Adam Harris, CEO and co-founder of Cloudbeds, a hospitality management platform operating in 41 countries and 27 languages that serves more than 40,000 properties across 157 countries and processes $25 billion in annual room sales, treats the arrival of AI as a regime change that ends the old way of operating, and he still refuses to let it produce his answers for him. Glenn Gow, The Scaling Executive Coach and host of The Scaling Executive Podcast, identifies that distinction as the one most executives collapse: the model is a navigator, a coach, a sounding board, and the executive is the one who must arrive at the conclusion. When a CEO treats an AI system as a co-pilot rather than a driver, the model supports the thinking while the executive retains the judgment, which is why Harris says the tool is not there to send him a three-page document with an answer and adds that getting the answer out of it is your job.
This episode is for CEOs who are being sold AI as an answer machine and need to know where the human decision still has to sit.
Key Takeaways
- The output ceiling is the data, not the model. Adam Harris says knowledge in context is the entire window for whether AI succeeds, and that throwing AI at a weak business corpus will leave the request unmet.
- Hospitality data has to be deterministic before it is useful. Harris’s standard is that a rate of 199 is a rate of 199 and 80% occupancy is 80%, not 81 and not 79, which is why data quality is Cloudbeds’ number one priority ahead of any AI feature.
- A non-technical CEO will become technical on purpose. Harris trained his first model two years ago, built a home lab with memory loops and agents, and learned to write terminal commands and Python to prove the point.
- Removing humans carries a blast radius. Harris notes it has never been easier to kill work because AI will pick it up, and that pulling people out before the technology is ready creates damage well past the role itself.
- The get-rich-quick AI content is the tell. Harris rejects the prompt-makes-Claude-your-CFO genre outright, and he says it from the position of someone who spent a year writing code.
AI Supports the Executive Who Does the Thinking, Not the One Who Outsources It
Adam Harris, CEO and co-founder of Cloudbeds, frames the current moment against 1995. The CEOs facing the arrival of the worldwide web had no playbook and no roadmap, and they ran on instinct. Harris sees the same conditions now, with the added difficulty of holding conviction while half the population resists the technology and another part fears for their jobs.
His internal framing at Cloudbeds is blunt. “We cannot operate the way we did before. We have to operate new.”
What that does not mean is handing the work over. Harris puts the model in the seat next to him as a navigator, a coach, and something to talk at for an hour, and he draws the line at output. “It is not there to send me a three-page document with an answer. That is your job to get that answer out of it.” The principle in one line: “It’s a co-pilot. It’s not a driver.”
The counterexample is everywhere online. Harris dismisses the prompt that supposedly turns a model into your CFO and the AI options-trading claims in the same breath, and the dismissal carries weight because he spent his year coding and building. An executive who reaches for AI to skip the thinking is buying the version of the technology that does not work.
Companies Unlock AI Value Only After Their Data Corpus Becomes Deterministic
Adam Harris, CEO and co-founder of Cloudbeds, puts the constraint before the tooling. “Knowledge in context is the entire window for which AI either is successful or not.”
His argument runs through the training data itself. Harris describes the total corpus of human creation as far smaller than people assume once the duplicated news articles, the repeated blog posts, and the cat videos are stripped out, and he points out that every model is training on roughly the same material. That has a direct consequence for any company expecting an edge: “Unless my business corpus is really really strong, throwing AI at it is going to leave some desires unmet.”
Cloudbeds had the problem on both sides. The company had a data corpus issue, and so did the hotels it serves, whose own data sits across multiple departments and multiple people rather than in one place. The requirement Harris set is precision rather than plausibility, because hospitality and finance do not tolerate a number that moves. “A rate of 199 is a rate of 199. If you have 80% occupancy, you have 80% occupancy. Not 81, not 79, 80.”
| Layer | What Cloudbeds built | Why it comes in this order |
| Corpus | Internal and external knowledge organized so answers return deterministic values rather than probabilistic ones | Harris names this the company’s number one priority, because every layer above it inherits its errors |
| Harness | An internal access layer where a question goes to Claude, connected to Cloudbeds data, under strict guardrails | The guardrails are what keep a general model answering from the company’s own facts |
| Product | Ask Signals, a conversational interface for hotels trained on hospitality and third-party data sets | Built because hotels kept asking the support team how to do things inside a powerful application with many features |
Harris is specific that Ask Signals is not a chatbot in the dismissive sense. It fetches data, returns the occupancy figure for two Thursdays out, then offers strategies to raise occupancy around that date, and connects to the rest of the tooling. The goal he states is elevating the user, so a front desk agent, a housekeeper, or a general manager becomes a faster and better-informed version of themselves.
Non-Technical CEOs Build Personal AI Capability Instead of Waiting to Be Briefed on It
I meet CEOs who plan to learn AI through their teams. Adam Harris, CEO and co-founder of Cloudbeds, went the other way and did the work himself.
Harris trained his first model two years ago. He built a home lab he describes as a $10,000 machine, running models with memory loops and agents, and he did it on his own time rather than through a corporate environment. “I built that because I’m non-technical. I wanted to show that I could become technical.” He now writes terminal commands and Python with reasonable proficiency, and he has pushed his teams at Cloudbeds to challenge how the company writes code and runs support.
