In late 2017, I watched a room of 150 retail investors nod along as I explained the dangers of centralized control. We were gathered in a Chicago community center, dissecting whitepapers that most of them had never fully read. That experience taught me something that has guided every analysis I have done since: the gap between what technology promises and what it actually delivers is where the real human story unfolds.
That gap is exactly where we find ourselves with agentic coding tools in 2026. A new wave of enterprise data suggests that 32% of organizations are choosing to build their own software using AI agents rather than buying off-the-shelf solutions. High-performing companies—those generating at least 5% of EBIT from AI—are nearly twice as likely to skip traditional software purchases. But here is the uncomfortable truth buried in the same reports: only 11% of agentic systems are production-ready, and 40% of agentic AI projects are predicted to be canceled by 2027.
This is not a failure story. This is a governance story.
Let me translate what these numbers really mean, because I have spent the last decade helping communities navigate exactly this kind of transition. When I co-designed the governance structure for UnityDAO back in 2020, we faced a similar chasm between aspiration and execution. We implemented quadratic voting to prevent whale dominance, and participation jumped 300% above industry averages. The lesson was simple: systems fail not because the technology is weak, but because the social and operational infrastructure around them is underdeveloped.
The same principle applies to agentic coding. MIT NANDA's research shows that internal build success rates hover around 33%, while purchasing vendor tools succeeds at roughly 67%. On the surface, this looks like a clear endorsement of buying over building. But I see something different. I see a market that has not yet developed the governance frameworks necessary to support either path effectively.
Here is what the data actually reveals. The gap between pilot adoption and production readiness—17% of organizations have deployed agents, while 75% are experimenting—mirrors what I observed in DAO governance. Everyone wants to vote, but nobody wants to build the infrastructure that makes voting meaningful. The same is true for agentic coding. Organizations are eager to experiment, but they have not invested in the evaluation systems, observability tools, and human oversight loops that separate successful deployments from expensive failures.
My experience with the Rebuild Chicago initiative in 2022 taught me something else about this dynamic. When we organized peer support for 200 former crypto employees after the FTX collapse, we discovered that technical failure was rarely the root cause of their struggles. The real issues were organizational instability, eroded trust, and the psychological toll of watching systems crumble. The same pattern appears in enterprise AI adoption. When internal builds fail, the cost is not just development time—it is organizational morale and institutional knowledge loss.
The industry is asking the wrong question. The real issue is not whether to build or buy, but how to create the conditions under which either approach can succeed.
Let me be more specific about what those conditions look like. McKinsey reports that 20% of organizations already feel the pressure of AI operational costs. Agentic coding workflows can consume 10 to 100 times more tokens than traditional chat-based AI interactions. This is not a technical footnote—it is a fundamental economic constraint that shapes every decision downstream. The organizations that succeed will be those that treat operational costs as a design constraint, not an afterthought.
But here is where I want to challenge the conventional wisdom. The narrative around agentic coding tools assumes that vendor tools are the safer choice because they have higher success rates. But my experience negotiating the Values First coalition in 2025 taught me that high performers do not simply choose between building and buying—they build the capacity to do both effectively. When we united 15 smaller DAOs to negotiate with BlackRock's venture arm, we succeeded because we had developed the internal governance structures that made our collective decisions credible. The same logic applies to enterprise software. The 67% success rate for vendor tools is not a vote of confidence in vendors—it is a reflection of the fact that these organizations have invested in the operational infrastructure to make external tools work.
The 33% internal build success rate, on the other hand, tells me that most organizations attempting to build their own agentic systems lack the governance frameworks necessary for success. They are treating technical implementation as a purely engineering challenge when it is actually an organizational transformation challenge.
This is where the contrarian angle emerges. The conventional reading of this data suggests that enterprises should default to buying. But I see the 33% success rate as an opportunity, not a warning. The high performers—the organizations generating meaningful EBIT from AI—are precisely the ones choosing to build. They are not doing this because they are naive. They are doing it because they understand that the organizations that develop internal AI-building capabilities will have a structural advantage that cannot be purchased.
Let me give you a concrete example from my own work. When I led the Human-First Protocols initiative in 2026, we faced a choice between adopting existing AI governance tools and building our own verification layer. The vendor tools had better initial success rates, but they could not handle the specific requirements of DAO governance. We built our own manual verification system, and while the process was painful, the result was a system that actually served our community's needs. The same logic applies to enterprise software. Off-the-shelf solutions work well for generic problems, but they fail when organizations face unique regulatory, cultural, or operational constraints.
The hidden truth in this data is that high-performing organizations are not simply choosing to build—they are building the infrastructure that makes both building and buying viable options. They are investing in model fine-tuning, evaluation systems, observability tools, and security frameworks that most organizations have not even considered. This is not a technology gap. It is a governance gap.
The Gartner prediction that 40% of agentic AI projects will be canceled should not be read as a market failure. It should be read as a market correction. The organizations that succeed will be those that recognize agentic coding tools are not a replacement for human judgment—they are a amplifier of human capacity. The question is not whether the technology works. The question is whether the organization has the governance structures to use it responsibly.
I have seen this pattern play out across every technology transition I have witnessed in my 27 years observing this industry. The ICO boom of 2017, the DeFi Summer of 2020, the institutional adoption of 2025—each cycle followed the same trajectory. Initial enthusiasm creates inflated expectations, followed by a period of disillusionment as the gap between promise and reality becomes apparent. But then something remarkable happens. The organizations that invested in the underlying infrastructure—the governance, the education, the human systems—emerge stronger than before.
Code without compassion is cold. And agentic coding tools without governance frameworks are just expensive ways to automate chaos.
Let me be clear about what I am not saying. I am not arguing that every organization should build its own agentic coding infrastructure. Most organizations do not have the capacity or the need to do so. The 67% success rate for vendor tools is real, and for many organizations, buying is the right choice. But the decision to build or buy should not be based on success rates alone. It should be based on the organization's strategic position, its competitive context, and its capacity to develop the operational infrastructure necessary for either path.
The organizations that will thrive in this new landscape are those that treat agentic coding as a strategic capability rather than a tactical tool. They will invest in the evaluation systems, security frameworks, and human oversight loops that make AI deployment sustainable. They will recognize that the real competitive advantage is not the technology itself, but the ability to integrate it into existing workflows while maintaining human judgment at the center.
I have spent my career building bridges between technical systems and human communities. The build-versus-buy debate is not really about software. It is about the kind of organization you want to become. Do you want to be a consumer of technology, dependent on vendors for your competitive advantage? Or do you want to be a creator of technology, capable of building systems that serve your specific needs?
There is no universally correct answer. But the data suggests that the organizations generating real value from AI are the ones choosing to build. They are not doing this because they are reckless. They are doing it because they understand that the future belongs to organizations that can create their own tools.
The 33% internal build success rate is not a reason to avoid building. It is a reason to build better governance frameworks, invest in organizational capability, and treat AI deployment as a long-term strategic journey rather than a short-term technical project.
As we move toward 2027, I expect to see the market continue to consolidate. The organizations that invested in governance, evaluation, and human oversight will emerge as leaders. The organizations that treated agentic coding as a quick fix will continue to struggle. The technology is not the differentiator. The human systems around it are.
This is the lesson I have learned from every community I have helped build and every crisis I have witnessed. Technology is a tool. Governance is the art of using that tool wisely. And the organizations that master this art will be the ones that define the next decade of software development.
We are not just building software anymore. We are building the capacity to build. And that distinction makes all the difference.