State of AI in 2025: key trends shaping artificial intelligence

AI SALES
ARTIFICIAL INTELLIGENCE
Featured - State of AI in 2025 - Key Trends Shaping Artificial Intelligence

The state of AI in 2025 tells a story of technology reaching a turning point. Two years after ChatGPT sparked widespread interest in generative AI, we’re seeing a shift from testing things out to actually putting AI to work.  

 

Organisations worldwide are deploying AI systems, governments are creating rules around it, and new capabilities are emerging that genuinely change how businesses operate. 

Agentic AI systems are moving beyond the hype  

The defining characteristic of the state of AI in 2025 is the emergence of agentic systems. Earlier AI tools could write an email or answer a question. Agentic AI can plan out multi-step projects, execute tasks independently, and adjust when things don’t go as expected. 

  • Early adoption shows promise but limited scale: McKinsey & Company data showed 23% of organisations are scaling agentic AI systems within at least one business function, while an additional 39% have begun experimenting with AI agents. These deployments concentrate primarily in IT departments and knowledge management teams, where specific jobs like automating help desk tickets and conducting research have matured quickly. 
  • Most implementations remain narrow: Most organisations scaling agents are doing so in only one or two functions, and in any given business function, no more than 10% of respondents (McKinsey) say their organisations are scaling AI agents. Organisations are proving the technology works in controlled environments but struggling to expand beyond initial use cases. 

The difference between traditional and agentic AI matters. Traditional AI waits for you to ask it something. Agentic AI takes initiative, coordinates across different systems, handles problems autonomously, and completes objectives without constant supervision.  

The technology is proven. Getting it rolled out everywhere remains the challenge. 

Enterprise adoption is accelerating but remains patchy  

AI adoption among businesses has more than doubled in two years, signalling a shift from experimental technology to operational reality. Yet this growth masks significant disparities in how different industries and organisations are actually implementing AI. 

 

  • Overall adoption doubles: Anthropic data showed AI adoption among US firms growing more than 2x in the past 24 months, rising from 3.7% in 2023 to 9.7% in mid-2025. This represents substantial progress, though the majority of firms still aren’t using AI in their production processes. 
  • Sector differences are dramatic: Early August 2025 US Census data showed 25% of businesses in the Information sector using AI, which is roughly 10x compared to Accommodation and Food Services. Tech companies and businesses with digital workflows are racing ahead, while industries dealing with physical operations struggle to find their footing. 
  • Pilot-to-production gap narrows slowly: According to a 2025 ISG report, only 31% of AI use cases it studied in 2024 reached full production. Progress is real, but most AI projects remain stuck in the testing phase, unable to scale beyond initial experiments. 
  • Strategy makes the difference: WRITER data revealed 80% AI adoption success rates among companies with formal strategies, compared to only 37% for those that don’t have them. Technology capability isn’t the bottleneck, but knowing what you’re doing with it is. 

The numbers reveal something crucial: technology isn’t holding companies back. The challenge lies in organisational readiness, strategic clarity, and the ability to redesign workflows around AI capabilities. 

Poor implementation is creating organisational chaos  

Whilst adoption grows, many companies are discovering that adding AI to their operations without proper planning creates more problems than it solves. The human cost of getting AI wrong is emerging as a significant barrier to successful deployment. 

 

  • Internal conflicts surge: C-suite executives report that AI adoption has been causing friction in their teams, or developed in private away from everyone else (WRITER). Power struggles between departments, conflicting priorities, and teams working against each other are undermining what should be transformative initiatives. 
  • IT-business alignment fails: Tensions between IT departments and other business areas are common, with most organisations struggling to coordinate AI development across functions. This results in duplicated efforts, incompatible systems, and wasted resources. 

The message is clear: organisations that rush into AI without addressing change management, governance, and cross-functional collaboration face significant risk of internal disruption that can outweigh any technological benefits. 

