Artificial intelligence remains one of the most consequential forces reshaping business, yet many organizations still struggle to distinguish meaningful breakthroughs from attention-grabbing headlines. While public discussion often centers on increasingly powerful models, digital assistants and speculation about artificial general intelligence, many enterprise leaders are discovering that the most transformative AI developments occur behind the scenes.
Ask 10 AI experts what will matter most a year from now, and you might expect 10 different answers. Instead, members of the Senior Executive AI Think Tank—a curated group of experts specializing in machine learning, generative AI and enterprise AI applications—arrived at a strikingly similar conclusion: The biggest opportunities—and risks—aren’t tied to the next model release. Across industries, they point to the infrastructure that makes AI useful in practice, from governance and security to evaluation, trust and workflow integration. At the same time, many are skeptical of some of today’s loudest predictions, particularly around fully autonomous agents replacing human judgment at scale.
As recent research from McKinsey suggests, organizations are increasingly finding that AI success depends less on access to cutting-edge models and more on the ability to operationalize them effectively. The experts featured here—those on the front lines of AI innovation—share the developments they believe leaders are underestimating, the trends they think are overhyped and where executives should be investing now to create lasting competitive advantage.
The Hidden Foundation of Agentic AI
Few experts see bigger changes ahead than Manu Agrawal, Chief Architect at Oracle, which develops healthcare infrastructure and AI systems used by providers and public-sector organizations around the world.
Agrawal argues that the most important AI development over the next year will not be smarter models themselves but the operational infrastructure required to make agentic systems trustworthy.
“The AI development that will matter more than people realize is not simply better models, but the infrastructure required to operationalize agentic AI reliably in production,” Agrawal says.
She points to orchestration, governance, evaluation, memory, observability and policy-aware execution as the components that will determine whether organizations can successfully deploy AI at enterprise scale.
“The biggest shift will be from building impressive AI demos to deploying trustworthy systems that enterprises can safely operate at scale.”
At the same time, Agrawal believes the market is overestimating how quickly autonomous agents will replace human work.
“The most overhyped area will be the assumption that fully autonomous agents can broadly replace human workflows in the near term.”
Instead, she expects governed autonomy supported by auditability, oversight and workflow integration to become the dominant model.
Falling Costs Will Reshape Competitive Advantage
Marc Massar, Founder of AURA Labs, sees economics—not intelligence—as the biggest underappreciated force in AI.
“Underrated: inference cost collapse for reasoning models.”
Massar believes organizations are dramatically underestimating how quickly AI operating costs will decline.
“What costs $10 today will cost cents in 12 to 18 months.”
That shift could fundamentally alter competitive dynamics and invalidate strategic assumptions built around today’s cost structures.
Meanwhile, Massar rejects predictions that knowledge workers will be widely replaced by autonomous systems in the near term.
“Production reliability for multi-step tasks remains poor. Most AI-attributed layoffs are over-hiring corrections with a better label.”
His advice to leaders is straightforward: Invest more heavily in observability, evaluation and governance than in expensive model experimentation.
“Most attention goes to the ceiling, meaning who ships the smartest flagship model. The bigger trend over the next year is the floor rising.”
The Rise of Open Models and Portable AI Architectures
Danish Shaikh, Engineering Lead for ML/AI at Meta, has spent more than a decade building AI systems that reach hundreds of millions of users. His experience spans Meta, Twitter, Roku, Alibaba and Rakuten, where he has led large-scale recommender systems, machine learning infrastructure and agentic AI initiatives.
While much of the AI industry’s attention remains fixed on the latest frontier models, Shaikh believes leaders are overlooking a more consequential trend.
“Most attention goes to the ceiling, meaning who ships the smartest flagship model,” Shaikh says. “The bigger trend over the next year is the floor rising.”
According to Shaikh, open-weight models continue to narrow the performance gap with proprietary frontier systems. For many organizations, a customized open model trained on company-specific data may soon deliver better results than a generic frontier model.
“A capable open model post-trained on your data will match a generic frontier model on the tasks your business actually runs,” he says. “That keeps your fate in your own hands.”
As model capabilities become increasingly commoditized, organizations are focusing more on proprietary data, evaluation frameworks and workflow integration as sources of competitive advantage.
“The overrated bet is that the frontier model is the moat,” Shaikh says. “Raw capability is a commodity you can rent.”
Instead, he argues that organizations should invest in model portability and infrastructure that allows them to switch providers as the market evolves.
“The real value is your data, evals and the harness around it.”
For leaders making investment decisions today, Shaikh offers a straightforward recommendation: “Own the harness, not just the model. Invest in the model-agnostic layers, architect for portability and build the ability to post-train open models.”
Security Becomes the Next AI Battleground
Egbert von Frankenberg, CEO of Knightfox App Design Ltd., brings more than 16 years of experience in AI strategy, cloud computing and digital transformation. Through Knightfox, he helps organizations implement AI-powered automation, analytics and customer experience technologies across multiple industries.
