Trend

Single LLM models often fall short in handling general tasks effectively, whereas specialized agents excel at performing micro-tasks.

Individual workflows and goal achievement will be augmented and scaled by using AI multi-agents, effectively transforming each person into a virtual manager. This approach reduces routine tasks and increases productivity across organizations. 

Rising labor costs and the demand for better ROI make agentic AI an ideal solution across many business sectors. Enterprises are already leading the way, with examples like Salesforce and ServiceNow offering AI agents, along with many others [1 2 3 4].

Opportunity To Invest

Comprehensive business solutions will involve thousands of LLMs, leading to performance bottlenecks and a high demand for efficient cost-effective infrastructure solutions.

Invest in initiatives supporting policy-based model routing, hierarchical multi-agents, AI swarms, and LLM fleet management. These are just some of the tools that will allow AI companies to deal with the influx of micro-task LLMs in production. 

Opportunity to Avoid

Startups that focus on creating agentic tools but do not provide self-serve turnkey solutions for non-technical business stakeholders, and startups that avoid pursuing deep-tech, cutting-edge agentic solutions that demand ambitious technological breakthroughs. 

Trend

The EU officially approved the AI Act in early 2024, paving the way for secure AI usage, complete with penalties, with key trends: AI governance, AI (cyber) security. Leading the way in this space is Safe Superintelligence, with $1.1 billion in funding. 

Gartner observed enterprises implementing a TRiSM (trust, risk, and security management) layer. 

However, most AI companies relying on non-deterministic statistical GenAI models treat them as though they are deterministic functions. They don't monitor model performance over time, don’t validate the performance state of prompts, don’t conduct AI-related testing for their code and CI, or implement AI observability practices.

Opportunity To Invest

In startups that validate their AI, and are likely to have an AI culture and internal expertise in AI products. These companies will experience fewer business challenges, such as user complaints, customer churn, unmet value expectations, negative growth, etc.

Companies today rely on a multitude of evaluation tools and platforms. Invest in startups that develop meta-evaluation platforms capable of aggregating data from various observability tools.

AI security will become a major focus and establish itself as a standalone trend in addition to TRiSM. Therefore, Invest in startups that offer multi-layered solutions addressing the data stage, the modeling stage, the post-model stages, and during runtime. 

Solutions such as LLM firewalls, plug-and-play TRiSM layers, LLM Cyber & security solutions, LLM search engines with the ability to unlearn [1, 2] information, and evaluation platforms for generative AI.

Opportunity to Avoid

Startups focused on backend log-based observability integration (SIEM). As well as those that neglect proper AI validation, and end up creating AI products that only perform effectively during the development phase.

Trend

The global AI market is set to grow until 2028, multiplying its value, and increasing the percentage of decisions made using AI, in which the global impact of augmentation on workforces is key for productivity, while automation will impact job displacement.

Opportunity To Invest

In AI companies focused on vertical applications, first to market, hyper-focused, with deep tech solutions led by founders with prior expertise in the business domain.

In companies that are augmenting day-to-day roles across specialized domains. From straightforward low-income jobs to complex high-income roles.

Opportunity to Avoid

Horizontal AI companies, focused on broad applicability that are lacking turnkey solutions and fail to effectively address the needs of non-technical business buyers.

However, this approach will shift over time as the global workforce and countries increasingly adopt LLMs tailored to specific industries, transitioning from role augmentation to full automation in specialized sectors.

Trend

80% of people will interact with smart robots daily by 2030. The field of robotics will undergo a complete transformation, affecting everything from mechanical design, electronics and hardware, software development, testing, and simulation to deployment, at scale.

Opportunity To Invest

In startups that disrupt traditional approaches to building and developing robots by enabling faster, more cost-effective manufacturing, new programming approaches, optimized simulation platforms, easier and quicker testing, large-scale manufacturing and deployment, and true multimodal models.
Startups that develop affordable robots will become essential for industries with high demands for manual labor, as well as for a range of complex tasks in workplaces and homes.

In startups that apply the same model across different robots, or fine-tune the same foundational model for various robotic functionalities, as well as those developing true multi-modal GenAI models, optimized and specialized Small Language Models (SLM) designed for specific tasks, that develop LLM models that learn context indirectly [1, 2], model routing, swarm robotics, or multi-agent solutions that manage of SLMs across robotic fleets. 

Opportunity to Avoid

Startups that merely integrate existing off-the-shelf technologies to simulate robotic autonomy, without introducing true innovation that could drive the business forward.

Trend

Edge AI with large language models (LLMs) is set to become a commodity. The integration of AI chips [1, 2, 3], specialized SLMs based on optimization or distillation, multi-modality LLMs, model merging, and runtime memory scheduling will enable smartphone companies like Apple or Samsung to minimize dependence on GPU-powered data centers, thereby reducing costs. This reduction can be passed on to consumers within the hardware price. Additionally, security will improve as fewer API calls to external LLMs are needed, reducing private data exposure.

Opportunity To Invest

In startups that develop compact, efficient AI chips, LLMs or SLMs, designed for easy repurposing that are easily adaptable and can be seamlessly integrated across a range of devices, modalities, and use cases. Keep in mind that these startups are more likely to be acquired by large companies rather than going public through an IPO.

Opportunity to Avoid

Startups with a strong focus on edge computing that overlook critical challenges like security and cost.

Trend

Professional service enterprises are investing in multi-agent AI, fueling global demand, with leaders like Cognizant and Accenture at the forefront.
In partnership with Nvidia, Accenture is conducting extensive AI training for 30,000 professionals and expanding its reach with service hubs to 57,000 AI practitioners worldwide.

Startups are building businesses that cater to the mid-market. They augment and automate workflows using multi-agent AI, which reduces OpEx and increases workforce productivity. These startups deliver value through automation and augmentation, leveraging AI to operate as a scalable professional service without expert human involvement. 

Opportunity To Invest

In startups that are early mid-market entrants, in which professional service companies, such as the Big Four, may lead or participate in their investments in order to secure a mid-market share.

Opportunity to Avoid

Once professional services market leaders acquire mid-market leaders, it will trigger immediate FOMO, likely defining the segment's top winners before the investment momentum fades.

Trend

System integrator startups, which focus on combining components to create a functioning system, lack strong competitive moats.

Similarly, popular GenAI developer tools, such as CoPilot, LangChain, LLamaIndex, CrewAI, and others offer both a free open-source version and a premium SaaS platform, but ultimately they all share the same core MLOps functionalities on top of their offering, for example, model deployment and model market places.

These companies may struggle to innovate horizontally because their solution can be easily replicated.

Opportunity To Invest

In startups that build automatic assembly lines using agents at scale, or have a robust business model that is deeply embedded in the domain and require specialized knowledge that cannot be addressed through simple and generalized integrations.

In open-source developer tools that offer and extend beyond their core functionality, and integrate within a platform that aims to support micro-agents, by delivering high availability, low latency, hyper-scalable, and LLMs fleet management.

Opportunity to Avoid

Avoid startups who treat AI as a modular component, connecting open-source AI solutions to address immediate functionality needs, i.e., they quick fix rather than invest significant efforts in solving complex, non-standardized business and product challenges.

Avoid open-source tools that don't offer added premium value beyond basic governance, observability, MLOps deployment, and marketplaces, which have largely become commodity features and are abundant.