2025 was quite a year for artificial intelligence. We asked for predictions from the people working every day to take AI into the future where they see AI headed in 2026. The response was so great, we are breaking down in two parts.
Let’s dive into AI Insider’s Expert Predictions on Artificial Intelligence: Part One.
Jiahao Sun, CEO of FLock.io
In 2026, the focus will shift from massive, general purpose models to specialized, verifiable intelligence. It could well be the year of Auditable Intelligence, where the market demands proof of privacy and provenance.
Many highly specialized, decentralized AI models are already deployed in high stakes sectors like healthcare and finance with blockchain governance. These vertical solutions will succeed where centralized general purpose models fail, because they solve the critical issues of data sovereignty, regulatory compliance, and domain specific accuracy.
The market will reward the protocols that can deliver real world impact and verifiable results, making industry specific DeAI the most valuable sector in the coming year.
Open Governance and Composable AI
The market has gone from celebrating proprietary models to scrutinizing their lack of transparency. The core issue is that centralized AI is a closed system. Training a massive, general purpose AI without violating these legal boundaries is simply impossible, exposing the fundamental weakness of the approach.
The next wave of innovation will be defined by open source models and composable AI architectures. Developers and enterprises are demanding the ability to audit, customize, and combine AI components, which is impossible with closed off models.
The open source ecosystem is already proving that AI innovation is now pluralistic, and this trend will accelerate as regulatory pressure forces more transparency.”
AI Tokens and the Institutionalization of DeAI
The market is recognizing that the true value in AI is not in the centralized infrastructure, but in the decentralized application layer. The surge in AI token market capitalization is a clear signal that the future of intelligence cannot be built on the same centralized models that govern today’s data economy.
Investors will realize they were valuing the shovel, not the gold being mined, as the gold is in the decentralized application layer. This is the institutionalization of DeAI. The market is moving from hype to valuing protocols that offer utility and a clear economic model. Tokenizing AI utility is the only mechanism that can align the incentives of the model creator, the infrastructure provider, and the end user.
Blockchain as the Provenance Layer
As autonomous AI agents take on more critical roles, the need for an immutable record of their actions, training data, and provenance becomes very crucial. Blockchain is emerging as the essential trust and verification layer for all high stakes AI applications.
The challenge is conceptual – many people do not understand what technologies like federated learning is, and they certainly do not understand how blockchain makes it better.
The industry needs a new architectural standard where raw data is kept on local devices and only insights are sent to a secure blockchain. This is how we move AI governance from corporate policy to auditable code, ensuring that the immense potential of AI serves all of humanity, alongside prioritizing ethical integrity and innovation.
Miguel Alvarez, Chief Data and AI Officer at Orange Business
Redefining Intelligence Through Agentic AI
Next year we’ll see autonomous agents do far more than just communicate. They will start to take action, make choices, and expand their knowledge over time. Yet there will be increased questions concerning accountability when an agent makes an error. There may even be emotional attachment issues, such as over-reliance and dependency, as users let agents into more aspects of their lives. In 2026, both businesses and regulators will need to address the realities of agentic AI decision-making sooner than they had expected.
Users will start to feel that AI technology truly understands them without having to continuously retrain it, as agents recollect past encounters. Commercially, agents will have the power to purchase, sell, and bargain on behalf of people and businesses, including machine-to-machine transactions between various enterprises.
Regulatory frameworks keep getting delayed, so there’s more emphasis on organisations to make sure they’re proactively prioritising users’ protection. Businesses must educate the public on how these agents will evolve over time, moving beyond just solving problems to becoming the new competitive frontier for many industries. When people understand what agentic AI truly does, and the boundaries that accompany it, the opportunities to use AI will be endless.”
Next year Agentic AI will Become Every Customer’s Personal Operator
Companies have been racing to bolt generative AI onto every customer-facing process, but in 2026, CX will move beyond passive chatbots and scripted automation. We’ll see the rise of agentic AI: Autonomous, goal-oriented systems that don’t just answer questions but take action on a customer’s behalf, from rebooking flights to resolving issues in real time.
These intelligent agents will turn support into a seamless, invisible layer of service that anticipates rather than reacts. The future of CX isn’t conversational, but operational. It’s AI that does, not just talks. But while automation gets smarter, the human touch will remain the differentiator, because no matter how smart the system becomes, trust will always be built by people, not processes.”
