amount of access in any given 5 hour window. In practice, I’ve found that they’ll cut you off in under an hour of sustained agentic use. For students facing a deadline, this can be the difference between something that is useable and something that isn’t. For relatively simple projects this level of access may be enough, but for anything like a final project that approximates a “real world” application, it won’t be. For that, a student would have to spend $100/month, far beyond what most can afford. At that level, students will be able to work continuously for several hours at a time. They can also learn to manage multiple parallel agents, learn to write agentic skills, manage agent tool access (e.g., MCP servers), etc. These are precisely the skills that employers are seeking right now.
So what can be done? The answer certainly isn’t to avoid teaching important skills simply because access is unequal. A chemistry department would not design a lab course around equipment that only some students could afford to bring from home. A better answer is for schools to budget for agentic AI access the way they already budget for journals, software licenses, lab equipment, and classroom technology. At the same time, companies should provide meaningful low-cost student access to their most capable tools, not just limited tiers that demonstrate the technology without allowing sustained use.
Agentic AI is becoming part of the basic toolkit of professional life. If we teach it only to the students who can afford ongoing subscription costs, we will reproduce exactly the kind of inequity that education is supposed to reduce. However, we should not lower the ceiling on what we teach because some students cannot yet reach it. We should raise the floor.
AI News Bits
Thinking Machines Lab, led by ex-OpenAI CTO Mira Murati, surprised many people by announcing new models specifically created for more realistic human-AI interaction.
Codex users can finally manage agents running on a desktop remotely through the ChatGPT mobile app. I’ve been expecting that for several weeks!
Anthropic launched Claude for Small Business, a package of agentic workflows for small companies that connects Claude to tools such as Canva, QuickBooks, PayPal, HubSpot, DocuSign, Google Workspace, and Microsoft 365.
Google released Gemini Intelligence on Android, a new set of AI features for Android mobile devices. I find this particularly interesting since Google also has the contract to power the next generation of Apple intelligence on iPhones, including an updated Siri.
Speaking of Google, the day this post comes out is Google’s I/O conference, where it is likely that they’ll make some big AI-related announcements. Check this section in next week’s post for the highlights!
David Bachman is a professor of Mathematics, Data Science, and Computer Science. He writes about AI and its real-world impacts. To learn more about his academic work, mathematical art, or AI speaking, consulting, and curriculum development, visit davidbachmandesign.com.