Technical capabilities are expanding rapidly  

AI systems in 2025 can handle tasks that were impractical or impossible just months ago. These advances aren’t just incremental improvements – they represent fundamental expansions in what AI can do. 

 

  • Advanced reasoning emerges: Models with advanced reasoning capabilities can solve complex problems with logical steps that are similar to how humans think before responding to difficult questions, with applications in fields like science, coding, mathematics, law, and medicine. These systems can review contracts side by side, write working code, and handle complicated workflows that used to require human judgment at every step. 
  • Multimodal processing becomes reliable: AI systems now work effectively with text, images, video, and audio simultaneously. This enables applications that truly understand context regardless of input format, from diagnostic tools analysing visual and audio symptoms to customer service systems that grasp nuance across channels. 
  • Compute demands increase: AI reasoning moves beyond basic understanding and into advanced learning and decision-making, which requires additional compute for pre-training, post-training, and inference. As capabilities mature, the technical infrastructure needed to support them grows substantially. 

These technical foundations keep strengthening, expanding the range of problems AI can tackle whilst simultaneously increasing the resources required to deploy it effectively. 

Economic dynamics shift as costs plummet  

The economics of AI are changing dramatically. Whilst training large models remains expensive, the cost of actually using them is falling at rates that unlock entirely new applications. 

 

  • Inference costs collapse: The cost per million tokens of some models has decreased from $20 to $0.07 between 2024 and 2025  (Search Engine Journal) Applications that were economically unfeasible in 2024 now make perfect business sense. 
  • Energy consumption rises despite efficiency gains: Despite gains in energy efficiency, overall power consumption is up. MIT Technology Review reported data centers being 48% more carbon intensive compared to average US values. Companies are getting more efficient with each AI query, but they’re running so many more queries that total energy use keeps climbing. 
  • Investment doubles: According to IDC, spending on AI by 2028 will reach $632 billion – more than double $307 billion in 2025. Capital is flowing to infrastructure, model development, and deployment at unprecedented levels. 

The economic equation is becoming clearer. Deploying AI is getting cheaper, enabling broader adoption, but the aggregate environmental and infrastructure costs are rising as usage scales. Organisations can now deploy AI across more use cases without costs spiralling, creating real tension between scaling capabilities and meeting sustainability commitments. 

Regulatory approaches diverge globally  

The global regulatory landscape for AI fractured significantly in 2025. Major economies are pursuing fundamentally different strategies, creating complexity for any organisation operating across borders. 

 

  • Europe enforces comprehensive regulation: Non-compliance with the rules will lead to fines of up to 7% of a company’s yearly turnover, which is around €35 million. This is dependent on the infringement and the company’s size. The EU’s AI Act implements strict requirements, though confusion about which systems fall under banned categories has caused many international companies to move cautiously with European deployments. 
  • United States pivots to deregulation: In January 2025, a dramatic shift in federal AI policy occurred with the transition from the Biden administration to the Trump administration. President Trump rescinded Biden’s order and replacing it with a new order prioritising AI innovation and US competitiveness. The new approach focuses on clearing away regulations that might slow development, with federal funding for state projects now depending partly on whether states avoid adding their own regulations. 
  • China implements content labelling: In March 2025, the Cyberspace Administration of China issued final “Measures for Labelling AI-Generated Content”, which took effect on September 1, 2025. This compelled all online services that create or distribute AI-generated content to clearly label such content. It’s a characteristically centralised approach with clear rules and strict enforcement. 
  • Global coordination remains elusive: The OECD tracks more than 2,083 AI governance initiatives worldwide, yet harmonised international frameworks are nowhere in sight. 

This regulatory patchwork creates significant compliance challenges for global organisations, who must navigate contradictory requirements across different jurisdictions whilst maintaining competitive advantage. 

Skills gaps are constraining deployment  

Companies are creating leadership roles and investing in AI capabilities, but they’re struggling to build the expertise needed to deploy AI effectively. The skills challenge extends far beyond technical roles. 