Von Frankenberg believes one of the most underestimated developments in AI has little to do with model intelligence and everything to do with security.
“AI agent infrastructure security” will become a defining issue over the next year, he says.
As organizations increasingly deploy networks of autonomous agents that communicate across enterprise systems, new vulnerabilities are emerging.
“We’re moving from ‘human prompts AI’ to hundreds of autonomous agents acting across enterprise systems via MCP,” von Frankenberg says. “But there’s no security layer for this.”
Traditional cybersecurity tools were not designed to monitor or govern autonomous software agents.
“Traditional firewalls don’t understand AI agents,” he says.
The challenge is already becoming visible. Von Frankenberg points to examples of weaponized AI agents and supply-chain attacks as indicators of what enterprises will soon face at scale.
In contrast, he believes predictions surrounding artificial general intelligence continue to dominate headlines despite more pressing operational realities.
“Mostly hype are AGI timelines,” he says. “The bottleneck isn’t model intelligence—it’s trust, reliability and governance.”
For leaders, that means investing in trust infrastructure now rather than waiting for future breakthroughs.
“Invest in agent governance infrastructure now,” von Frankenberg says. “Starting late means starting from zero while competitors have millions of interaction patterns.”
The Quiet Supply Chain Revolution
Uttam Kumar, Engineering Manager at American Eagle Outfitters, has spent years delivering technology solutions for some of the world’s largest retailers. His expertise spans point-of-sale systems, inventory management, order fulfillment and customer experience optimization.
While many organizations are captivated by AI-generated content and design, Kumar believes the most important advances will occur much deeper inside operational systems.
“AI-driven predictive demand forecasting at the local store level will prove to be a massive, quiet revolution,” Kumar says.
By predicting demand with greater precision, retailers can reduce waste, improve inventory placement and create more responsive supply chains.
“It eliminates waste and optimizes inventory placement before a shopper even orders.”
At the same time, Kumar believes some of today’s most visible AI applications are receiving more attention than they deserve.
“Generative AI product design tools will prove overhyped for immediate market impact.”
The reason is simple: Physical operations still move at a different pace than software innovation.
“A brilliant design means very little if your manufacturing infrastructure cannot deliver it to shelves.”
As organizations evaluate their AI roadmaps, Kumar advises leaders to prioritize operational capabilities over marketing excitement: “Leaders must stop chasing flashy AI-generated marketing concepts that lack operational depth.”
Instead, he recommends investing in foundational systems that enable responsiveness and execution.
“Capital should be aggressively funneled into data pipelines and automated micro-fulfillment networks to build a highly responsive supply chain.”
“From my seat, the loss of jobs is definitely hype that will become evident.”
Agents Become Teammates, Not Replacements
Anisha Manvatkar, Sr. Technical Program Leader at NVIDIA and author of Hello AI Transformation, has advised organizations across SAP and Capgemini while mentoring startups and technology leaders globally. Looking ahead, she expects agentic AI to become a routine part of organizational operations.
“A year from now, agents and sub-agents will be the norm,” Manvatkar says.
Rather than functioning as isolated tools, she believes AI systems will increasingly operate as embedded collaborators supporting employees across departments.
“Org structures will include them as routine members due to their ability to reason on their own.”
She also anticipates early advances in personalized medicine as agentic systems become capable of synthesizing increasingly complex healthcare information.
“Personal medicine will start to see early breakthroughs with the help of agentics.”
Despite persistent headlines about job displacement, Manvatkar sees a different outcome emerging.
“From my seat, the loss of jobs is definitely hype that will become evident.”
Instead, she expects AI to drive workforce evolution rather than workforce elimination: “Team productivity will increase with the help of AI as the roles evolve.”
Her outlook reflects a recurring theme throughout the Think Tank’s responses: AI’s greatest value may come from augmentation rather than replacement.
Trust in Digital Content Becomes Essential Infrastructure
Pradeep Kumar Muthukamatchi, Principal Cloud Architect at Microsoft, helps startups and enterprises deploy secure, scalable AI solutions globally. He also serves in advisory and standards-focused roles that shape the future of responsible AI adoption.
Muthukamatchi believes one of the most important AI developments over the next year will receive very little public attention.
“Digital provenance infrastructure—like C2PA cryptographic watermarking—will prove far more critical than people realize.”
As AI-generated content becomes increasingly difficult to distinguish from authentic media, organizations will need reliable methods for verifying provenance and authenticity.
“It is the invisible, vital plumbing required to preserve digital media trust and defend against deepfakes.”
At the same time, Muthukamatchi views fully autonomous enterprise agents as one of the market’s most overhyped narratives.
“Fully autonomous enterprise agents that promise to run entire departments without human oversight will prove to be mostly hype.”
He expects concerns around legal liability, governance and operational risk to slow widespread adoption.
“High error latency, edge-case failures and legal liabilities will stall their widespread adoption.”
His advice is to prioritize foundational investments: “Capital should be directed toward strengthening foundational data architecture, ensuring strict data integrity and securing fully compliant, legally cleared IP pipelines.”
