What Stephen Koza's Squiggly Path From Engineer to CEO Reveals About Making Decisions Without a Crystal Ball
Every senior technology leader is being asked to commit budget, headcount, and architecture to a market that reprices itself every quarter. The five-year plan still gets built, the board still expects it, and the honest version of that document contains more assumptions than facts. The leaders navigating this well have stopped waiting for certainty and started building the things that hold up regardless of which way the market breaks.
Stephen Koza has spent a career operating that way, and this time he is the one answering the questions. TechPod Talks producer Lauren Hochman Dort turned the mic on the show's usual host for our latest episode, walking us through the path from electrical engineering student at UT Austin to enterprise sales, through Silicon Valley startups and investing, to acquiring and running EverOps as CEO since February 2022. He has described that route as a “squiggly path,” and the phrase undersells it.
The conversation moved from career pivots to the decisions engineering leaders are making right now, covering what authentic leadership looks like as an observable behavior, how to hire for ceiling when credentials say otherwise, and why most companies are still failing to get returns from AI. His views on that last point come from running AI within his own operations, which gives them a different weight than a slide would.
Read on for what Stephen has learned about planning under uncertainty, the gap between a working demo and a production system, and the parts of engineering leadership that hold their value as autonomy increases.
Planning for Certainty Is the Wrong Bet
The pressure to map out a career or a technology roadmap in advance shows up early and rarely serves anyone well. Stephen described advising an undergraduate intern who had his entire trajectory planned out, including graduating in three years, and telling him to pump the brakes. The advice came from Stephen's own experience, since he took an engineering job out of UT Austin while knowing he wanted the business side, lobbied his way into sales about a year in, and built everything after that on pivots he could not have predicted at the start.
That same skepticism shows up in how he handles the five-year plan as a CEO. When people asked him for one at the beginning of his EverOps journey, he turned the question around and asked for an example of a five-year plan that landed exactly where it said it would. EverOps builds the plan and does the strategy work, and Stephen holds it with an appropriate grip on how much of it will survive contact with the market.
"Anybody that can confidently tell you what the world is going to look like in five years is full of it, in my humble opinion."
For a VP of Engineering or CTO, this lands directly on the architecture bets and vendor commitments sitting on this year's roadmap. Stephen's framing of AI's scale helps explain why uncertainty is unusually high right now. As he put it, the shift is "personal computer plus cloud plus mobile plus internet, all wrapped into one, to the power of ten or a hundred." A market moving at that speed rewards decisions that stay reversible and foundations that stay useful across several possible futures.
Vibe Coding Is Easy & Production Is Hard
One of the most useful things Stephen has learned from running AI across sales, marketing, finance, and delivery within his own company is how misleading early wins can be. Getting AI to do something impressive takes very little effort now, which is exactly what makes the next step so easy to underestimate.
"It's really easy to vibe code something, but it is really hard to get that vibe-coded thing into a production app that's stable and secure and doesn't break and doesn't require full-time people to babysit and fix all the time."
That distance between a demo and a durable system is where most enterprise AI programs are currently stuck. Companies today are investing in AI and adopting agents left and right, while the majority of those efforts fall short of the returns they expected. Stephen sees this pattern play out across the industry and in the client work EverOps does. The causes cluster into a few categories, such as culture and change management, picking the wrong use cases, and, most often, a messy foundation beneath the whole effort.
The term foundation means something specific here. Starting with data organized and consolidated enough to be usable, plus a consolidated infrastructure layer of security, governance, guardrails, observability, and clean pipelines, carries a clear business case long before agents entered the picture, and still delivers cost savings, faster shipping, and less downtime. The agentic era adds a new reason to care about it, which our recent analysis of the infrastructure gap blocking AI ROI examines in depth. Stephen's version of the point is blunt. Your house has to be in order before you can expect a return, and the alternative is explaining to a CFO why token spend climbed while earnings per share stayed flat.
Probabilistic Systems Need a Human Layer
Stephen has said publicly that companies may eventually run production systems autonomously, and he is specific about how far away that is. Companies are not doing it today, and that will hold for a while because large language models are probabilistic, whereas production systems demand determinism.
The math makes the constraint obvious. Ask a model the same question twice, and the answers can differ, which is tolerable in a chat window and disqualifying in a system serving customers. Downtime budgets do not have room for a system that is wrong some percentage of the time, and the failure mode is what makes it dangerous.