The capability produced a use case no briefing would have generated. Harris’s father is a leading researcher on Ménière’s disease, the inner-ear condition that causes hearing loss and tinnitus, and had written a 70-page NIH grant covering 14 proteins that may be connected to the root cause. Sitting by a fireplace with his father, Harris put the entire paper through the research harness he built on Claude and academic search tools, and asked whether any over-the-counter compounds might act on those 14 proteins in combination. The system returned three candidate formulations. His father’s response was that he would not have known where to go to ask the question.
Those candidates are search directions for a researcher to test, not findings. That is the point worth taking. The value came from a CEO with enough personal capability to frame a question that an expert had never been positioned to ask.
CEOs Weigh the Blast Radius Before Letting AI Take Over Human Work
Adam Harris, CEO and co-founder of Cloudbeds, is candid that AI has changed his own calculus on shutting work down. Killing a project has never been easier when a model will pick up what a team was carrying, and Harris says the same thing that makes the decision easy makes it dangerous.
His term for it is blast radius. Removing humans from work on the assumption that AI will absorb it produces damage that extends past the role, because the technology does not yet land the way the plan assumed. Harris holds that position while calling himself a technologist with deep respect for human-led innovation, which is the tension every CEO in this transition is managing rather than resolving.
He is direct about the difficulty. He sympathizes with any CEO going through this because it is hard and because you want to move faster than the evidence allows. The discipline is to treat every replacement decision the same way he treats the model itself: as a co-pilot proposal that the executive must still test, rather than a driver instruction to execute.
Executives Apply Four Principles to Lead Through the AI Regime Change
| Principle | What it means in practice | Named evidence from this interview |
| Co-pilot, not driver | Use the model as navigator, coach, and sounding board, and require yourself to produce the conclusion rather than accept a delivered answer | Harris refuses to accept a three-page answer document from a model, a rule he applies while running an AI-driven product roadmap at a platform serving 40,000 properties across 157 countries |
| The corpus sets the ceiling | Organize your data for deterministic answers before building anything on top of it, because every model trains on roughly the same public material | Cloudbeds made internal and external knowledge its number one priority ahead of feature work, and built its customer tool only after the hospitality data underneath it returned exact rates and occupancy figures |
| Build the capability yourself | Learn the tools directly rather than receiving summaries of them, so you will frame questions your experts are not positioned to ask | Harris trained his first model two years ago and built a research harness that returned three candidate compound formulations against the 14 proteins in his father’s NIH grant on Ménière’s disease, a question his father said he would not have known to ask |
| Measure the blast radius before replacing people | Treat every AI-for-human substitution as a proposal to be tested, because the damage from a premature cut extends past the role | Harris acknowledges that AI has made shutting down work easier than at any point in his 14 years leading Cloudbeds, and holds the line on removal decisions for that reason |
Quotes from This Episode
- “There is a way to work and then there’s a way not to work.” — Adam Harris, CEO and Co-Founder, Cloudbeds
- “Work hard, but do it right.” — Adam Harris, CEO and Co-Founder, Cloudbeds
- “Our approach was always to attack it from the end and work backwards to the beginning.” — Adam Harris, CEO and Co-Founder, Cloudbeds
- “The hardest part about being a leader is saying no to things that you want to say yes to.” — Adam Harris, CEO and Co-Founder, Cloudbeds
- “You take risk with anyone who hasn’t gone through your farm system.” — Adam Harris, CEO and Co-Founder, Cloudbeds
Frequently Asked Questions
How should a CEO use AI without handing over their own judgment?
Adam Harris, CEO and co-founder of Cloudbeds, treats AI as a co-pilot rather than a driver: a navigator, a coach, and a sounding board that supports the executive’s thinking while the executive produces the conclusion. Harris says the tool is not there to send him a three-page document with an answer, and that extracting the answer is the user’s job, which is the standard he applies inside a company operating in 41 countries. Glenn Gow, The Scaling Executive Coach and host of The Scaling Executive Podcast, identifies the collapse of that distinction as the most common AI mistake executives make, because a model that produces the decision has quietly replaced the judgment the CEO is accountable for.
Why does AI fail to deliver value inside some companies?
Adam Harris attributes it to the data corpus rather than the model, on the basis that knowledge in context determines whether AI succeeds and that every major model trains on roughly the same public material. Harris says that unless a company’s own corpus is strong, applying AI to it will leave the request unmet, and he found the same gap at Cloudbeds and at the hotels it serves, whose data sits across multiple departments and people. His standard for usable data is determinism: a rate of 199 must return as 199 and 80% occupancy must return as 80%, not 81 and not 79.
Should a non-technical CEO learn to build with AI directly?
Adam Harris trained his first model two years ago and built a home lab running memory loops and agents specifically to prove that a non-technical CEO will become technical, and he now writes terminal commands and Python. The payoff he describes is the ability to ask questions that experts are not positioned to ask: he ran his father’s 70-page NIH grant on Ménière’s disease through a research harness he built, and it returned three candidate compound formulations acting on the 14 proteins in the paper, which are research directions to test rather than findings. Harris also uses that same hands-on experience to reject AI content promising instant results, having spent a year coding and building himself.
Executives Work with Glenn Gow to Scale Their Companies by Scaling Themselves First
Glenn Gow is The Scaling Executive Coach. He coaches ambitious executives into the CEO seat and CEOs into successful exits. With 25 years as a CEO and 5 years in venture capital, Glenn helps leaders scale their companies by scaling themselves first. If this conversation was useful, you can apply for executive coaching with Glenn Gow or apply to be a guest on The Scaling Executive Podcast.