Practical Automation Delivers the Greatest Value
Will Conaway, President of Tuxedo Cat Consulting and a recognized healthcare technology leader, has advised organizations on AI strategy, organizational transformation and operational improvement across multiple industries.
Conaway believes the most transformative AI advances will emerge from practical automation rather than attention-grabbing breakthroughs.
“AI’s ability to autonomously chain complex tasks will prove more transformative than most realize.”
He points to applications such as legal research and supply chain management as examples of where organizations will begin seeing measurable value.
“The real impact will come from practical, agentic AI quietly reshaping workflows and boosting productivity.”
Meanwhile, Conaway believes public fascination with AI-generated art and debates surrounding machine consciousness distract from more important developments. His recommendation is for leaders to focus on practical outcomes rather than novelty.
“Prioritize investments in applied AI solutions that streamline operations and solve real business problems.”
From Copilots to Execution Systems
Aditya Vikram Kashyap, Vice President of Firmwide Innovation at Morgan Stanley, focuses on enterprise innovation, AI governance and large-scale transformation initiatives across the financial services sector.
Kashyap argues that the next major shift in AI will be less about generating content and more about driving execution.
“The most underestimated AI shift is not bigger models, but context-aware systems connected to workflows, memory, enterprise data and decision chains.”
Organizations that successfully integrate AI into core business processes will gain a significant advantage.
“The real winners will be organizations embedding AI into execution across operations, compliance, healthcare and finance.”
Conversely, Kashyap believes the market continues to underestimate the complexity of replacing human judgment.
“The biggest hype risk is the belief that autonomous agents will reliably replace human judgment at scale anytime soon.”
For enterprise leaders, the solution is investing in infrastructure that supports transparency and accountability.
“Governance, observability, trust and orchestration are far harder problems than generating language.”
He advises leaders to invest in “auditability, data lineage, retrieval quality and human oversight.”
“Everyone’s racing to deploy agents, but the unglamorous work of measuring whether outputs are actually correct will quietly separate the winners from the losers.”
The Companies That Win Will Be the Ones That Measure
David Obasiolu, AI Security, Governance and Systems Consultant at Vliso AI, specializes in the intersection of artificial intelligence, cybersecurity and trustworthy systems.
Obasiolu believes one of the most important AI categories remains largely overlooked: “AI evaluation and verification tooling” will prove more important than most organizations realize. As companies race to deploy increasingly capable agents, he argues that the ability to measure performance accurately will become a critical competitive differentiator.
“Everyone’s racing to deploy agents, but the unglamorous work of measuring whether outputs are actually correct will quietly separate the winners from the losers.”
According to Obasiolu, trust—not capability—is the central challenge.
“Trust, not raw capability, is the real bottleneck.”
That is why he believes expectations around fully autonomous agents remain unrealistic.
“The demos dazzle, but reliability collapses the moment these systems meet messy, real-world tasks.”
For leaders, the path forward is clear.
“Shift focus away from moonshot autonomy and toward measurement, guardrails and human-in-the-loop workflows,” Obasiolu says. “The teams who can prove their AI works will out-execute those who merely promise it.”
Where Leaders Should Place Their Bets Now
- Build AI infrastructure before chasing autonomy. Governance, observability, memory and evaluation systems will determine long-term success.
- Prepare for dramatically lower AI costs. Business models based on current inference economics may quickly become outdated.
- Own your AI harness, not just your model. Data, evaluation frameworks and portability create sustainable advantages.
- Treat agent security as a strategic priority. AI-native trust and governance layers will become essential infrastructure.
- Invest in operational intelligence. Predictive forecasting and supply-chain optimization may deliver greater ROI than generative creativity.
- Expect agents to augment teams before replacing them. Productivity gains are likely to outpace workforce displacement.
- Prioritize digital provenance and trust. Verification and content authenticity will become increasingly important as synthetic media expands.
- Focus on applied automation. The greatest value will come from AI systems solving specific business problems.
- Embed AI into execution workflows. Context-aware systems connected to enterprise processes will outperform standalone copilots.
- Measure AI performance relentlessly. Evaluation and verification capabilities will separate successful deployments from failed experiments.
Beyond the Hype Cycle: What Really Matters Next
Across industries, the members of the Senior Executive AI Think Tank identify a clear pattern emerging beneath the noise surrounding AI. The technologies likely to create lasting competitive advantage are not necessarily the most visible. Governance, trust, observability, evaluation, provenance, security and workflow integration consistently emerge as the capabilities that will determine enterprise success.
For leaders making investment decisions today, the signal from these perspectives is hard to ignore: The real work is happening beneath the headlines. Competitive advantage is shifting toward the less visible layers—clean data pipelines, reliable evaluation, security controls, observability and workflows that hold up under pressure, not just in demos. Organizations that get serious about operational readiness now will be far better positioned to absorb whatever comes next in AI, regardless of which models or vendors end up leading the market.
