Riding the Technology Waves: Doug Williams on What the Internet, Mobile, and GenAI Have in Common, and What Makes This Wave Different
Q1. The pattern that repeats across waves
Looking across the internet, mobile and GenAI, what is the single most important pattern that separates those who successfully ride a new wave from those who get left behind, and how clearly do you see it playing out right now?
I have been very lucky in my career, able to ride successive technology waves. For every wave, focused on how we could apply the technology to better solve problems for real people and use cases.
At FTP Software, Empirix, and Family Education, got to ride the internet wave moving from the underlying TCP/IP layer to the web application layer to create the leading K12 education site. At Zipcar got to ride the mobile wave, as we allowed our drivers to access our cars using their phone, and then at Runkeeper we grew to over 50 million users tracking their runs from their phone. At 1upHealth, the FHIR healthcare data standards and new federal interoperability rules finally opened up health data that had been stuck in silos for decades.
GenAI is the same pattern, just faster. What took a few years with the internet and mobile now happens in months, and the models change every quarter. At the Trust Center we had a head start. Bill Aulet spent years building the Disciplined Entrepreneurship frameworks to codify best practices from his work at MIT for increasing the odds of startup success.
When ChatGPT landed, students were already asking us for guidance, so we automated MIT’s core Disciplined Entrepreneurship frameworks. That became the Orbit AI Jetpacks. Because the frameworks that Bill and our MIT entrepreneurship leaders had developed broke down our best practices into distinct steps with considerations for each step, they turned out to be a great fit for original GenAI technology, and are even better today. Our focus was to ensure that our user interfaces were easy to learn and use, and allowed for quick updates, as our users validated the Orbit AI Jetpack data with real people.
Q2. What “AI That Works” requires
As an independent MIT judge for the InterSystems Impact Awards 2026, what distinguished submissions where AI was delivering impact from those where it was present in name only, and what does “AI That Works” require that most organizations still underestimate?
I judged the 2026 awards with Andrea Kates and Phil Budden, all three of us working with MIT entrepreneurs. The entries that stood out had their data in order first. You can’t get reliable AI out of messy, fragmented data. That was a big lesson for me working with healthcare data as we normalized it for the FHIR standards, and it held across the healthcare, public sector and supply chain entries.
The strong submissions also put AI into a workflow people already used. A clinician gets a summary of the record inside the system they already open every day. A logistics planner sees the risk right where they make the decision. The weaker entries described a model, a pilot or a chatbot and stopped there. Lots of capability, but no owner and no measured result.
What most organizations underestimate:
- The data work. Cleaning, mapping and governing data is most of the effort. It isn’t glamorous.
- Measuring the result. Wait times, hours saved, error rates. If you can’t say what changed, it’s not working yet.
- Who owns it after launch. Someone has to watch it and fix it when the model or the data changes.
The winners ran AI like any other production system. Someone owned it, they measured it, and they had a plan for when it broke.
Q3. Democratizing entrepreneurship, or widening the gap?
Is GenAI democratizing entrepreneurship, or primarily accelerating the advantage of people who were already well positioned, and what would it take to ensure the former?
Both. Anyone with a laptop can now get a decent first pass at a market segmentation or a pitch in minutes. That’s real, and it wasn’t there three years ago. But a model doesn’t give you judgment, a network or capital. People who already have those just go faster, and the gap can grow.
That’s the gap we’re working on with Orbit. Everything we do starts from Bill Aulet’s work at the Trust Center. Disciplined Entrepreneurship has been taught at MIT for years and is used by hundreds of schools, and our view is simple: MIT’s approach helps increase your probability of success, so let’s get it to more people. Paul Cheek started Orbit, and alumni funding let us build the Jetpacks on top of it. They take Disciplined Entrepreneurship, Paul’s Startup Tactics, and Climate and Energy Ventures, and walk founders through them step by step. Over 10,000 people in the MIT community use Orbit, as well as our partner universities and accelerators. The Jetpacks have around 12,000 users who have generated over 20,000 ideas. Given the efficiencies with the most recent AI technologies, a handful of us can develop and support the Orbit AI Jetpacks across both our MIT and non-MIT users.
We also wanted to know if it works. Billy Dale, an MIT System Design and Management student advised by Bill, studied multiple semesters of Jetpack use in Bill’s 15.390 course. Our best teams completed more of the Disciplined Entrepreneurship steps and went back to revise their work more often, especially market segmentation, and course ratings went up. It also showed where AI helps less. It was strongest on the analytical steps and weaker on creative, messy synthesis, which is where mentors and customers come in. Bill wrote up the findings. He saw the same thing in delta v, where the Jetpacks lifted the baseline quality of the whole cohort.
A few things matter if you want this to help more people. Focus on the user interface, so users enjoy using it and iterating on the data. We run on fast, low-cost models, so an extra user costs very little and we can make our platform widely available. Build workflows and structure around the AI. The Jetpacks help steer users adopting our MIT entrepreneurship frameworks in a step-by-step approach. We encourage use of the Jetpacks to accomplish secondary market research faster so that the entrepreneurs can spend more, higher quality time with real people to validate their findings and build more unique insights based on real use cases and problems to be solved.
