Since earning my doctorate at Stanford several decades ago, I have worked with universities as a teacher, a lecturer, and an advisor to their startup programs. This companion piece applies the argument of the When Software Gets Cheap series to those institutions.
The series asks who keeps the economic value when software becomes cheap to produce. Its short answer is that the barriers to creating software are falling faster than the barriers to building a durable software business.
A university needs a second question as well. It exists to educate, to create knowledge, to serve the public, and to support economic development, and capturing financial surplus is rarely its main purpose. So the question is also how a university can use cheaper software to create more value for the people it serves.
This piece's answer is that as AI lowers the cost of producing software and other knowledge work, universities may become more valuable for what they alone provide: new knowledge, validation of what is true, professional judgment, and the preparation of people who take responsibility for consequential decisions.
A university plays four roles at once
A company usually meets cheaper software in one role. A university meets it in four, often through different offices with different incentives.
Role | What cheaper software changes | The strategic question |
|---|---|---|
Technology operator | Internal applications become easier to build, while governance and integration stay hard | What should the institution build, buy, and control? |
Research and innovation partner | Code becomes less scarce than research expertise and specialized facilities | What does the university contribute to industry partnerships? |
Startup incubator | Companies can start with less capital, while conventional venture funding may be harder to win | How should the university support and evaluate startups? |
Educator | AI changes which technical and professional capabilities stay distinctive | What should students learn, and how should they be assessed? |
The university as technology operator
Cheaper software lets a university do things it could not afford before. Student services can become more personal. Faculty can build specialized research tools without large software budgets. Smaller institutions can build capabilities that were once available only to large, well-funded universities. These are possibilities rather than established outcomes, and most campuses can already reach them.
The governance challenge is particular to higher education. Universities are decentralized. Research labs, departments, faculty, and students choose and build their own tools, often for good reasons. Coding agents lower the skill needed to build an application without lowering the skill needed to run one safely. The question is how to govern AI agents connected to university systems without preventing the experimentation a university exists to encourage.
Institutional risk. Someone has to answer when an application built by a lab, an office, or a student reaches university systems. A light path into production, with a review before a tool touches institutional data and a named owner afterward, settles that question without forbidding the building.
Information stewardship. Student records fall under FERPA, and research data often carries privacy, consent, sponsor, or export-control obligations. Controls work best when they sit with the data itself, so that a tool built in an afternoon cannot reach what it should not.
Commercial control. Before building agents over a vendor's platform, such as a student information system, the institution should secure documented access to its own data at a known price for the length of the contract. In an October 2026 webinar, Gartner's Patrick Connaughton warned that an AI layer over a system of record can deepen a customer's dependence on it. Renewals with point-tool vendors are where the university gains bargaining power, and core platforms call for more care.
The university as research and innovation partner
Universities commercialize research through patents, software licenses, sponsored research, startups, and industry partnerships. AI may change the relative economics of these mechanisms. For software-intensive research, handing a company a finished codebase may become less valuable when the company can reproduce much of the functionality itself.
University software keeps legal and practical value. Patents, copyrighted implementations, and specialized scientific algorithms remain protected, and reproducing a program's functions does not remove licensing obligations. What is shifting is the relative value of the different parts of commercialization. Industry can increasingly produce code on its own. It cannot easily reproduce what surrounds the code:
Faculty expertise and the judgment that comes with it
Validated experimental methods and benchmark results
Specialized instruments, facilities, and test environments
Permissioned datasets and the knowledge needed to interpret them
Multidisciplinary teams that companies rarely assemble internally
A pipeline of graduate students and postdoctoral researchers
Freedom to explore hard problems without near-term commercial pressure
Data needs particular care. Research data often cannot be licensed freely, because consent, privacy, sponsor agreements, and institutional policy limit its use. Its value usually reaches industry through collaboration with the researchers who collected it.
As software development becomes cheaper, university-industry partnerships may focus less on transferring a finished piece of code and more on capabilities that are hard to reproduce. Universities that organize those capabilities into sustained partnerships, such as industry-funded research centers with member companies, may capture more value than those relying mainly on individual licensing transactions. Intellectual property remains. What changes is where industry puts its money.
