I Build AI-Native Companies, But I Still Need to Be the CEO.


AI has transformed how I work. But the more I use it to build businesses, the more I realize that doing more work and making better decisions are not the same thing.
I run a software company, and I'm also building AI-native startups. As a software engineer by training, I've spent much of my career exploring how technology can help businesses operate more effectively. Artificial intelligence has accelerated that process in ways I wouldn't have imagined just a few years ago.
Today, I use AI extensively. It helps me develop software, analyze marketing performance, conduct market research, write emails, and explore business opportunities. I also use it as a thinking partner, challenging my assumptions and helping me consider perspectives I might otherwise overlook.
Much of the thinking still begins with me. An idea might come from a conversation with a customer, an experience running my company, or something I've observed in the market. AI helps me develop and articulate that thinking, often much faster than I could on my own.
The productivity gains have been extraordinary. As a developer, I can accomplish in hours what previously required days. As a CEO, I can examine more information and explore more alternatives without proportionally increasing my workload.
Yet despite building businesses around AI, I wouldn't trust it to run my companies entirely on its own.
Not because I doubt where the technology is heading. I expect AI systems to become substantially more capable and autonomous. My hesitation comes from something I've repeatedly experienced: being able to execute a task isn't the same as understanding whether that task should be executed, how far it should go, or whether it matters in the first place.
When Building More Is the Wrong Decision
Recently, I was developing an early version of a clinical software product. We were testing a specific hypothesis and needed a functional prototype to determine whether our approach made sense to potential users.
We weren't preparing to deploy the product in a hospital or support real clinical operations. We intended to use synthetic data and potentially test the functionality with a small group of trusted people. Our objective was to build enough to demonstrate the concept, observe how people interacted with it, and learn whether the idea deserved further investment.
The AI assisting with development, however, kept recommending additional infrastructure: sophisticated access controls, backup mechanisms, privacy safeguards, HIPAA-related requirements, and other capabilities associated with a more mature clinical application.
Many of these recommendations were technically reasonable. Healthcare software requires serious attention to privacy, security, and regulatory obligations. Once real patient information or clinical workflows are involved, those considerations become essential.
But that wasn't the context of our experiment.
I repeatedly asked the AI to simplify its approach. We didn't need a production-grade architecture. We needed enough functionality to test an assumption without introducing unnecessary risk.
Every additional component meant more code, more dependencies, and more time before obtaining the evidence we needed. A sophisticated system might eventually become necessary, but building it prematurely would consume resources without necessarily improving our understanding of the problem.
The AI was helping me build a technically more complete product. I was trying to discover whether we should build the product at all.
That experience captured a distinction I've encountered throughout my career: building something correctly and building the right thing are fundamentally different challenges.
As engineers, we naturally think about robustness, scalability, security, and maintainability. Those considerations matter, but their relevance depends on the product's stage, its operating environment, and the consequences of failure.
The appropriate level of sophistication isn't determined solely by what technology can accomplish. It's determined by what the business needs to learn, which risks are present, and what commitments the available evidence justifies.
In this case, the AI could generate the code and explain the benefits of additional safeguards. What it wasn't consistently doing was recognizing that the most valuable decision was to stop adding things.
Judgment Is More Than Generating Answers
I've spent years developing software, managing projects, and building businesses. Some decisions produced excellent results. Others created unnecessary complexity, wasted resources, or forced us to reconsider assumptions we should have challenged earlier.
Those experiences influence how I evaluate opportunities today.
When reviewing a proposed architecture, I don't think exclusively about engineering quality. I consider whether we're solving an important problem, how much we know about the requirements, what might change as we learn, and how difficult it would be to reverse our decisions.
Sometimes that reasoning is explicit. Other times, I recognize a familiar pattern almost immediately. A proposal might look technically impressive, yet its scope feels disproportionate to the problem.
I don't consider this intuition infallible. Experience can create biases, and founders can become overly attached to their interpretations. That's one reason I value AI: it can challenge my thinking rather than simply reinforce it.
But experience also provides context that isn't always captured in specifications or prompts. I've seen what happens when teams build too much before validating demand. I've experienced the consequences of premature architectural decisions and watched projects become complicated because nobody questioned whether every requirement was necessary.
