From Invention to Adoption: Building the Human Infrastructure for Innovation in the AI Era
The team wanted to build an AI agent.
The idea had momentum. The engineers were capable. The technical path looked plausible. In many organizations, that would have been enough: a new tool, a motivated team, a fast-moving mandate, and a prototype waiting to happen.
But one member of the team hesitated. She had recently returned from SAP’s Institute for Product and Engineering, an organization designed to help engineers and leaders to innovate more effectively in conditions of uncertainty. Before the program, she told me, she probably would have started building right away. This time, she stopped the group to ask a more uncomfortable question: what problem were they actually solving?[i]
The business case was thin. Customer demand was untested. Long-term maintenance costs were unclear. The issue was not whether the team had the talent to build the agent. It was whether building it would create value.
The reaction was immediate. Her pause introduced friction. Progress appeared to stall. “My interruption didn’t feel like good judgment,” she said. “It felt like I was slowing everyone down.” What she experienced as responsibility was read as obstruction in a system optimized for forward motion rather than course correction.
This is the real management challenge of the AI era. Not simply whether organizations can build new tools, but whether they can create the human conditions that allow new practices to take hold in a reliable and persistent way.
The innovation barriers created by AI
Across industries, “We must do AI” initiatives push systems faster than they can change. Risks go unspoken. Decisions default to activity over judgment. We celebrate the exponential curve of our tools while ignoring the linear limits of the humans expected to build and use them. And that mismatch quietly breaks innovation from the inside out.
AI has made invention faster, cheaper, and more visible, but it has not solved the harder problem of adoption: a 2025 MIT/NANDA report found 95% of enterprise GenAI pilots had no measurable P&L impact, with flawed workflow integration and learning gaps cited as central causes. The bigger picture is that adoption focus was missing: deciding what is worth building, surfacing risk, challenging weak assumptions, and coordinating change across systems where mistakes carry consequences.
AI acceleration is uneven: it multiplies activity while leaving necessary human practice changes untouched. The mechanisms that once supported exploration and coordinated risk-taking are now optimized for revenue, predictability, and execution.
And so, AI is driving an unintended consequence: by optimizing the creation of technology, it is making an existing blind spot bigger: the necessity of considering the human adoption factors which are just as important to any innovation story, especially in complex enterprises.
How AI exacerbates the problem of mistaking invention for adoption
These patterns are exacerbated by AI, but they’re not new. Innovation scholar Peter Denning explains that innovation is not the invention itself, but the adoption of a new practice by a community. That distinction matters most in complex enterprises, where new capabilities must survive handoffs across departments, adapt to existing workflows, and operate inside systems built for continuity and accountability. In these environments, innovation doesn’t fail only because technology is weak. It fails because organizations often lack the human infrastructure required to absorb it.
That distinction reframes the management task. Leaders should not ask only, “What can we build?” or “How quickly can we deploy it?” They also need to ask: “What human conditions must exist for a new practice to be trusted, tested, challenged, modified, and sustained?”
Those conditions are not peripheral. They include trust, psychological safety, judgment under uncertainty, cross-boundary collaboration, the ability to surface risk early, and the discipline to listen before prescribing a solution. And all the AI in the world will not fix an organization whose people cannot safely question weak assumptions, negotiate new commitments, or change how decisions are made. In that sense, the AI era doesn’t make the human side of innovation less important. It makes it impossible to ignore.
The SAP Institute offers one example of what it looks like to treat these human conditions as central to innovation, not as soft benefits around the edges. Its programs suggest that large technology organizations may need more than “innovation theatre”: hackathons, innovation events, and technical upskilling. What’s missing: institutions designed to help people become carriers of adopted practice.
Internet pioneer Vint Cerf, co-designer of the TCP/IP protocols – arguably one of the most important innovations of the 20th century – has advised the Institute since its creation. Cerf echoes this understanding: “Innovation is nothing without adoption! To be effective, an innovation needs to be implementable and sometimes interoperable with existing practices. This is a major design challenge. Innovation in vacuo may be sterile.“[ii]
This distinction matters most in enterprise environments – not because they resist innovation, but because they represent one of its largest remaining opportunities. Much of the world’s critical work still runs on systems designed for stability and accountability, while dominant innovation models assume cheap failure and easy reversibility.
This is a far cry from consumer-product software innovation, which has dominated the conversation. When these models are projected onto enterprises, innovation stalls because adoption is misaligned with systems that cannot afford to break.
Another Institute experience illustrates that the difference between invention and adoption becomes clearest when innovation survives contact with real work.
One SAP cohort shows what this looks like when adoption is built into the work from the start. Engineers worked with SAP’s Business AI group on ten possible Joule for Consultants agent use cases. All ten became prototypes; only two became products: fit-gap analysis and document generation.
That narrowing was the point. The work took place with internal customers, integration requirements, and accountability for actual use already present. The result was not more innovation activity, but a filter for what could survive adoption.
One of the agents later contributed to an early $50 million agent deal, while the broader J4C opportunity was estimated at $1 billion.
