6 lessons from top execs as they build the future of AI

We had the opportunity to sit down with six leaders at the forefront of AI innovation — founders and builders who are not just scaling technology but defining its impact. Through our SoundBytes series, they shared the challenges, breakthroughs, and strategies shaping the next era of AI.

Below, we share their most compelling insights.

1. Set a clear vision and be intentional when measuring against it

AI can be most impactful when aligned with specific business goals. Any new venture, whether a project to implement AI or otherwise, needs a clearly defined thesis, said Carmine Visconti , CEO of Quantive . He explained, “AI is playing a critical role in challenging the leaders who are running strategic initiatives and evaluating their ‘theory of victory.’” And added: “AI helps you diagnose and understand better than you can do on your own” and can stress test, iterate, and evaluate your project vision.

“AI is playing a critical role in challenging the leaders who are running strategic initiatives and evaluating their ‘theory of victory.’” — Carmine Visconti

Pipefy ’s Founder and CEO Alessio Alionco , described a similar experience regarding adoption: Enterprise boardrooms need to show progress with AI but don’t always know to what end. He suggested working backward from the ultimate vision: “Think about what you would want to achieve if you had unlimited time and resources and the best professionals on the planet. What would that experience look like? Then, work backward from that vision to see how you’re going to transform your day-to-day.”

Writer ’s Chief Strategy Officer Kevin Chung discussed the importance of a mindset shift and collective buy-in to a business’ AI objectives. “Everybody in a company has the mission to drive a particular goal,” he said. “And if we can use AI to help accelerate and get to that goal faster, then everybody’s on board.”

He also explained the importance of knowing how to measure the success of an AI project: “Ultimately, the end state is: Is it driving ROI in a meaningful way? Is it making my life better? Is it giving my employees superpowers?”

2. Use proprietary data to personalize AI

Using proprietary data sets is central to gaining a competitive advantage with AI. SingleStore 's Chief Innovation Officer Rahul Rastogi explained how vector similarity search (VSS) — finding items in a dataset similar to a given query vector — can unlock enterprise data when building AI applications.

“It’s an important piece in the puzzle when you're working with large existing corporate data sets…no longer do you have to go across and run a query in a transaction database.” Businesses can easily retrieve and use their proprietary data at scale using VSS and, through integration with existing systems that use standard SQL queries, can use it to unlock personalized AI applications for every employee.

“The proprietary information of a company is its unique secret sauce. That’s something nobody else has,” explained Chung. “It feels like the first time we've had the capability to tap into and understand a wealth of data across many different systems.” And, as Parker Mitchell , founder and CEO of Valence , said, tapping into a company’s large data sets is unlocking hyper-personalization, adding value by “moving from general AI to specific AI for individual users.”

“If an AI co-pilot has a lot of or all of the contextual data on myself as the manager, on my team, on the organization, that becomes a really effective agent to help me do my role better, and really build and drive a high-performance team.” — Jenny Podewils

This personalization can lead to faster, more data-driven decisions. As Leapsome Co-CEO and Cofounder Jenny Podewils noted: “If an AI co-pilot has a lot of or all of the contextual data on myself as the manager, on my team, on the organization, that becomes a really effective agent to help me do my role better, and really build and drive a high-performance team.”

Acceldata ’s CEO Rohit Choudhary explained that future companies will be either the ‘AI haves’ or the ‘AI have-nots.’ He added: “The companies that have got their data sorted have already reaped the benefits of their data being in the right place and in the hands of the right people,” predicting that those who don’t organize their data will struggle.

3. Overcome security barriers through confidential computing and strong governance frameworks

Many of AI’s critics have pointed to data security challenges, which can hinder adoption in some companies. Anjuna Security ’s CEO and Cofounder Ayal Yogev discussed how confidential computing can meet privacy and regulatory requirements: “Traditionally, when an application needs to process data, it has to decrypt the data, and at that point, the data is exposed.” But a data clean room containing only the data and AI model means the raw data is never exposed. For Yogev, this shift to confidential computing is “enabling collaboration, regardless of how sensitive the data is.”

Similarly, Skyflow 's CEO Anshu Sharma spoke about their polymorphic encryption engine, which allows a user to do a lot of work with data without ever converting it into plain text. Anshu explained: “With a polymorphic engine, you can interact with these models with all the controls in place. But at the same time, you have governance rules imposed on it.”

“Having a framework that is part of the same technology stack helps you govern thousands of these Agents, ensuring they operate in a way that is CFO compliant and operationally compliant.” — Chandar Pattabhiram

These governance frameworks could include role-based access controls and human-in-the-loop systems to balance security and functionality, explained Chandar Pattabhiram , chief go-to-market officer at Workato . “The agile governance piece is very important. Having a framework that is part of the same technology stack helps you govern thousands of these Agents, ensuring they operate in a way that is CFO compliant and operationally compliant.”

Nexus Cognitive ’s CEO Anu Jain explored how a clear governance model for data can make for faster, more effective AI initiatives: “Having a clear data owner in the data governance area helps data scientists receive consistent and accurate data. They're no longer being data wranglers or data engineers and really get to focus on what they do…increasing speed to value.”