Jean-Francois Colin, Head of Customer Experience, SVP at Orange Business
The New CX Frontier – Human-AI Partnership Drives Authentic Value
We’ve spent the past few years using AI to make things faster and cheaper, but in 2026, the real advantage will come from uniting humans and agentic AI as equal partners in delivering value. AI will take care of more complex, routine tasks, freeing people to do what machines can’t, such as making connections, empathising, and building trust. The best experiences will be designed around this collaboration, where virtual and human agents each play to their strengths to create interactions that feel both intelligent and deeply personal. Authenticity and empathy will become the new luxury of the customer experience. In the race to automate everything, the brands that remember the power of feeling seen, supported and truly understood will be the ones that customers go back to.
Michael Stewart, Managing Partner at M12, Microsoft’s Venture Fund.
- Massive capacity partnerships across the value chain were what surprised me most in 2025 – AI Labs, hyperscalers, neoclouds, memory and hardware providers, real estate and energy with deals $10B in size up to hundreds of billions in size. The entire IT industry is gearing up for a new phase of technology growth, in a weirdly similar, but uncoordinated way. Microsoft is one of the largest spenders but not by a huge margin, and partners are forming their own webs of partnerships.
- 2025 was shaped by two big shifts: the early-Q1 “Deepseek moment,” when confidence wavered around whether massive pre-training budgets were still a real moat against cheaper, lower-overhead open-source models, and then the huge hyperscaler capacity ramps in Q3/Q4, which are still going. Does this mean we no longer take the threat seriously—or that we’ve simply found better ways to guard against model “copying” through checkpoints or synthetic data? Or does it mean that with all the new capacity, we can now afford to run both high-overhead frontier models and lower-cost open-source models within the same economic envelope, choosing higher throughput when it matters and higher-quality, more compute-intensive inference when that’s the better tradeoff?
- Seems that we have more concerns about startup competition enabled by not just AI model capability, but the ease of vibe coding to copy successful apps that have gained meaningful traction in the business and enterprise markets. This is an expected result of the “model blast radius” whether or not AI Labs use their latest, greatest, improved models to move into apps, or if other people do it with the same models. It’s a sign of an efficient market if competition is vigorous and firms can enter or leave verticals at will to follow where others are finding quick success.
- Local AI is not being talked about, but it should be. Not just the model types that work well on local devices, but the rapid progress in improving local consumer hardware to run AI tasks. There is a lot of distance to make up, but local models are improving in “intelligence per Watt” by 3-4x per year while tooling that improves model inference efficiency proliferates and people start to want more use of AI on their own devices. Robotics is a key use case of local AI although it involves a lot of risk around the chassis or body that will be tested over the next few years. The question is: will the public accept and embrace the humanoid robot product or treat it more like AR/VR/XR which is still niche in its use despite huge investments by major tech companies?
Lior Shaltiel, PhD, CEO of NurExone Biologic Inc.
Advance of AI Applications in Bioprocessing
AI will play an increasingly important role in bioprocessing, including monitoring and optimizing the production of biological materials. It can help identify and characterize new molecules and enables analysis of massive datasets with rapid iteration. As drug delivery technologies advance, and as new nanoscale delivery systems capable of crossing the blood–brain barrier mature, AI may also assist in selecting and engineering cargo with the right biological signals for highly specialized or previously inaccessible targets.
Anurag Gurtu, CEO of Airrived
2026 Will Be Brutal for Legacy Tech. AI-First Platforms Will take the Throne
For more than a decade, enterprises have been drowning in point products, dozens of tools stacked on top of each other, each promising visibility, automation, or analytics. Yet none of them delivered what organizations actually needed: outcomes.
In 2026, that changes. AI is no longer an assistant bolted onto legacy systems. It’s becoming the operating system for cybersecurity, IT, and enterprise workflows. And the companies built on agentic architectures not cloud-first or workflow-first models are the ones rewriting the rules.
Below are the major forces that will define 2026, and why they signal the most significant shift since the rise of cloud computing.
Agentic Platforms Become the New Digital Workforce
The most important shift of 2026 is the rise of agentic platforms where systems composed of autonomous and semi-autonomous agents collaborate, learn from feedback, and deliver operational outcomes.
Unlike chatbots or workflow engines, agentic systems operate more like digital workers:
- They understand enterprise context.
- They coordinate across identity, endpoint, network, and cloud.
- They adapt from human reinforcement.
- They reduce manual workloads by orders of magnitude.