"When it's wrong, AI runs off a cliff at 200 miles an hour very confidently."
Confidence without correctness is the specific risk, and it is why Stephen places the human layer where he does. Teams need people who understand how to build, architect, and operate these systems, who can set the right guardrails and catch failures before they propagate. He pairs that with genuine optimism, noting that he is not dumb enough to bet against technology and that models will continue to improve. The load-bearing question for a technology leader is which decisions keep a person accountable for them, and that answer changes more slowly than the models do.
His read on the job market follows the same evidence-first pattern. Stephen was uncertain about the displacement narrative six months ago and has landed on job change and displacement over wholesale loss, pointing to software developer openings at a three-year high and to frontier labs hiring aggressively while being demonstrably good at AI. He also flags another discipline worth adopting: drawing information from many sources and questioning both the claim and the source, because sentiment around AI runs hotter than the data supports in either direction.
Betting on Ceiling When the Credentials Say Otherwise
Hiring consumes an enormous share of a senior leader's time and results in a painful number of misses. Stephen's most reliable pattern is one he picked up early and now applies deliberately, which he calls the “rising star profile.”
These are candidates who do not quite have the experience or the credentials the role nominally calls for, and who show clear evidence of having taken on something above their heads and risen to it. The reason the profile works is that it predicts those people have a high ceiling and a genuine growth mindset. Stephen named the tell that separates them.
"They're not just open to feedback. They want it. They ask for it."
His second signal probes how a candidate thinks by walking through a problem they solved, particularly one involving ambiguity, no straight-line answer, or real personal or professional adversity.
The follow-through matters as much as the hire. Stephen frames his role as providing people with a platform and good advice, which starts with understanding what someone actually wants, since helping with a goal requires knowing what it is. He extends that to the part many leaders quietly resent, having built an environment where someone chasing something the company cannot offer can say so and get help finding it, whether through his network, advice, or a recommendation. He is also self-critical about the frequency, acknowledging that the press of running a business makes this the kind of work that slips when heads go down.
Hard to Work With and Toxic Are Different Problems
EverOps has had a no-jerks policy since before Stephen arrived, and he treats it as an operating hiring principle and a core cultural value. His reasoning is straightforwardly operational, since a person who is excellent at the job but toxic to the organization has a negative impact that far outweighs the positive.
The next distinction he draws is where the leadership value lies, because conflating these two categories is costly in both directions. A genuinely toxic situation calls for a decision. A person who is hard to work with is usually a repairable relationship, and Stephen has lived the repair, describing a colleague he butted heads with until the two of them sat in a room and talked it through, after which the working relationship improved substantially. Some things needed to be said, and neither person had understood where the other was coming from.
"Hard to work with and toxic are not the same thing."
That repair depends on the same foundation as everything else in his leadership approach. Stephen points to Radical Candor and its instruction to care deeply, along with Ben Horowitz's argument that culture is the behavior a leader models, well beyond the perks in the office. His own summary of authenticity leaves no room for technique, by making it clear that there is no magic advice for caring and a leader simply has to care. Teams read the difference immediately, and the leaders people remember are the ones who show up supportive and still hold the line on accountability.
How to Apply These Frameworks This Quarter
The conversation maps onto decisions engineering and technology leaders are making right now. A few concrete places to start:
- Build the plan and hold it loosely: Keep the five-year plan for the board, and treat the commitments within it as revisable, favoring two-way door decisions and investments that aren’t hard-coded. A market moving this fast rewards optionality over precision.
- Audit the foundation before funding the next agent: Check whether your data is sufficiently organized and consolidated to build on, and whether security, governance, guardrails, observability, and clean pipelines are in place. Most stalled AI programs trace back to this layer well before they trace back to model choice.
- Name the decisions that keep a human accountable: Map where a probabilistic system touches production and decide explicitly which actions require a person to own the outcome. The failure mode is a confident wrong answer moving at machine speed, and guardrails must be designed in advance.
- Hire for ceiling on at least one open role: Look for the candidate whose evidence shows they took on something above their head and delivered, and probe how they navigated a problem with no straight-line answer. Ask whether they seek feedback or merely accept it.
- Separate your hard-to-work-with people from your toxic ones: Make the distinction explicitly for anyone currently creating friction, then have the direct conversation with the first group. Most of those situations resolve into a stronger working relationship once both sides say what needs to be said.
