Over three days last week, Gartner hosted 285 different sessions as part of its Application Innovation & Business Solutions Summit in Las Vegas. The somewhat broad theme allowed more than 150 speakers to explore topics from the esoteric to the universal, digging into the weeds of application migration one moment and philosophizing on the role of trust the next. For attendees, this resulted in a rich tapestry of lessons — no matter which combination of lectures you chose.
Still, the variety of subject matter would eventually begin to converge on some key themes. This felt more reassuring than repetitive; it left little ambiguity over what to take back to the workplace.
Speaking across every aspect of the business, presenters pulled at the same threads: the importance of human ingenuity, the struggle to successfully navigate the final mile of AI deployment, the need for governance now more than ever. Each mention added more nuance and context, helping attendees translate thoughtful insights into meaningful, organized action.
The action piece is critical because execution is where many organizations have stumbled, whether due to a lack of information or the pressure to move quickly. And the stakes are high; as Gartner Senior Director Analyst George Sellner explained, “organizations that skip steps often end up scaling inconsistency instead of capability.”
The first part of not skipping steps is knowing what the steps are in the first place. It would be impossible to condense 285 sessions into a single, digestible summary. Some insights will remain in the minds of the attendees alone. But Gartner’s event did highlight some specific lessons that CIOs, enterprise architects and IT leaders would be wise to heed.
The human factor remains paramount
For an event focused on application innovation and business solutions, the real hero of the day wasn’t technical at all: it was the human touch. In the opening words of Tuesday’s keynote, human ingenuity was called out as the singular compounding factor for successful AI-in-the-enterprise initiatives. Rather than seeing AI as a potential replacement for human activity, senior analysts Jason Wong and Brent Stewart emphasized that AI was an amplifier.
“When human-AI partnership is done right, AI compresses the cost of iteration and exploration so humans can spend more time doing the work only humans can do: innovating and transforming your business capabilities,” Stewart said.
This need to keep humans “in the loop” would be referenced throughout the next three days. Aaron Lord, a senior director analyst at Gartner, likened AI models’ current cognitive ability to that of a child; they know enough to understand that a tomato is a fruit, but not enough to realize it doesn’t belong in a fruit salad. And just like a child, AI needs human guardians to keep it developing on track.
Gartner Senior Director Analyst Birgi Tamersoy focused on AI’s innate fallibility and the inevitability of mistakes; a 99% success rate means a 1% failure rate, by definition. “There is too much ambiguity in real-world situations to ensure a binary understanding,” he said. This makes the human ability to interpret and manage risk not just valuable, but vital. AI can be a useful tool, but only when wielded with nuance by people.
Even the most ambitious AI perspectives placed humans in an important role.
Deepak Seth, a senior director analyst at Gartner, argued that human performance can sometimes set a low bar, and therefore AI can appear like a tempting improvement; he gave the example of human-caused driving accidents vs. the safety of autonomous cars. But actually, Seth reasoned, AI cannot do everything that executives might think it can. And even if — or when — it has been improved to meet these expectations, the “human-in -the-loop” will be the crucial bottleneck that stops any error in its tracks.
“AI agents are not inherently a direct replacement for humans,” Seth declared. And it seemed like everyone agreed.
Everyone is still getting AI wrong
Humans may be invaluable, but AI is still the future. And yet, despite years of talking about it, experimenting with it, throwing money at it, it seems like very few people have AI figured out. The data doesn’t lie: Gartner has found that 90% of AI pilots don’t actually move past the pilot phase. Only 5% are in production.
“Your organizations are in constant motion: pilots, demo, prototypes,” said Stewart in the opening keynote. “But the real product, the actual business value, keeps collapsing.”
He and Wong attribute this to AI being a multiplicative system, rather than an additive one. They created an equation for business value from AI: Business value = (model capability x workflow fit x trust x governance) to the power of human ingenuity.
If any value inside those parentheses falls to zero, the total sum of those parentheses falls to zero as well. And unfortunately, too many people are focusing on just one or two of those values, neglecting the others at the expense of the project’s actual impact on operations. There’s no use in optimizing for the best possible model if there is no useful workflow fit or low adoption due to poor trust from employees.
This miscalculation echoed through the presentations that followed. Before each exploration into a new AI strategy or application, speakers would acknowledge all the ways organizations currently misunderstand the technology. The good news is that anyone failing to move an AI pilot into production is in good company and has yet to truly fall behind. There’s still hope, as long as they step up their game in time to take advantage of the next best thing in AI.
Agentic AI is no fad
As impressive as ChatGPT was when it launched, the tech community is already moving away from LLM-supported chatbots and hyping the next generation of AI advances. Agentic AI was mentioned in nearly every talk, and it was the unspoken undercurrent in all others, save perhaps the final guest keynote, “Leading with Levity.” But while agentic AI isn’t particularly funny, it is exciting.
ServiceNow’s Jithin Bhasker, general vice president and general manager of AI application platform and developer products, positioned agentic development as the final phase of app development, following low-coding, AI-assisted development and then vibe coding.
Gartner’s Seth predicted a future where agentic AI is able to assume all basic-level roles where decision complexity is low, for example, customer service, IT SDLC and ITSM. We may not be there yet: A 2025 Gartner survey of 360 IT leaders found that only 15% of organizations were considering, piloting or deploying fully autonomous AI agents. Yet the same survey found that 75% were piloting, deploying or had already deployed some form of AI agent; truly agentic AI would be the next step.
Camunda , a process orchestration software vendor, offered perhaps the most provocative demonstration of what isn’t just around the corner but already taking shape. Peter Vaccarella, global head of solutions consulting at Camunda, emphasized that too many companies are focusing on bolting on solutions to existing systems — which is insufficient in the world of AI. “Every single process that you have today in your enterprise is legacy,” he said.
Vaccarella described the vendor’s new platform, ProcessOS, as not just another agentic AI tool, but an operating system designed for AI agents themselves. Rather than taking an existing process and trying to embed agentic AI into it, ProcessOS aims to use AI agents to re-engineer the workflow entirely to achieve better outcomes. Instead of giving a caterpillar a jetpack, Vaccarella explained, Camunda is trying to invent the very process of metamorphosis.
Trust is the goal, governance the path
Given the growing public wariness of artificial intelligence and the mass layoffs that have run rampant across enterprise companies, it’s clear that trust in AI is on the decline. But in all its forms — whether in the output of an AI model, between employee and leadership, or between vendor and client — trust is paramount for business success.
The keynote described trust as the “new first principle of user experience in the AI era.” Simply put, you can’t expect adoption of a new technology without sufficient trust. This is true both internally, in terms of employees picking up new AI-assisted workflows, and externally; Plat4mation’s Greg Clock said that building trust early with a customer is what can make the difference in clinching long-term contracts. And more often than not, trust goes hand in hand with governance.
It may not be the most exciting factor, but governance is central to successful enterprise IT. Speakers repeatedly emphasized that governance needs to be a part of the development process; “every moment you spend putting out fires in governance is an hour stolen from winning the race in mission impact,” Sellner said.
By building governance into systems from the beginning, enterprises can also get ahead of one of the biggest challenges with AI deployment: scaling. After all, slow scaling is a sign of poor governance, not of governance itself, as Wong noted. That’s not to say that this is something you can set and forget; good policies should be revisited consistently and adapted to evolving circumstances. But taking the time to lay a governance foundation will actually save you headaches in the long run and make it easier to build future policies.
Choose to tack on regulation at the end, and organizations may find themselves undermining their larger investment. “When governance is fragmented across programs and functions, AI becomes risky and inconsistent,” Sellner warned.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.