How to Invent the Future I - Stanford CS183F: Startup School

Summary

Alan Kay argues that entrepreneurs seeking outsized impact should aim to create new industries, not merely improve existing products. Incremental innovation starts from today’s assumptions; invention starts with a compelling vision of a future that does not yet exist, then builds the technologies and organizations needed to make it real.

His central case study is Xerox PARC and the research ecosystem that preceded it. Kay credits that work with producing technologies that shaped personal computing and networking, while emphasizing that the results depended on long-term funding, strong researchers, tolerance for exploration, and the ability to build complete working systems—not just demonstrations.

Business and Organizational Playbooks

Use a Future-Back Invention Process

1. Start with a “cosmic goodness” intuition: a problem or capability important enough to pursue deeply.
2. Identify a favorable exponential trend, such as Moore’s Law.
3. Imagine the future roughly 30 years out and ask what would seem inevitable or absurdly absent.
4. Work backward to a 10–15-year horizon, when it may be possible to build an early version by spending heavily.
5. Build and test the future product now, rather than extrapolating only from current products and constraints.

Fund People and Visions, Not Narrowly Specified Projects

Kay’s account of ARPA emphasizes giving capable researchers room to find the right problems. On difficult, unfamiliar work, choosing a precise project too early can lock a team into the assumptions of the present.

Use Milestones Rather Than Rigid Deadlines

Treat uncertain research more like baseball than golf: failure is expected in a hard problem, and a meaningful success rate can justify substantial exploration. Licklider’s analogy was that batting .350 is excellent; the many unsuccessful attempts are part of the cost of pursuing difficult work.

Build a Community and Develop Future Talent

The research model depended on successive generations of trained researchers, not isolated projects. Kay notes that PARC researchers were often second- or third-generation products of earlier ARPA-supported work.

Make What the Vision Requires

If existing hardware, software, operating systems, or languages cannot support the goal, develop new ones—provided the team has the expertise. Kay recommends starting with the software experience, then designing hardware to run it effectively.

Engineer Beyond a One-Off Demo

PARC’s reported rule was to make 100 usable units: an Ethernet should serve 100 machines, a time-sharing system 100 users, and a personal computer should be manufacturable in a batch of 100. This forces teams to confront reliability and practical use.

Argue for Clarity, Not Victory

Technical disagreement should help a team understand a complicated problem, not create winners and losers.

Protect Long-Term Work from Short-Term Management Incentives

Kay argues that quarterly cycles and management attachment to existing businesses often obstruct disruptive research. Leaders need to keep learning and allow new efforts to challenge the company’s current sources of success.

Set an Absolute Quality Bar

Don’t make a product look better merely by lowering expectations or comparing it with a weaker baseline. First establish a genuinely high standard, then explore improvements beyond it.

Case Studies and Concrete Examples

Xerox PARC and the Alto

The Alto was built in 1973, about 11 years before the Macintosh, and Kay says it anticipated commercial personal-computing development by roughly 15–16 years. PARC built about 2,000 machines. The system combined bitmap displays, a pointing device, a graphical interface, WYSIWYG, desktop publishing, and authoring tools that let users edit what they read—an approach Kay contrasts with web experiences designed mainly for consumption.

A Portfolio of Enabling Technologies

Kay attributes to PARC a set of roughly eight and a half inventions, including the Alto, laser printing, PostScript and outline fonts, Ethernet, peer-to-peer and client-server computing, and network technologies related to the Internet. He argues that creating a usable ecosystem—not just a single product—helped establish new industries.

Xerox’s Return

Kay estimates that the PARC effort involved about 25 researchers over five years, at an annual cost of roughly $10–12 million in today’s dollars, and puts the resulting economic value at about $35 trillion at the time of his estimate, or around $40 trillion by the lecture. He also says Xerox earned roughly 250 times its total PARC investment, particularly through laser printing. These are Kay’s estimates, not audited financial measures.

He argues that Xerox’s failure to commercialize the entire PARC portfolio did not mean the research had no value; its technologies spread across an industry.

The Broader Research Runway

PARC built on roughly eight years of prior ARPA research. Kay says the full path to the major results took about 12–14 years, illustrating why long-horizon research often cannot be judged on a five-year project cycle.

Sketchpad

Ivan Sutherland’s early interactive graphics system demonstrated drawing with constraints, reusable objects and instances, and dynamic problem-solving. Kay presents it as an example of working on a real problem rather than limiting ambition to what seemed easy or conventional.

SAGE and Interactive Computing

Earlier air-defense research produced interactive displays and large computer systems. Kay highlights how research built for one purpose can create capabilities with much broader future uses, including the foundations of air-traffic-control systems.

Tools Grounded in Observation

Kay describes a RAND tablet effort whose researchers examined people’s discarded work to see how they actually thought—finding diagrams, flowcharts, and other visual material—and then designed tools around those behaviors.

Metrics and Timelines Mentioned

• PARC Alto: Built in 1973; about 11 years before the Mac and 15–16 years ahead of commercial development, according to Kay.
• Alto production: Approximately 2,000 machines.
• PARC research group: About 25 researchers for five years.
• PARC annual cost: Estimated at $10–12 million in today’s dollars.
• Kay’s estimated economic value of the resulting work: $35 trillion, later revised to roughly $40 trillion.
• Kay’s estimate of Xerox’s return on its total PARC investment: About 250×.
• Broader ARPA-to-PARC research arc: Approximately 12–14 years, including about eight years of prior ARPA work.
• PARC engineering benchmark: Build enough to support 100 machines or users, rather than stopping at a demonstration.
• Future-back planning horizon: Imagine roughly 30 years ahead, then work toward a buildable system about 10–15 years ahead.

Actionable Recommendations

• Choose a problem worth sustained effort, rather than one selected only because it fits today’s market or tools.
• Define a compelling long-term vision while leaving room to discover the right intermediate problems.
• Invest in teams with the skill to invent missing tools and infrastructure.
• Test concepts in working systems and at meaningful scale; don’t confuse a prototype with a deployable product.
• Accept a portfolio of unsuccessful experiments when the potential payoff is a major new capability or industry.
• Evaluate leaders and research programs over a horizon long enough for capability-building and talent development to compound.
• To escape present-day assumptions, imagine a desirable future first, then bring the necessary products and technologies back toward the present.

Presenters and Sources

Presenter: Alan Kay Introduction: Sam Altman

Key people, programs, and references cited: J.C.R. Licklider, Ivan Sutherland, Bob Taylor, Larry Roberts, Marvin Minsky, John McCarthy, Douglas Engelbart, Gordon Moore/Moore’s Law, Nicholas Negroponte, Paul MacCready, Picasso, and Wayne Gretzky; ARPA, Xerox PARC, MIT radar and SAGE research, and RAND’s tablet work.