Image source: Generated by Unbounded AI

In 2024, AI agent technology will undoubtedly be the focus of the technology world.

Through an in-depth survey of 3,400 developers from more than 100 countries, a report recently released by Langbase revealed key issues in the development of intelligent agents. Among the respondents to this survey, 46% are business leaders and 26% are engineers.

Langbase focuses on providing a serverless AI cloud development experience. As an AI platform that provides composable infrastructure, they processed a total of 184 billion tokens and 786 million API requests from 36,000 developers in 2024.

You may not be familiar with the name Langbase. In fact, this company is very young, having just been established in 2023.

According to data disclosed by Crunchbase, they just conducted a pre-seed round of financing in September, but have already received financial support from executives of major companies such as Google, Apple, OpenAI, and Microsoft.

Let us take a look at what is mentioned in this illustrated report released by the young and promising Langbase.

Question 1: What base large models are developers using?

In this competition for AI agents, OpenAI’s large model service Google dominates the market, but Google is quickly emerging as a serious competitor, with Anthropic not far behind.

Among them, although Meta’s Llama, Mistral and Cohere have less influence, their growth momentum cannot be ignored, showing fierce competition in the base large model market.

Question 2: The specific uses of different large models

OpenAI is widely used in translation tasks, Anthropic is popular in technical tasks, and Google's models dominate in health and translation fields.

In addition, Meta has been widely used in technology and scientific applications, and Cohere has also been valued in many fields, including science and marketing.

Question 3: What factors hinder your application of large model technology

In the expansion and deployment of the model, data privacy and security compliance have become the main concerns. The two factors of "lack of monitoring tools" and "high infrastructure costs" also hinder the implementation of the technology.

Resistance or skepticism toward AI-driven solutions reflects long-term concerns among users and suggests that providers of models and technologies need to be morePlus a transparent and user-friendly AI platform.

Question 4: What factors affect the selection of large models

In When choosing a large language model (LLM), accuracy is the most important factor, followed by safety and customizability, and cost has a relatively small impact.

Question 5: What is the biggest challenge you encounter when deploying large models?

Deploying large models and agents in production presents key challenges, including difficulties with customization, limited quality assurance assessment methods, and a lack of reusable infrastructure.

Additionally, tool fragmentation, integration issues, and scalability concerns further complicate the process.

Question 6: What is your main goal when adopting large model technology?

"Automation" and "simplification" are the top priorities for AI applications, benefiting companies in terms of efficiency and process simplification; in addition, customized solutions and improvements Goals for areas such as collaborative processes reflect the growing flexibility of large models and consumer interest in shared access to systems.

Question 7: How does your company use large model agents?

LLM is widely used in software development, especially in marketing, IT operations and text summarization, customer service, human resources and legal fields Interest is also growing.

Question 8: Which platform features are important to you

Most respondents require multi-agent retrieval-augmented generation (RAG) capabilities to improve contextual information processing, while evaluation tools are also important to ensure AI systems work as expected.

Question 9: Which tool do developers prefer when arranging AI pipelines

Most respondents prefer development tools that provide flexible, basic primitives for designing customized AI pipelines. Pre-built, problem-specific solutions may directly solve specific problems, but they are less customizable.

Question 10: What factors affect the choice of large model agent development tools

Developers regard version control of AI agents as the most important function of the development platform. A strong ecosystem of SDKs or libraries and local development environments are also valued.

Question 11: The extent to which large models are used in the company

Most developers use AI for experimentation and production. In contrast, the proportion of experimental uses is much larger than that of production uses, but the proportion of the latter is still in the process of steady progress.

As the AI ​​agent infrastructure matures, more and more companies will try to develop agents, and with the development of multi-modal and large-model operating computing environments (such as Claude interacting online through interfaces) Development, the application of intelligent agents will be more extensive in 2025, and it is not even limited to the fields such as software development, marketing, IT operations and text summarization mentioned in the report.

However, developers revealed in the survey are concerned about issues such as placing more emphasis on accuracy, security and customizability, while cost factors have less impact, and preferring flexible and customizable AI pipeline development. tools, and pay attention to the AI ​​agent version control function.

No matter how technology advances, the results of this survey will be of reference value to intelligent agent platforms and developers.