Emerging Tech: Agentic AI Innovation Will Foster Autonomous Business
18 July 2025 - ID G00836945 - 13 min read
By Danielle Casey, Eric Goodness, and 5 more
Upcoming shifts in the agentic AI market will significantly change agentic products and use cases. Product leaders must future-proof for success by investing in emerging technologies that are key to unlocking next-level agency applications.
Overview
Key Findings
- Chief product officers must plan for the agentic AI market to undergo massive shifts, including market consolidation, scaled adoption of agentic AI and a stalling of innovation at the expert agency level. This requires them to align their investment plans, capability roadmaps and use-case strategy accordingly.
- Expert agents — systems optimized for complex, domain-specialized workflow support — will proliferate in three to six years, marking a massive shift in the types of tasks being automated.
- The progression of agentic AI from digital to physical actuation will inevitably unlock transformative opportunities where agentic AI operates across digital and physical environments, such as physical assets and critical infrastructure.
Recommendations
- Develop a “land and expand” AI agent strategy by building role-, industry- or use-case-optimized agents that enable the scalability of application complexity and the deployment footprint. Win the AI agent race by designing solutions that solve business problems and by delivering value that can be extended through existing workflows and across additional roles.
- Develop a product roadmap that plans for a proliferation of diverse, specialized agents in three to six years by investing in domain-specialized language models (DSLMs), multimodal capabilities and industry tools use. These emerging technologies will enable the offering of specialized, better-performing expert agents.
- Future-proof your use-case roadmap by identifying how your current use cases can extend from digital processing to include physical actuation.
Strategic Planning Assumptions
Through 2027, a majority of agentic AI deployments will use conditional autonomy with human-in-the-loop oversight and review.
By 2028, only 15% of agentic AI deployments will be highly autonomous, with expert agents up from less than 5% in 2025.
By 2030, 30% of organizations will be using AI agents in three or more business units.
By 2033, 15% of agentic AI deployments will be fully autonomous compared with less than 1% in 2025.
Analysis
Agentic AI Market Evolution
As agentic AI evolves from “intermediate” to “advanced” to “expert” and, finally, “superior” agency capabilities, so too will the applications of agentic AI. This means that future agency stages represent both new technical achievements as well as use-case opportunities. Product leaders must be prepared for these upcoming agency stages and the new capabilities and automation it will unlock, as well as new challenges that must be navigated.
Technology description: Agentic AI refers to a class of systems developed using various architectures, design patterns and frameworks, encompassing both single-agent and multiagent designs. These systems are capable of performing unsupervised tasks, making decisions and executing processes. They range from semiautonomous to fully autonomous, are software entities that utilize AI techniques to perceive their environments, make decisions, take actions, adapt and achieve specific goals in both digital and physical settings. See Emerging Tech: The Key Defining Characteristics of Agentic AI for more details.
This research considers the market implications, use-case evolution and enabling technology associated with the AI agency’s maturation over the next decade. This document builds on Emerging Tech: Agentic AI Maturity Roadmap, which outlines agentic AI’s evolution and agency capability progression.
This research will evaluate the following:
- How agentic AI will automate simple and complex tasks both now and in the far future (see Figure 1)
- The key, foundational technologies that will enable agentic AI’s capability progression (see Table 1)
Product leaders can use the agentic AI roadmap to:
- Differentiate and future-proof for success by investing in emerging tech that will enable future agentic functionality.
- Develop a use-case strategy that capitalizes on the new task automation that agentic AI advancements will unlock.
The agency maturity roadmap will trigger several major market moments that product leaders must prepare for as it progresses along the agency spectrum to Level 5 (L5) agency. These moments are illustrated in Figure 1.
The agency maturity roadmap in Figure 1 reveals key timeline insights aligned with the agency levels (L1 through L5, novice to superior agency):
Now, 2025 — intermediate agency: A majority of agentic products in the market today exhibit L2 intermediate capabilities. The key hurdle to overcome in moving to “advanced agency” is customer trust. According to Emerging Tech: Customer Trust Is a Critical Barrier to Agentic AI Adoption, a top go-to-market challenge cited by vendors is customer trust of agentic solutions. Massive market hype, agent-washing and hypercompetition further compound this problem by creating customer confusion.
One to three years — advanced agency: The market will concentrate at L3 — advanced AI — agents in the very near term, as vendors work on building trusted, reliable offerings that organizations are willing to adopt at scale. Gartner anticipates massive market consolidation in this time frame, as agentic offerings either prove out or fail.
Early majority adoption of agentic AI will be achieved near the end of L3 and the beginning of L4 agency.