AI can reason about these situations and draw on enormous amounts of information. What it doesn't necessarily possess is the same understanding of the specific commitments, relationships, and consequences surrounding a decision.
Of course, better-configured agents, with clearer objectives, constraints, and risk tolerances, might behave differently. My experience doesn't prove that AI will never develop stronger judgment.
Nevertheless, someone must determine which objectives matter, which risks are acceptable, and when sufficient evidence exists to move forward. Those decisions have consequences beyond the technical system.
For me, that's where leadership responsibility becomes particularly important.
A Company Is More Than a Collection of Tasks
Much of the discussion about autonomous companies assumes that if enough activities can be performed by AI agents, the organization itself can eventually operate without meaningful human involvement.
I understand the reasoning. If agents can develop software, manage customer support, analyze financial performance, and coordinate operations, it's reasonable to ask how much human management will remain necessary.
I expect increasingly autonomous organizations to emerge. But running a company involves more than completing tasks.
A business is also a network of decisions, responsibilities, commitments, and relationships. Its success depends not only on whether work gets completed, but on whether people make appropriate trade-offs, respond to changing circumstances, and maintain the confidence of those who depend on them.
Consider my relationship with customers.
AI can research a customer's organization before a meeting, identify potential challenges, and suggest questions worth exploring. It can summarize our conversation afterward and help me evaluate the next steps.
These capabilities make me better prepared. But I still want to sit across from that customer.
I want to hear how they describe their challenges, understand what concerns them, and observe how they respond when we discuss difficult choices. I also want them to understand how I think, what I value, and whether they can trust me to make responsible decisions.
Understanding a person and building a relationship with that person aren't necessarily the same thing.
Trust develops through experience, consistency, and the willingness to stand behind commitments. When someone chooses to work with me, they're not only evaluating technical capability. They're also making a judgment about the people who will be responsible when circumstances become difficult.
The same applies to my team. AI can help organize priorities, monitor progress, and identify operational problems, but I don't want every interaction with employees reduced to an automated exchange. Some conversations matter because we're having them as people, with different experiences, concerns, and ambitions.
I also think about former customers and colleagues I've met throughout my career. AI could remind me when to reconnect and even draft a thoughtful message. But when I genuinely want to know how someone is doing, I'd rather make the call myself.
That personal attention is part of what gives the relationship meaning.
Can AI Agents Build Trust?
I don't assume these distinctions will remain unchanged.
AI agents may increasingly negotiate agreements, coordinate business activities, and make decisions on behalf of organizations. In transactional contexts, this could be far more efficient than requiring humans to participate in every exchange.
But I wonder what happens when the relationship itself matters.
If two agents negotiate an agreement, the transaction might be perfectly executed. Yet when something unexpected happens, who is ultimately responsible? Who decides whether preserving a relationship is more important than enforcing a contractual provision? Who recognizes that a customer needs reassurance rather than another technically correct response?
Perhaps future systems will become excellent at making those judgments. Perhaps people will become comfortable developing long-term relationships with AI representatives. I wouldn't dismiss either possibility.
Still, there's a psychological dimension to human interaction that current AI systems don't fully reproduce.
When we meet someone personally, we're forming impressions about their character, intentions, and reliability. Those impressions aren't always accurate, but they influence how we establish trust and decide whom to work with.
I suspect human relationships will remain particularly important when decisions involve uncertainty, significant commitments, or shared responsibility.
The question isn't whether AI can participate in business relationships. It already can. The question is which relationships benefit from automation and which derive part of their value from personal accountability.
What Should a CEO Do With the Time AI Gives Back?
All of this has made me reconsider what I actually want from AI.
The obvious answer is productivity. If I can develop software faster, prepare better analyses, and automate repetitive activities, my capacity increases.
But capacity alone isn't the outcome I'm looking for.
I've seen how easily organizations confuse activity with progress. A company can have everyone working at full capacity while important decisions remain unresolved. Teams can deliver more features without improving a product's value. Executives can spend their days reviewing reports and attending meetings without addressing the problems limiting their businesses.
AI could make that situation worse if we use it simply to multiply output. We might generate more reports than anyone can review, pursue more opportunities than we can evaluate, and launch more initiatives without improving our ability to prioritize them.
The technology would make us more productive while the organization became increasingly distracted.