This points to a larger possibility. AI may make universities more valuable as research partners even as it makes some research outputs less scarce. As AI produces more plausible analysis and code, the ability to establish what is actually true, through experiment, peer review, and replication, becomes harder to come by and more valuable.
The university as startup incubator
Cheaper software lets students and faculty start companies with far less capital, and it lets an incubator support more teams on the same budget. At the same time, the series' piece on startups, The Squeeze on Software Startups, describes early-stage investors funding fewer conventional software companies and showing more interest in deep tech. Venture capital is also concentrating, a separate trend: AI companies took 86% of US venture dollars in the first half of 2026, and rounds of $100 million or more took 87.5%. Much of that money is likely going to AI model developers rather than to robotics or chip companies.
Many university entrepreneurship programs measure success the way venture capital does. The economics now support two distinct paths, and programs may need to support both.
Venture-scale startup | Capital-efficient business | |
|---|---|---|
Market | Very large | Narrow or specialized |
Pace | Rapid scaling | Sustainable profitability |
Defensibility | Hard-to-copy technology or position | Expertise and customer relationships |
Funding | Institutional venture capital | Bootstrapping, grants, customer funding, modest angel rounds |
Measure of success | Growth and exit value | Customers, revenue, profitability, durability |
Typical university fit | Research-intensive deep tech, such as chips and robotics | Student and faculty products for specialized markets |
The first path stays important for research-intensive companies whose technology takes years and significant capital to build. The second may become the more common route for students and faculty who can now build a specialized product without millions of dollars. Programs can support both by asking early what a team holds that stays expensive to copy, by preparing teams for funding beyond venture capital, and by counting customers kept and revenue earned alongside capital raised.
The university as educator
Writing code is becoming a smaller part of technical work, though unevenly. On narrow tasks, developers using AI tools have worked far faster in controlled experiments, while experienced developers on mature codebases have shown little gain or a loss. Across organizations, Google's 2025 DORA survey found AI associated with faster software delivery and still associated with less stable delivery. AI makes first drafts cheap faster than it makes finished, dependable work cheap.
As AI produces more capable technical and professional work, the central skill shifts from producing that work to evaluating it, supervising it, and taking responsibility for it. Graduates will need domain knowledge that tells them when an answer is wrong. They will need judgment about what to build or buy, and care for an organization's critical data. Above all, they will need habits of accountability, so they can show which agent acted and on whose authority.
This raises a question for every program: how to assess students when AI can complete the assignment. Assessments that ask students to critique, verify, and defend AI-produced work, and to explain the decisions they made along the way, test the capabilities that keep their value.
Practitioner disciplines such as law, business, and medicine are where these skills meet daily work, because those professions run on records, rules, and responsibility. An emerging technology curriculum that pairs technical fluency with those disciplines prepares graduates for the part of the work that stays expensive.
What universities provide when software is cheap
Across the four roles, the same pattern appears. Producing software and other knowledge work is getting cheaper. Generating new knowledge, establishing what is true, developing judgment, and preparing people to answer for consequential decisions are getting more valuable, and those are what universities exist to do. The institutions that organize around them may find their value rising as the cost of software falls.
Questions for university leadership
As technology operator. How can the institution give labs, departments, and students room to build with AI while knowing who owns each tool that touches student or research data?
As research and innovation partner. Which partnerships and licenses still assume that code is the main asset, and how could the university offer its expertise, facilities, and validated methods through sustained relationships instead?
As startup incubator. Does the program measure success only by capital raised, and what would a track for capital-efficient businesses look like?
As educator. Which courses teach students to evaluate, supervise, and answer for AI-produced work, and how are students assessed when AI can do the assignment?
Sources
The Future of Enterprise Applications: Navigating AI, SaaS Disruption and Vendor Lock-In, Gartner webinar with Patrick Connaughton, October 9, 2026
Q2 2026 PitchBook-NVCA Venture Monitor, PitchBook
The Impact of AI on Developer Productivity: Evidence from GitHub Copilot, Peng, Kalliamvakou, Cihon, and Demirer, February 2023
Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, METR, July 10, 2025
Announcing the 2025 DORA report, Google Cloud, September 23, 2025