Report “AI Agent Trends To Watch in 2025” — CB Insights

The Age of Autonomous Agents Has Begun

2025 marks a strategic turning point in the recent history of artificial intelligence. Following the first generation of generative “copilots,” AI agents are taking over: systems capable of carrying out complex tasks autonomously, with minimal human intervention. These agents — powered by advanced LLMs — are infiltrating business use cases, from sales to regulatory compliance. CB Insights, in its report “AI Agent Trends to Watch in 2025,” provides a dense and instructive overview of this shift. It is no longer just about experimenting, but about integrating. And the front lines are clear: digital giants want to control everything, while specialized start-ups battle to carve out a place in high-value-added niches.

 

  1. The Big Tech Steamroller

From the very first lines, CB Insights sets the tone: mentions of AI agents quadrupled between Q3 and Q4 2024. And 2025 confirms the rising power of “general purpose” solutions driven by OpenAI, Google, Anthropic, and Amazon. Thanks to model costs being divided by ten every twelve months and massive distribution (ChatGPT now has 400 million weekly users), these players capture most of the value.

The strategy is clear: crush the market for versatile assistants by leveraging network effects, computing power, and the appeal of major brands. Uber, Klarna, and Lyft have already deployed their AI customer service agents in direct collaboration with OpenAI or Anthropic. These moves leave little room for start-ups that cannot compete in terms of infrastructure, brand recognition, or functional coverage.

But behind this apparent dominance, CB Insights highlights a paradoxical effect: standardization is pushing needs toward greater specialization.

 

  1. The Specialists’ Revenge: Toward Vertical Agents

While versatile agents are dominated by Big Tech, the field of specialization remains up for grabs. Nearly 50% of fundraising in AI agents still targets horizontal applications such as customer support, software development, or sales. Players like Sierra ($20M in revenue) are betting on tight integration with brand policies, internal knowledge bases, and ticketing systems.

The case of Cursor, which understands millions of lines of code to automate development processes, illustrates this dynamic well: the agent becomes an expert colleague, rooted in the business. Meanwhile, Glean and LlamaIndex are building bridges between knowledge management platforms and adaptive agents.

But the real frontier of innovation lies elsewhere. It is the verticals — healthcare, finance, legal, industry — that concentrate the hopes for differentiation. Hippocratic AI tailors its agents to the traceability requirements of the medical sector. Norm AI, backed by Citi Ventures and New York Life, targets financial compliance workflows with advanced explainability tools. These players are betting on an alliance between sector expertise and generative AI.

 

  1. Infrastructure: The New Silent Eldorado

Behind the visible agents lies an invisible but crucial mechanism: infrastructure. CB Insights observes a gradual structuring of the technology stack, currently fragmented. Three sub-categories are emerging:

  • Data curation: Solutions like LlamaIndex or Unstructured make it possible to convert companies’ unstructured data into usable contexts for agents.
  • Access to tools and the web: Platforms like Browserbase offer AI-controllable headless browsers to automate web interactions.
  • Evaluation and observability: Langfuse, Haize Labs, and Coval are developing monitoring tools to make agents more reliable and anticipate errors.

This “agent enablement” market is becoming one of the most dynamic, according to CB Insights. It addresses a strong demand from companies: to create robust agents capable of interacting with complex systems while ensuring secure human oversight. Orby AI embodies this “human-in-the-loop” trend, which secures autonomy through intelligent supervision.

“Full-stack” platforms — offering no-code or low-code solutions — are also attracting a clientele of companies seeking rapid deployment. Low-code is thus becoming a lever for democratizing AI agents.

 

  1. From POCs to Implementation: The Era of “Digital Workers”

63% of companies surveyed by CB Insights report placing “high importance” on AI agents in their plans for the next 12 months. The experimentation phase is coming to an end. Established vendors like Twilio are already integrating agents into their tools, to filter leads or respond to customers before human intervention.

But the transition to mass implementation faces three major obstacles:

  • Reliability and security: 47% of decision-makers express concerns about privacy, accuracy of responses, and customer-facing uses.
  • Technical integration: 41% point to difficulties connecting to existing systems.
  • Talent shortage: 35% highlight the insufficient number of hybrid profiles (AI + organizational transformation).

Here again, solutions are emerging: from co-creation with supervised agents to building solid internal knowledge bases. The challenge is not only technical but cultural: we must rethink the way employees work with “agent colleagues.”

 

Toward a New Economy of Augmented Work

This CB Insights report sheds light on a reality already perceptible on the ground: the AI agent is no longer a laboratory fantasy, but an actor in organizational transformation. Its potential goes beyond mere automation to achieve a form of instrumental, specialized, and connected intelligence.

The strategic choice is not limited to “whether or not to adopt” an agent, but rather “which agent, with what granularity, for which value chain.” Large enterprises will tend to integrate generalist agents via their historical technology partners. SMEs, regulated sectors, or knowledge-intensive industries will opt for vertical agents tailored to their constraints.

This evolution requires decision-makers to rethink their technology integration frameworks. Building a compatible data architecture, training their teams to interact with autonomous agents, developing appropriate audit and security policies, but also rethinking human roles around these agents.

 

Some Avenues for Reflection for Companies

  • How should the internal governance of agents be organized (assignment, supervision, audit)?
  • Which jobs will be transformed by seamless collaboration between humans and AI agents?
  • Can the AI agent become a new form of interface for customers or employees?
  • What specific performance indicators should be put in place to evaluate an agent?
  • Should investments be pooled with other players in the same sector to co-develop sovereign vertical agents?
  • What role for regulators in supervising the behavior of AI agents in critical contexts (healthcare, finance, justice)?

 

Main Reference
CB Insights. (2025). AI Agent Trends to Watch in 2025.
https://app.cbinsights.com/research/enterprise-ai-agents-copilots-market-map/