It’s also important when building any technology, certainly true with AI, to connect it to a community. I mentor at a lot of accelerators. When a program has no common framework, I spend the session on fundamentals. With founders who know Disciplined Entrepreneurship, we skip past what a beachhead market is and talk about how to reach customers. That’s the kind of head start we want everyone to have.
Q4. How the entrepreneurial mindset needs to evolve
What has surprised you most about how the entrepreneurial mindset needs to evolve for the GenAI era, when the underlying capability is available to almost everyone at the same time?
I’m always aware and surprised at how fast a technology advantage can go away. Everyone can call the same models. I’ve seen student teams build in a weekend what used to take a seed round. That’s great, and the next team can do the same thing the following weekend.
So the fundamentals matter more. Who exactly is your customer? What do you know about them that nobody else does? In Disciplined Entrepreneurship the 24th step is the product plan, and that’s where every entrepreneur wants to start, including me. The first steps are all about the customer, that’s where the entrepreneurs’ advantages are as they gain unique real world insights.
I think of entrepreneurs driving their startups at high speed to bring their innovations to market. Our Orbit AI Jetpacks couple with MIT’s Disciplined Entrepreneurship frameworks, help them know which highways they’re on with which customer. Entrepreneurs are pulled in different directions as they talk with more people.
They can go really fast ahead, but if they keep weaving on and off the road, they won’t get there any sooner. The Jetpacks are your steering wheel. Create several ideas, one for each customer segment you’re considering, keep revising as you learn, and be clear which one you’re focused on right now. Use AI, but don’t let it drive. Start with the framework to stay focused, validate with customers, then take that work into Claude, Gemini or whatever you like for surveys, research, a landing page or outreach.
On the build side, stay mainstream and don’t lock yourself to one model provider. Keep the AI layer separate so you can use the right model for each task, and be a fast follower on new architectures instead of betting the company on them. The models keep getting better and cheaper. You want to be able to use the best models for the particular task, the week they show up.
Q5. What has changed, and what has stayed constant
Looking across nearly two decades of ODBMS.org innovation interviews and three technology waves, what has changed about how innovation happens, and what has remained stubbornly constant?
What changed is the cost and speed of building. At FTP Software, shipping meant disks, boxes and long release cycles. Mobile brought app stores and weekly releases. Now a small team can build and test a working product in a few days, even hours, so a lot more people get to try an idea.
What hasn’t changed is everything around the actual people involved. You still have to find customers with a real problem. You still need a team that trusts each other when things go sideways. Go-to-market is still hard, as you work to cut through the noise. And how you handle people’s data matters more each wave. In Orbit, founders’ ideas stay private, and we’re set up so the model providers don’t train on them.
Going back through your series, from Vint Cerf and Alan Kay on, the people building things keep saying it in different ways: stay curious, talk to customers, keep building. The tools change. That part doesn’t.
Q6. What I wish I had known at the start
Looking back at the hardest decisions and most instructive moments, what is the one thing you know now about building technology products and organizations that you wish someone had told you at the beginning of your career?
Build for supportability from day one.
Early on I treated the first version as something to get out the door, and we’d clean it up later. Later came, but often after we had accumulate technical debt. Every early version became something the next team had to live with, and that team was usually bigger, under more pressure and serving more customers. At Zipcar we were growing fast and integrating acquisitions in the US and overseas. The systems that had documentation, monitoring and clean interfaces were the ones we could scale. The others slowed us down, sometimes at the worst possible times.
I tell founders now: if you can’t support it, monitor it and hand it to someone else, it’s a hobby project, not a production application. That doesn’t mean over-engineering. It means picking well-supported technology where you can, and assuming someone new will join in a year and need to understand what you built.
The other lesson is about people. Hire people who are better than you at the things you’re weakest at, and give them real ownership. Most of my hardest decisions were about people, and the ones I waited too long on cost the most.
Qx. Anything else to share?
I’ve learned the value of mission-driven startups, focused on problems that matter. You’re going to spend over half your life working. It’s a gift to be able to spend each day working with great people applying the latest technology to solve problems that really matter.
And thank you, Roberto, for keeping these conversations going for twenty years.
………………………………………………………
Doug Williams, Technology Lead, Orbit AI Jetpacks, Martin Trust Center for MIT Entrepreneurship
Doug Williams is Technology Lead for the Orbit AI Jetpacks at the Martin Trust Center for MIT Entrepreneurship, where he leads Orbit, MIT’s GenAI-enabled entrepreneurship platform for entrepreneurs worldwide. He is founder of CentanniPark, a strategic consultancy for mobile, web and digital health. Previously he was Chief Product Officer at 1upHealth (FHIR interoperability), CTO at iSpecimen, VP Engineering at Runkeeper (50M+ users, acquired by ASICS) and Zipcar (IPO, acquired by Avis Budget Group), and led engineering at Empirix, Family Education Network and FTP Software, startups acquired for over $1B in aggregate. He served as an independent MIT judge for the 2026 InterSystems Impact Awards. BSEE, Princeton University; MBA, Harvard Business School.
Links: MIT Trust Center profile | LinkedIn | orbit.mit.edu