Three to eight years — expert agency: Expert agents emerge and are characterized as being highly autonomous and deeply specialized. Model innovation will unlock a new class of agents and use cases. There will be a diverse proliferation of specialized agent offerings ranging from dietitian agents, to shopping agents, to construction design agents, to security agents.
However, this will also introduce a slew of new challenges. The scalable use of expert agents will also introduce technology, business and regulatory challenges that the market will need to navigate.
Moreover, there is a massive technology step-gap to progress beyond L4 — expert agency. The market will level off at the expert-level agency for several years as agents proliferate and the underlying technology continues to advance.
Eight to 10 years — agent ecosystems: L5 agency will be difficult to attain, as many of the key, foundational technologies to support L5 capabilities remain theoretical or experimental. Key model and orchestration breakthroughs are needed to unlock the reasoning, planning, adaptability and physical actuation capabilities of L5. Moving these innovations out of R&D and into production will be a telling indicator of the emergence of agent ecosystems.
Product leaders should use this research as guidance in developing their agentic AI product strategy, R&D investment plans and use-case roadmap. This will prepare them for success in a rapidly changing future (see Figure 2).
Critical Insight: Future Agents Will Be Specialized and Highly Autonomous, and Execute Across Ecosystems
Future agentic systems will undergo a seismic shift in task automation, where they progress from being optimized for simpler, routine tasks (Levels 1, 2 and 3) to being optimized for increasingly complex, expert tasks in dynamic environments (Levels 4 and 5). This advancement in task and workflow support is illustrated in Figure 3.
L3 agency represents an inflection point where agents transition from primarily automating simple, routine tasks to complex, skilled tasks.
L1, AI assistant — Assistants offer limited task automation support and are primarily used for question-and-answer support.
L2, AI agent — AI agents advance automation support for simple and routine tasks and workflows quite significantly, for example, scheduling meetings, processing orders and updating customer accounts. Up to 75% of these tasks can be automated with AI agents. At this level, agents start offering limited support for more complex tasks.
L3, advanced AI agent — This is where complex tasks and workflows start to be notably supported, with up to 50% of traditionally “skilled” tasks capable of being supported with agentic AI. These tasks can be multi-intent and multistep, and generally require analysis and judgment, for example, fulfilling prescription refills, conducting limited stock trading and monitoring predictive maintenance.
L4, expert agents — The key distinction at L4 is the depth of expertise required for task performance. For this reason, expert agents will unlock a lot of use cases and automation previously unachievable. Here, agents will operate in increasingly complex, dynamic and multimodal environments spanning the digital and physical realm. It is likely that a whole slew of new agent offerings will enter the market at the L4 level, such as legal agents, clinician assistant agents, factory operations agents, financial trading agents and personal buying agents.
L5, agent ecosystems —The most defining characteristic of L5 agents is the ability to negotiate and transact with third-party systems and agents representing outside entities and undefined stakeholders. This function will fundamentally change the nature and scale of agent applications and the actors involved. L5 tasks are not only infinitely more complex but will also introduce a slew of new challenges.
Table 1 identifies some of the enabling technologies that will advance AI offerings up the agency spectrum. For detailed profiles on many of these technologies, see the Emerging Tech Impact Radar: Generative AI.
Many emerging technologies are key to the rise of L4 and L5 agency — expert agents and ecosystem agents. As illustrated in Table 1, a lot of technology innovation is required for L4 — expert agents. There will also be significant variability in the types of agents and the tasks they support (such as lawyer versus clinician agents). The step-gap to move beyond L4 and to L5 agency is massive, as many of the enabling technologies for L5 agency remain theoretical or experimental. Consequently, the agentic AI market will struggle to move beyond expert AI agents and will continue innovating at L4 for at least five years before making breakthroughs to L5, ecosystem agents.
Table 1 outlines some of the key technologies that will enable future AI agent stages. Various emerging technologies are placed under the agency category that they are key to enabling. Though these technologies may emerge sooner, they will likely reach early majority customer adoption with the associated agency stage. For example, DSLMs are key to enabling expert agents, and both the enabling technology and application will accelerate along similar timelines.