This is what innovation looks like when adoption is treated as the primary design problem rather than a downstream concern. Most ideas are discarded. A few survive. But the ones that do arrive already embedded in practice, not waiting for permission to matter.
The two kinds of innovation stories we tell
Most of the stories we tell about innovation come from environments where failure is cheap, reversibility is high, and decisions carry limited downstream consequence. Advertising, content, recommended products. In those settings, innovation is often measured by speed, iteration, visible experimentation, and the production of new artifacts: pilots, prototypes, workshops, labs, demos, and tools.
That story travels well. But it doesn’t travel far.
In complex enterprises, the problem is different. New practices must survive handoffs across departments, adapt to existing workflows, align with incentives and governance, and operate inside systems where mistakes carry consequences. Here, a working prototype is often only the beginning of the adoption problem.
AI makes this distinction harder to ignore. It can accelerate the creation of new tools, but it doesn’t automatically create the human, organizational, and operational conditions required for those tools to change how work gets done. That is why so many innovation efforts end in the same frustrating place: a capable system, a polished demonstration, a trained cohort—and no reliable, persistent change in practice.
The dominant innovation narrative emerged largely from digital-native environments, where products can be tested quickly, changed continuously, and sometimes abandoned with limited systemic consequence. That narrative has become so familiar that it is often treated as a general account of how innovation works. It is not.
Governments and large enterprises operate a different class of systems: manufacturing, finance, healthcare, energy, agriculture, logistics, public administration, and the business processes that connect them. These systems are deeply interconnected. A change in one area can ripple across many others. They are persistent; decisions accumulate over time, increasing both impact and inertia. And they are externally constrained by legal, regulatory, audit, and customer obligations that assume continuity and predictability.
These conditions do not make enterprise innovation slower by nature, nor do they reflect a lack of technical ambition. They make it structurally different. In these environments, innovation is less about introducing novelty than about integrating change without breaking what already works.
That requires a different approach to innovation management: one that treats adoption not as an afterthought, but as the central design problem.
Why most innovations fail after they’re built
Most innovations do not fail because they do not work. They fail because invention is mistaken for adoption. Once a system performs as intended, organizations assume the hard part is over, only to discover that little has changed in how work actually gets done.
It’s tempting to diagnose innovation failures as due to cultural drift or leadership issues. But the roots run deeper. Frederick Winslow Taylor helped popularize production management logic built around production-line measurement: eliminate variation, reward compliance, optimize for efficiency. That logic still quietly governs many large organizations. It is useful for predictable execution, but it undermines innovation because innovation depends on people doing something new before the outcome is fully known.
Panasonic founder Konosuke Matsushita saw this problem clearly. He argued that western management would struggle as complexity increased, because methods designed for efficiency and control were no longer enough. Once work becomes complex enough, organizations need more than better plans, better metrics, or better instructions. They need people at many levels who can think, adapt, coordinate, and take responsibility for change.
This is where Denning’s distinction becomes essential. Innovation is not complete when a technology works. It is complete only when a community adopts a new practice.
Adoption failure isn’t user resistance
The language of “people resisting change” is widespread, but wrong. In most cases, people are not resisting innovation itself; they are responding to the absence of the conversations that make adoption possible.
Denning’s work points to a different diagnosis. Innovation takes hold through eight conversational practices: sensing, envisioning, offering, adopting, sustaining, executing, leading, and embodying. Expertise in these practices turns a working invention into a shared way of working. Says Denning, “[These conversations capture] moods, contradictions, power. Once you understand this you can…navigate effectively through innovation”[iii].
Yet when organizations skip the right conversations, there are predictable failures. The prototype works, the demo is polished, the training is complete, yet behavior doesn’t change. Leaders then blame culture, inertia, or users. But the deeper failure is that no durable conversational system ever formed around the new practice. The result: fewer than one-third of digital transformations succeed, and far fewer sustain long-term performance improvement.
Social theorist Howard Bloom has argued that human groups behave less like collections of independent individuals than like “superorganisms,” whose survival depends on both variation and coherence. Innovation requires difference—people willing to introduce new possibilities—but adoption requires enough shared structure for those possibilities to be absorbed rather than rejected.
Until the invention vs. adoption distinction is owned, completely and explicitly, the result will be a growing graveyard of technically impressive innovations—built as if the future were predictable, deployed into systems where it is not—that never quite change how work gets done.
Innovation at enterprise scale
If innovation is constrained by human systems, that constraint should be most visible where failure is costly and accountability is durable. SAP—and the enterprises it serves—operate squarely in that terrain. The opening story shows this constraint in miniature: a pause that looked like friction was actually good judgment.
This tension is becoming more common, not less. Deloitte’s 2023 Global Human Capital Trends report describes a workforce increasingly shaped by worker agency, cocreation, meaningful work, and personalized career paths—yet most enterprise cultures still reward speed and certainty over judgment.
Nothing in this story is about tools or talent. The system didn’t fail because the technology was weak. Instead, the system had no way to reward the judgment that would have kept it on course.
The Institute as a human-capacity experiment
Institute graduates described it as a protected environment—temporarily insulated from execution pressures that shape enterprise work. Inside the program, disagreement could surface without immediately triggering defensiveness. That protection was intentional.