For enterprises, this means the first real path to scalable digital labor. For vendors, it means the next $50–100B category is already emerging.
By the end of 2026, agentic platforms will become the most sought-after acquisition target for hyperscalers and cybersecurity giants, the same way cloud platforms were a decade ago.
Legacy Vendors Lose Ground In an AI-First World
Legacy enterprise software companies built around rules engines, ticketing systems, or static analytics will face their toughest year yet. The problem isn’t that they lack AI features. It’s that their architectures were never designed to support reasoning systems that operate continuously and autonomously.
AI-first companies move faster, deploy faster, and deliver outcomes traditional vendors simply can’t match. Their systems:
- require fewer integrations,
- learn from operational behavior,
- and collapse dozens of workflows into a single agentic mesh.
In 2026, customers will start questioning why they need four to five categories of tooling when a single AI-native system can do the work. Expect several well-known cybersecurity and IT vendors to see revenue declines, restructuring, or to become acquisition targets themselves.
Token Costs Collapse And AI Becomes Nearly Free To Use
The economics of AI shift dramatically in 2026. Token costs the basic unit of LLM inference will drop by 70–90% as specialized AI hardware, quantization techniques, and MoE-based architectures mature. This reduction is not incremental it’s transformative. Once inference becomes cheap, always-on agents become feasible.
Security operations, IT workflows, compliance cycles, and monitoring systems will all run continuous, AI-driven processes that were previously cost-prohibitive. Cheap tokens unlock a new era: AI not as a feature, but as infrastructure.
China Introduces LLMs That Shift the Global Balance
Perhaps the most geopolitically important trend of 2026 will be the rapid rise of Chinese LLMs that rival and in some cases surpass the performance of leading Western models at a fraction of the training and inference cost.
Chinese labs have leaned heavily into:
- low-precision training,
- lightweight architectures,
- massive multilingual corpora,
- and integrated agentic reasoning modules.
These models will expand quickly across Asia, Africa, LATAM, and emerging economies, forcing Western enterprises and regulators to confront a new competitive reality: AI supremacy is no longer regional. The global cost curve for AI will fall even faster, and the competitive pressure this creates will be profound.
AI Talent Wars Escalate to Unprecedented Levels
The war for AI talent will intense in 2023–2025 and reaches a new peak in 2026. Microsoft, Google, OpenAI, Anthropic, Amazon, Meta, and a handful of elite AI-first startups will offer compensation packages that break previous benchmarks.
Top-tier LLM researchers, agentic system architects, and reinforcement learning experts will routinely receive $10M–$20M+ packages. Acqui-hires will become one of the fastest ways for companies to acquire scarce skills.
This has a secondary effect where legacy vendors simply won’t be able to keep up. Without competitive talent, they’ll struggle to innovate and will fall even further behind AI-first competitors.
The Collapse of Point-Product Cybersecurity
Point products have officially reached their saturation point. Enterprises are running 70–120 tools across security and IT, yet still struggle with alert fatigue, analyst burnout, and fragmented data.
In 2026, the market finally acknowledges what insiders have known for years: no single-feature product can compete with an AI system that can read, reason, act, and improve over time.
AI agents are now capable of executing complex workflows end-to-end, from threat detection to investigation, response, and audit. Instead of stitching together dashboards, enterprises are shifting to horizontal, agentic platforms that break down silos. This is the first year CISOs begin sunsetting tools rather than adding more.
Expect the valuation gap between AI-first platforms and legacy cybersecurity vendors to widen sharply. A wave of consolidation will follow.
Cybersecurity and AI Fully Converge
The most meaningful shift of 2026 is cultural: cybersecurity and AI cease to be separate domains. SOCs won’t just use AI — they will operate with AI. Agentic systems will:
- suppress alerts automatically,
- run investigations in seconds,
- correlate exposures across cloud, identity, endpoint, and network,
- generate remediations,
- validate changes,
- and maintain continuous controls.
By the end of 2026, large enterprises will see 30% or more of SOC workflows executed by agents, not humans. This is the year AI transitions from a co-pilot to a co-worker.
The Bottom Line: 2026 Belongs to AI-First Companies
The market is entering a once-in-a-generation transition. The companies that win will be the ones that rebuild for an agentic, AI-native world not those that bolt AI onto legacy workflows. AI is the new architecture, agents are the new workforce and 2026 is the year enterprises finally realize the old stack cannot compete with what comes next.