For readers interested in tracing the full arc of Stephen’s journey, he has shared his story with us before here in How EverOps Quietly Built 'Two Nikes': A Candid Conversation with CEO Stephen Koza, where he digs into outcomes-driven delivery, the company's AI-native philosophy, and what's kept EverOps' client base growing for years. Additionally, we conducted an extensive executive interview with him earlier in 2025, titled "From Texas to Silicon Valley: An Executive Q&A with EverOps’ Stephen Koza," in which he walks through the early career decisions and people-first leadership approach that shaped the company from the start. Both are worth a read for anyone looking for more of his perspective on his journey to date and on what sets EverOps apart from other firms.
Partner with EverOps to Put the Guardrails Around Your AI Ambitions TodayÂ
Our insightful conversation with Stephen kept circling back to one practical idea. Nobody knows how this market resolves, so the smartest response is to build what holds up regardless of which way it breaks.Â
For a VP of Engineering or CTO weighing this year's architecture bets and vendor commitments, that translates into a short list of concrete work such as ensuring the data and infrastructure are sound enough to support whatever comes next, building guardrails before autonomy expands rather than after something breaks, and finally backing that up with people who have the judgment to catch a confident mistake before it reaches customers.
If you’re in search of some foundational next steps, try kicking things off with our AI Opportunity Assessment, which identifies and prioritizes AI use cases genuinely worth a team's time, evaluating data readiness and technical infrastructure before a single line of code is written. For organizations ready to move past assessment, our AI Quick Start engagement delivers rapid proof of value through proven playbooks, and our AI Adoption TechPod embeds a dedicated AI engineering team for continuous delivery and scaled adoption.
Whatever path you’re on currently, if your team is working through any of the questions or problems Stephen touched on in this episode, we encourage you to reach out to our team today to start the conversation. Real AI ROI may not be as far off as it feels.
Keep Up With TechPod Talks
Stephen Koza joined TechPod Talks producer Lauren Hochman Dort for Episode 11 of TechPod Talks. Subscribe today to listen to the full conversation on Apple Podcasts, Spotify, YouTube, or the EverOps Podcast Page now.Â
Follow Stephen on LinkedIn and the EverOps page for more from the series.
Frequently Asked Questions
What is TechPod Talks?
TechPod Talks is a podcast hosted by EverOps CEO Stephen Koza featuring candid conversations with technology leaders, engineers, and operators. Each episode explores how real teams build, scale, and operate modern systems, with a focus on practical takeaways.
What topics does Episode 11 cover?
Episode 11 turns the mic on the host, with producer Lauren Hochman Dort interviewing Stephen Koza on the squiggly career path and planning under uncertainty, maintaining credibility with technical teams, what caring looks like as a leadership behavior, the difference between hard to work with and toxic, the rising star hiring profile, and why most AI initiatives are falling short of production and returns.
Who is Stephen Koza?
Stephen Koza is the CEO of EverOps, a position he has held since February 2022. He studied electrical engineering at the University of Texas at Austin, began his career as an engineer before moving into enterprise sales, and spent years in Silicon Valley at companies including Cisco Meraki and Samsara. He later founded an investment firm, which led to the investment in EverOps and his move into the CEO seat. His earlier executive Q&A covers more of that background.
What is the rising star hiring profile?
The rising star profile describes a candidate whose ceiling exceeds their current credentials. They may lack some of the experience a role nominally requires, and their history shows a clear instance of taking on a challenge above their level and succeeding. Stephen looks for a growth mindset and a candidate who actively seeks feedback, which he treats as a strong predictor of how quickly that person will grow into a larger role.
Why are most companies failing to get ROI from AI?
Stephen points to a few recurring causes, including culture and change-management gaps, poorly chosen use cases, and, most commonly, a messy foundation. Data that is disorganized or spread across too many places, combined with missing security, governance, guardrails, observability, and pipeline work, prevents pilots from reaching stable production, which is where the expected return actually lives.
Where can I listen to TechPod Talks?
TechPod Talks is available on Apple Podcasts, Spotify, YouTube, and the EverOps website, with episodes released in both audio and video formats.
How does this episode connect to EverOps' work?
EverOps helps technology leaders navigate the same questions Stephen covers in this episode, including AI adoption, strategy, cloud and infrastructure readiness, and the operational foundation that turns AI investment into production results. Services like AI Opportunity Assessment, strategy consulting, and embedded operations map directly to the work he describes.