The harder test came afterward. One participant returned just as a senior sponsor changed roles. The work itself had not changed, but the surrounding conditions had. Decisions became more guarded. Questions that had once been welcomed now carried greater risk. “The hard part wasn’t learning new behaviors,” he said. “It was learning when and where they would hold.”
This tension is not unique to SAP. Large-scale studies of digital and organizational transformation consistently show big investments with limited success, such as this one showing that less than a third achieve sustained impact. Not only must adoption be built into projects from the start, but leadership must also sustain the conversations that adoption requires.
Psychological safety as the missing infrastructure
In environments where failure carries consequence, innovation depends on whether people can speak honestly despite the risk. This is where psychological safety stops being cultural aspiration and becomes functional infrastructure.
Unless thoughtfully managed, fear doesn’t announce itself; it quietly shapes which conversations happen at all. People soften assessments until they carry no information. Requests go unmade. Commitments become ceremonial.
What makes this insight particularly compelling in the context of the SAP Institute is that it emerged again—independently—through practice. Institute director VR Ferose arrived at the same conclusion without reference to Edmondson or Denning. Faced with repeated reports of the failure of promising ideas to take hold, he found that the bottleneck was more human than technical. Teams avoided socially dangerous yet innovatively necessary behaviors. And AI made it worse.
Ferose went beyond advocating for safety, to personal vulnerability. He spoke openly about his own missteps, and named uncertainty without rushing to resolution. This was not confession or charisma. It was a signal of the value of navigating uncertainty together.
Ferose and his team also invited other SAP executives into the Institute and asked them to show up in the same way. For participants, this was often disorienting. One engineer described watching a senior executive she had known for years—someone she associated with polish, positivity, and certainty—speak candidly about burnout and mistakes. “I realized,” she said, “that the person I thought I knew at work was only a fraction of the reality. And if they could say those things out loud, maybe the rest of us could do so as well.”
That moment, echoed across my interviews with Institute graduates, marked a shift. It wasn’t that the room suddenly felt safe. It was that the cost of honesty visibly dropped. Students described relief: the sense that the conversations innovation requires were finally permitted, and the tension between the necessity for innovation and the lack of its preconditions was reduced.
Institute graduates explained the benefits. Engineers who would normally default to building first learned instead to stop and ask, “Why?” Managers admitted uncertainty. Challenges surfaced earlier, when they could still alter direction. And none of this was framed as a lesson in vulnerability. It emerged because the social risk of Denning’s conversations had been lowered.
Stanford psychologist Jamil Zaki studied Institute graduates, discovering that leaders resisted vulnerability not because they doubted its value, but because they expected these interactions to be awkward and unproductive, when they are often the opposite.
Using an instrument co-created with author and social entrepreneur Jane Chen, the Institute measured its impact. Across cohorts, teams reported measurable gains in trust and leadership behavior under stress. Institute-trained managers showed nearly twice the improvement in Leadership Trust over a year and sustained engagement longer than matched peers. Supporting its findings, Gallup research also focuses on engagement, showing that it’s tied to productivity, profitability, turnover, and other business outcomes.
The common thread of these studies is that psychological safety, rather than belief or buy-in, proved decisive in whether more effective innovation practices could take hold before decisions hardened.
Innovation worked, until it didn’t
Innovation’s dependence on adoption is nothing new, of course. But the dominant innovation model was built for settings where failure is cheap, iteration is fast, and technical progress looks like the main constraint. AI exposes where that model breaks.
AI accelerates capability, but not the human capacity required to absorb it. In complex organizations, the scarce resource is judgment under pressure: the trust to challenge weak ideas, surface risk early, and coordinate change.
SAP’s Institute points to one institutional response: build adoption capacity, not technical activity. For leaders, the question is no longer what can we build, or how fast can we deploy it. It is whether a new practice can be trusted, tested, modified, and sustained.
This is the hard part for complex organizations now: building sociotechnical systems that stay coherent as capability accelerates. Progress depends on people capable of carrying consequence: able to hold uncertainty, resist premature certainty, maintain a disciplined beginner’s mind, and bridge those who introduce new ideas with those responsible for sustaining coherence at scale.
Questions for Leaders
- Are we measuring innovation activity, or whether new practices are being adopted?
- Are we pressuring for AI invention but not creating the conditions for its sustained adoption?
- Do our programs lower the social cost of surfacing risk?
- Where are people mistaking speed for progress?
- Do teams have the trust to challenge technically successful but strategically weak ideas?
- Are we building repeatable human capacity, or relying on exceptional individuals?
[i] The author’s account of Denning’s innovation framework draws on interviews with Peter J. Denning, as well as Denning’s published work on innovation, adoption, and conversations for action. The account of SAP’s Institute for Product and Engineering draws on interviews with VR Ferose, Rana Chakrabarti, and SAP Institute participants conducted by the author.
[ii] Vint Cerf, personal communication with the author, January 21, 2026
[iii] Peter J. Denning, Personal communication with the author, October 202
Reprinted from Zenodo.
