AI-ming for the Stars Part 4: Agentification of Application Layer

You might think the AI frenzy of 2023 has settled, but leading AI innovators are maintaining their relentless pace, unveiling mind-blowing releases with ruthless execution nearly every month. In the past month alone, OpenAI announced the launch of SearchGPT, an AI-powered search engine designed to deliver real-time information from the web, positioning itself as a direct competitor to Google. Adding to the excitement, OpenAI is set to release ChatGPT-4o, a voice assistant capable of creating custom character voices, generating sound effects while telling a story, and even acting as a live translator. Moreover, OpenAI introduces GPT-4o mini, its most cost-efficient small model yet. Not to be outdone, Meta has officially unveiled LLaMA-3.1 405B, the world’s largest open-source LLM to date, boasting 405 billion parameters and trained on 15 trillion tokens. Meta claims it outperforms GPT-4, GPT-4o, and Claude 3.5 Sonnet across various tasks.

The foundational layer has seen impressive advancements over the last year, spanning from LLMs (Large Language Models) to LAMs (Large Action Models) and LAEs (Large Agentic Models). These innovations are enabling startups to revolutionize business operations by automating numerous repetitive and manual tasks previously thought impossible. These advancements are bringing unprecedented efficiency to systems and setting the stage for the next wave of innovation in the application layer.

We have witnessed founders evolve from creating basic copilots that answer questions, retrieve information, and summarize text to developing AI-native solutions and AI agents. These agents possess reasoning capabilities, can plan sequences of actions, and interact with software to execute complex tasks. While many incumbents have integrated copilots into their existing products, startups have the advantage of building AI agents that delve deeper into workflows and enhance operational efficiency.

Source: https://lilianweng.github.io/posts/2023-06-23-agent/

AI agents represent a significant shift towards more comprehensive and human-like AI applications, enabling businesses to:

• Reduce Costs: Intelligent agents help businesses minimize unnecessary expenses by addressing process inefficiencies, reducing human errors, and automating manual processes.
• Automate Higher-Level Tasks: These agents go beyond basic automation to support strategic decision-making and planning.
• Enhance Customer Experience: By providing personalized, empathetic, and human-like interactions, AI agents improve customer engagement, conversion rates, and loyalty.

We have looked at many startups building specialized agents for different functions and verticals. As we keep refining our thesis in this ever-changing field, our views of different segments are:

1. Sales and Marketing: Have always remained the most crowded segments in B2B SaaS. Also saw a huge influx of AI startups building to automate various workflows as it is closest to the revenue and time to value is faster. However, major questions remain about the differentiation and stickiness of the product. Major sales tech/ martech giants hold the majority value as they are the system of records and it makes it difficult for startups to build a very large business in this.
2. New System of Records: Disrupting horizontal systems of records like Salesforce, HubSpot, and Workday has traditionally been challenging. However, advancements in LLMs and AI agents are changing this dynamic. The way we interact with systems and data is evolving, creating opportunities for new systems of record. AI agents can now access and analyze data from emails, calls, and other sources, providing more accurate and comprehensive insights than those entered manually into CRMs. This shift means that instead of interacting through traditional dashboards, users will increasingly engage with AI agents directly, transforming how data is managed and utilized.
3. Back-office Functions: Functions such as HR, accounting, finance, and legal present significant opportunities for AI innovation. These areas often involve pattern-based workflows with high usage and habitual engagement, making them ideal for automation. These are internal non-customer-facing functions, where businesses are using some outdated clunky software and there is a lot of manual data manipulation and analysis involved. AI agents can streamline and automate these repetitive tasks, improving efficiency and accuracy.
4. Recruitment tech: We are seeing many AI recruitment tech startups, this is again a good area to show a quick time to value. The hiring process is a major pain point for large organizations, often involving long processes with many repetitive steps and extensive back-and-forth between candidates and employers. Candidates often complain about the poor interview experience, lengthy coordination and long hiring timelines. For large organizations and traditional industries with high turnover, AI agents for hiring are a game-changer. These agents can automate the entire recruitment process, including initial interview rounds, thereby improving candidate experience, and reducing time to hire.
5. Vertical industry-focused agents: We are seeing vertical AI applications in healthcare, biotech, financial services, wealth-tech, insurance, and legal. They are using industry-specific models/data sets for a deeper understanding of specialized data, enabling the extraction of insights and automation of workflows with high precision. These focused agents are more reliable as they are tailored to specific domains or defined sets of tasks. We've observed startups creating multiple specialized agents within a single industry to automate a wide range of tasks effectively. This is a large opportunity as startups begin with a wedge of data insights and gradually expand to automate more complex workflows.
6. Services as a Software: Approx 80% of the economy is service-based and has historically resisted tech disruption due to unstructured data and complex reasoning requirements. However, AI is now making it possible to revolutionize the services industry. We are seeing early signs of business model innovation, with outcome-based service models built on top of AI solutions. These models provide better, faster outcomes and are easier to scale, all with minimal human intervention.

These are exciting times to build AI applications. The time to develop software has significantly reduced, giving the ability to experiment and iterate quickly. In the competitive AI market, the race is on to build the best product, create deep and sticky workflows, and quickly acquire customers. Targeting high-value, high-volume work, or areas facing labor shortages can provide immediate and substantial benefits. The fundamentals of building a large business remain the same: build for specific end-users, ensure the workflow is valuable and critical to their job, which allows for effective pricing, and successfully execute go-to-market and distribution strategies.

We are always on the lookout for interesting startups in B2B applications and are happy to jump on a call to brainstorm. Get in touch with Gaurav Chaturvedi (gaurav@kae-capital.com)