That's not the future I want to build.
If AI could eventually execute 95% of the operational activities requiring my attention, I wouldn't necessarily use that capacity to run twenty more companies. I'd want to spend more time understanding customers, strengthening relationships, and exploring opportunities I currently can't evaluate with sufficient depth.
I'd want more time to think carefully about consequential decisions rather than constantly reacting to immediate demands. I'd want to spend more time with my team, understanding what they need to succeed. And I'd want to reconnect with people whose relationships I value but haven't maintained as consistently as I should.
Those are the activities that attracted me to entrepreneurship in the first place. I enjoy building technology, but what makes the work meaningful is understanding problems, working with people, making decisions, and creating something valuable.
The possibility that AI could free me to focus on those activities is more interesting than simply completing more tasks.
Of course, time saved through automation doesn't automatically become time for better leadership. Without deliberate choices, it can easily be absorbed by additional projects and responsibilities.
The real benefit requires deciding what deserves the attention technology makes available.
The Company I Want to Build
I don't believe humans need to remain involved in every business decision. Many activities I perform today should eventually be handled by intelligent systems operating within appropriate boundaries.
Nor do I believe human judgment is always superior. AI can identify patterns we overlook, challenge biases, and sometimes make more consistent decisions than people.
The challenge is determining how intelligence, responsibility, and decision-making should be distributed throughout an organization.
Which decisions can be delegated safely? Which require human oversight? How do we recognize when automation creates more complexity than value? And how do we ensure that greater autonomy doesn't weaken the relationships and accountability on which the business depends?
These are organizational questions as much as technological ones.
As execution becomes easier and cheaper, deciding what deserves to be executed may become even more important. AI makes it possible to pursue ideas and build functionality at a fraction of the previous effort. That's an extraordinary advantage, but it also makes it easier to build things we don't need.
The scarce resource may increasingly be neither code nor information, but the judgment required to determine where to focus and when to stop.
I'm still learning how to build organizations that take full advantage of AI without allowing increased capability to create unnecessary complexity. I don't yet have a model for a completely autonomous company, and I'm not convinced achieving one should be the ultimate objective.
What interests me more is building companies that operate with greater intelligence and independence while preserving the judgment, accountability, and human relationships that make them worth building.
I don't want AI simply to make me a faster CEO. I want it to help me become a more effective one.
The goal isn't to remove myself from the company. It's to remove the work that keeps me from leading it.
Questions this article raises.
Can AI run an entire company without a CEO?+
AI is becoming increasingly capable of managing operations, developing software, analyzing information, and making decisions within defined boundaries. However, running a company also involves determining priorities, managing uncertainty, building trust, and taking responsibility for outcomes. While greater autonomy is likely, the question isn't simply whether AI can replace a CEO, but which responsibilities should be delegated and which still benefit from human judgment.
Why can AI make a business more productive without making it more effective?+
AI can dramatically increase the speed and volume of work, but greater output doesn't necessarily create greater business value. Companies can develop more features, generate more reports, and pursue more initiatives without addressing their most important problems. Effectiveness depends on choosing the right priorities, reducing unnecessary complexity, and ensuring that execution contributes to meaningful outcomes.
What should CEOs focus on as AI takes over more operational work?+
As AI assumes more execution responsibilities, CEOs have an opportunity to dedicate greater attention to strategic decisions, customer relationships, organizational priorities, and emerging opportunities. The objective shouldn't simply be to accomplish more tasks or manage more businesses. It should be to use the capacity AI creates to exercise better judgment, strengthen trust, and lead more effectively.

Alejandro Zakzuk
CEO
Alejandro writes about reducing decision risk before software, AI, and operational systems are built. His perspective focuses on validation, executive clarity, and building only what the business can defend.
Continue with related insights.
Explore more thinking connected to this topic, from software decisions to AI systems and operational execution.

The Software Works. Now What?
A couple of weeks ago, a friend called me with a business idea.He understood the problem he wanted to solve, had a clear idea of how the product should work,...

Alejandro Zakzuk
CEO

How to Add AI to Legacy Software Without Rebuilding Everything
Most executives asking this question are focusing on the wrong problem.The assumption behind the question is understandable. Organizations spend years...

Alejandro Zakzuk
CEO

The Rise of AI-Native Software Delivery
For years, software development followed a relatively predictable model. Companies gathered requirements, teams wrote code, QA validated functionality,...

Alejandro Zakzuk
CEO