Key Emerging Technologies Enabling Agentic AI’s Progression
Enabling Tech | L1, AI Assistant | L2, AI Agent | L3, Advanced AI Agents | L4, Expert Agents | L5, Ecosystem Agents |
Models | Natural language processing/understanding models LLMs | Customized LLMs (fine-tuning, retraining) Machine learning models Neuro-symbolic AI Reinforcement learning | Reasoning models Simulation twins | DSLMs Small language models (SLMs) Unsupervised learning | Active inference Causal AI Self-supervised learning |
Data | Retrieval-augmented generation (RAG) | Generative adversarial networks Knowledge graphs | Synthetic data (augmentation of existing data) GenAI engineering tools Multimodal AI Vector databases Advanced RAG techniques | Hypersynthetic data (new, specialized data) Diffusion AI models Multimodal AI (advanced/niche data modalities) “In situ” data processing | Simulation training |
Security | AI runtime inspection | Memory leak and manipulation detection Reasoning and decision logging | Tool calling inspection (Model Context Protocol [MCP] gateway) | Machine identity and access management (IAM) | Interagent protocol inspection Model red teaming |
Orchestration | Operates in silos | Intra-agent orchestrator (that is, communication among agents from one provider such as Orchestrator Agent) Tool calling (Example: MCP) | Interagent protocols (that is, between agents from various vendors such as A2A) Deep, domain-specific integrations | Intersystem communications (that is, an abstraction layer between first- and third-party systems) | |
Ecosystem | Agent marketplaces | Data marketplaces | Model marketplaces | ||
Source: Gartner (July 2025)
Near-Term Implications for Product Leaders
Models — The agentic AI market is starting to innovate on the model level in an attempt to address the accuracy and performance challenges noncustomized LLMs present. Currently, this is primarily achieved via model fine-tuning. However, advancements in reasoning models, open-source models, and DSLMs and SLMs will help agentic AI providers better support advanced and expert agentic AI use cases.
There are already emerging innovators in this space (see Emerging Tech: Techscape for Startups Using and Enabling Domain-Specific Language Models for more details). Model advancements will be key to both technology differentiation and business value support. Model innovation is strongly linked to agentic reasoning, planning, autonomy, sensing and proactively, and is a key indicator of an agent’s performance threshold. Model innovation will improve agents’ ability to adapt and respond to unpredictable variables, proactively navigate uncertain and complex environments, and enable physical actuation. The shift from digital to physical actuation will unlock entirely new use cases where assets and physical systems are monitored and managed using AI agents. Consequently, model innovation will likely play an important role in determining the winners and losers in the upcoming market consolidation, where current AI agents progress to advanced and expert support.
Comparatively, the model advancements for ecosystem agents (such as active inference and causal AI) remain in very early R&D stages.
Data — How data is organized and how agents interact with data will change in the coming years. The use of vector databases allows agents to make connections and see patterns in data otherwise obscured. Also, as agents become more expert and specialized, so will their ability to interact with specific or niche data types. This may require the use of synthetic data generation to optimize the models to the data modalities of the targeted use case, industry or business function. At L4 and L5 agencies, simulation use cases are unlocked, requiring the use of hypersynthetic data and simulation training.
More fundamentally, though, future agents will not be fed AI-ready data. Agents will be designed to dynamically interact with data when and where it sits.
Security — Security will play an increasingly important role as agents grow in autonomy. For example, inspecting tool calls and agent protocols for authentication. Machine IAM will be key in managing an agent’s credentials, authority hierarchies and authorizations for fraud prevention. Security will become increasingly important as agents become more autonomous and collaborative. The complexities that L4 and L5 agencies will introduce around physical assets and third-party networks will prove particularly challenging.
Orchestration — Orchestration is a key foundational technology that still has a ways to go. Advancements in agent orchestration are what will enable agents to progress from more communicative to collaborative interactions, as well as shift deployment environments from applications to ecosystems. The emergence of agentic ecosystems is conditional on the development of more advanced orchestrators for cross-system negotiation and authorized transactions. Do not underestimate the importance of orchestration. These advancements will play a key role in the shift from agent communication to true multiagent collaboration across environments. Effective orchestration will enable scalable value realization from agentic systems.
Ecosystem — Supporting ecosystems will also radically change in the coming years. Currently, vendors are developing agent marketplaces to accelerate time to value and agent innovation. As domain specialization becomes more important, data marketplaces and model marketplaces will grow in popularity.
Recommended Actions for the Next Six to 18 Months
- Develop a land-and-expand use-case strategy capable of scaling in application complexity and deployment footprint by building role-, industry-, or use-case-optimized agents. Winners in the AI agent race will design solutions that solve business problems and meet current market demand, while enabling scalability of workflow automation complexity to extend the value chain across additional roles and task types.
- Develop a product roadmap that plans for a proliferation of diverse, specialized agents in around three to six years by investing in DSLMs, multimodal capabilities and industry tools use. These emerging technologies will enable the offering of performant expert agents.
- Future-proof your use-case roadmap by identifying how your current use cases can extend from digital processing to include physical actuation.
Gartner’s agentic AI case-based research project ran from December 2024 through March 2025. As part of this research, 24 vendors participated in two interviews each. The first interview focused on the agentic products, technology capabilities, R&D roadmap and adoption challenges, among other things. The second interview focused on the case studies to include the business problem addressed, the processes that were changed, the outcomes achieved and the unmet needs. This document was informed by insights from both interviews.