 

  • Leadership roles proliferate: Executive leadership in Gen AI adoption has resulted in 61% of companies having Chief AI Officer roles. Organisations are establishing dedicated positions to drive AI initiatives, signalling serious commitment. 
  • Capabilities lag behind ambition: Capability building is falling short of ambition, with enterprises reporting gaps between their AI aspirations and their ability to execute effectively. According to Deloitte research, workforce skills and readiness rank among the top barriers to AI adoption, with only 20% of executives feeling ready to make the switch. 
  • Multiple skill types needed: Organisations need technical specialists who can build systems, domain experts who understand where AI creates value, change managers who can navigate resistance, and strategic leaders who can balance innovation with risk management. 

The technology is available and increasingly affordable. The human capacity to deploy it strategically—understanding where it adds value, how to integrate it into workflows, and how to manage the organisational change it requires—remains the limiting factor. 

Workers see AI as augmentation, not replacement  

Public attitudes toward AI reveal a nuanced picture. People recognise that AI will change their work, but most aren’t expecting it to eliminate their jobs entirely. 

 

  • Expectations vary significantly: In McKinsey’s research, 32% of respondents anticipate cuts in the workforce in 2026, 43% expect no change and 13% actually anticipate an increase in jobs. Larger organisations are more likely to expect reductions, whilst high performers predict meaningful change in either direction. 

This relatively optimistic outlook suggests that organisations have an opportunity to position AI as an augmentation tool that enhances human capabilities rather than as a replacement technology that threatens livelihoods. 

Where we are right now  

A sales rep using an AI sales bot while talking to a client

The state of AI in 2025 comes down to this: the technology keeps getting better while organisations struggle to use it effectively. Technical performance improves, costs fall, and new use cases emerge constantly. But for most companies, there’s still a massive gap between running pilots and deploying AI at scale. 

 

The organisations succeeding share these characteristics: formal strategies aligned with business objectives, visible leadership commitment from executives who use the tools themselves, governance frameworks for managing risk and validating outputs, willingness to fundamentally restructure workflows, and substantial resource commitments that enable deployment beyond pilots. 

How this applies to sales execution  

As enterprises navigate these trends, specialised platforms are emerging to address specific business challenges. AI sales execution platforms represent one practical application of these technological advances, helping sales teams apply agentic capabilities, multimodal understanding, and advanced reasoning to actual customer conversations. 

 

For organisations evaluating where to invest in AI, the critical question isn’t whether to adopt. It’s figuring out where AI creates the most value for your specific workflows and objectives. The technology is evolving rapidly. But successful implementation still requires clear strategy, strong governance, the right skills, and focus on measurable outcomes. 

 

Book a demo to see how Captivate’s AI sales execution platform helps teams apply these advancements to improve sales performance. 

Frequently Asked Questions

Common questions about this topic

What distinguishes agentic AI from earlier AI systems? 

Agentic AI can independently plan and execute multi-step workflows, adapting based on results with minimal human oversight. Unlike traditional chatbots that just respond to single questions, agentic systems can break down complex tasks, coordinate across multiple tools, and handle problems autonomously. Currently, 23% of organisations are scaling agentic AI in at least one business function, though most deployments remain concentrated in IT and knowledge management where specific use cases have matured. 

AI adoption varies dramatically by sector because of differences in digital readiness and data centralisation. The Information sector shows 10 times higher adoption rates than Accommodation and Food Services because digitised workflows and centralised data enable faster implementation. Organisations with dispersed physical operations or fragmented data face much greater challenges integrating AI effectively, regardless of how much money they invest or how good their strategy looks on paper. 

High performers fundamentally redesign workflows rather than simply adding AI to existing processes—they’re nearly three times more likely to do this than other organisations. They also benefit from visible senior leadership commitment, with executives who role model AI use and demonstrate real ownership of initiatives. Additionally, high performers invest substantially more resources, define clear processes for when human validation is needed, and treat AI as a transformation catalyst rather than just another efficiency tool.