A strong governance model, combined with robust security, can help ensure that confidential data is handled in the appropriate manner while still gaining the maximum benefits.

4. Reduce errors and amplify humans to optimize productivity gains

Blending generative AI and rules-based automation can reduce manual errors and enhance workflows, said Templafy ’s Cofounder Christian Lund . “You can have different areas of a document being informed by AI. Blend that with different AI models and more classic rules-based automation. That's the way to get to these highly reliable outputs that otherwise can be super difficult.”

In critical decision-making processes, use multiple models to validate AI-generated insights, ensuring accuracy. Exiger 's CEO Brandon Daniels said: “We [challenge] our AI adjudication of risk of error [against] what we're seeing coming from another generative model, and then compare the two answers. And if those decisions match, then we know that we can move forward to our customers with high confidence.”

“You can have different areas of a document being informed by AI. Blend that with different AI models and more classic rules-based automation. That's the way to get to these highly reliable outputs that otherwise can be super difficult.” — Christian Lund

Then, AI can simplify more processes across systems and teams by using orchestration tools to eliminate silos, said Pattabhiram. “The LLM can then say, ‘Here are all the things you need to do’…have a human in the loop, if needed, and then an orchestration runtime technology to execute the process.”

In marketing teams, this orchestration is being used for real-time content iteration and performance feedback, enabling rapid decision-making in marketing campaigns — and beyond, as Jasper ’s CEO Timothy Young pointed out. One of its customers is using AI to produce “banner ads at scale, and then publish those ads, watch the performance, and do a complete iterative loop.”

Alionco believes the use cases for AI will be almost limitless and that those who don't take AI seriously will be outpaced. “Almost every job on the planet,” he said, “will be amplified by the usage of AI. If you do not use AI as the extension of your brain, you’re probably going to be left behind.”

5. Build AI initiatives with scalability in mind from the start

As AI applications scale, their implementation becomes increasingly difficult. To combat this, Raphael Ouzan , CEO and cofounder of A.Team , stressed the need to establish robust systems for data observability, compliance, and cost efficiency upfront — and have a recipe to go from prototype to production. “There's almost this Pareto Principle moment, where you’ve got to 60% or 80%, and then it takes a very long time to get to the finish line,” he said.

With increased human-in-the-loop AI workflows, the amount of code generated and pushed will continue to scale enormously, resulting in a paradigm shift for deployments and version control systems.

“You need the ability to target a subset of users; you need the ability to roll back if things are going wrong.” — Dan Rogers

LaunchDarkly ’s CEO Dan Rogers said, “The core concept for solving ‘bad ship’ is progressive rollouts or targeted releases. The second is the ability to roll back very quickly.” This is because AI applications have a particularly unique risk to them, as many of them are non-deterministic and stochastic. As Rogers shared: “You need the ability to target a subset of users; you need the ability to roll back if things are going wrong; and you need the ability to experiment very quickly on different prompts and different configurations.”

“Some of the best builders from tech giants want to work with, for example, an education company where they have the potential for scale and impact,” said Ouzan. Builders aren’t just looking to scale their coding output with AI; engineers want to build AI tools that scale their impact on the world and work on meaningful problems.

6. Implement cross-functional AI Agents to empower teams

CrewAI 's CEO Jo?o (Joe) Moura explained that orchestrating all your AI Agents into a dynamic system will become increasingly important. AI Agents should move beyond isolated tasks to integrated systems that collaborate and adapt dynamically across business processes. Moura stressed that giving these Agents genuine agency “is going to be crucial for the long-term success of agentic AI implementations.”

And, once you give AI Agents agency and eliminate silos, they can carry out cross-functional workloads and handle increasingly dynamic tasks. But, at least for now, having a human in the loop provides extra safeguards as the technology matures.

Pattabhiram explained: “The technology enables you to get the human in the loop at any step in this process, depending on the task.” And, as a rule, “The more an agentic system interacts with external parties in their process, the more important it is to have a human sit in the loop,” suggested Pattabhiram.

AI agentic systems will orchestrate the future, according to Copado 's COO Sanjay Gidwani . He explained: “AI Agents are the plan: The build Agent, test Agent, release Agent, and operations Agent. They can span the entire DevOps life cycle and empower teams to focus on innovation.”

“AI is here to help humans. The human brain is the most powerful computer that can never be matched.” — Sanjay Gidwani

But he conceded that AI won’t replace us. “AI is here to help humans. The human brain is the most powerful computer that can never be matched.” Instead, he argued that AI will allow us to focus on the most complex tasks and stretch people to achieve their true potential.

AI continues to transform how businesses innovate and operate, but the experts are clear: Successful initiatives demand vision, a clear way to measure success, a robust data strategy, first-rate security, and deployment at scale.Check out the full SoundBytes library for more inspiration and advice from some of the leading executives driving AI initiatives today.

Thanks for reading Flight School. Be sure to subscribe and share it with friends and colleagues here. We're excited to have you in our community of innovators and leaders.

Where do you see the biggest opportunities for AI in the next five years? Let us know in the comments below ??????

Note: The use of genAI poses a number of risks, including privacy, intellectual rights, and information security concerns.? We suggest conducting a comprehensive risk assessment before engaging GenAI or incorporating it into your processes.