Retool has just released its latest “State of Artificial Intelligence Report” for the first half of 2024, collecting opinions from about 750 technicians, including developers, data teams, and leaders from various industries, to understand how to use artificial intelligence to make a real impact.
Key Findings of Report
AI Emotion Changes
In the 2023 report, most respondents were cautious about AI, believing it to be slightly overrated. In 2024, this sentiment did not change significantly, with the average rating for each role still around 5 points (“reasonable evaluation” to “slightly overrated”).
Respondents believe that although AI has great potential, it is still in a relatively immature stage and many applications have not truly realized their value.
Current state of AI applications
Most respondents said their companies have not yet reached the highest level of AI adoption, with only about 30% of respondents believing their companies are at the “operational” or “flight” stage in terms of AI adoption.
Among different industries, AI application is highest in consulting (46%), real estate (46%) and consumer goods (37%).
Respondents from nonprofits and the materials industry said they lag behind in AI adoption.
Investment and productivity
The vast majority of respondents use AI tools at least once a week, and 56.4% use them almost every day. Small companies (1–9 employees) and larger companies (1,000–4,999 employees) use them daily at 72% and 43%, respectively.
Among different roles, the product and engineering departments have the highest daily usage rates, at 68% and 62.6% respectively, while the design department has the lowest, at 39%.
Respondents who frequently use AI tools report significant productivity gains, with 64.4% of daily users reporting significant productivity gains, compared to 17% of weekly users and 6.6% of occasional users.
Secret use
The percentage of companies that allow the use of AI at work has increased from 54% to 64.2% in 2023. However, 27.3% of respondents still use AI secretly when company policies are unclear or not allowed.
Only about 9% of respondents use AI secretly in violation of company policy, while the motivations for the others are mainly internal alignment and mixed views on the use of AI.
Alternative Risk
45.7% of respondents believe that junior employees are most likely to be replaced by AI, followed by middle managers (13.2%), while senior ICs and executives have a lower risk of replacement.
Senior executives (such as the C-suite) are more likely to see AI tools as a productivity boost, which suggests that AI is seen as a tool, not a threat.
AI technologies and tools
Model Usage
OpenAI’s models (such as GPT-4 and GPT-3.5) are most commonly used in production, accounting for about 76.7% of the market share. Among them, GPT-4 has the highest usage rate, which is 45%, followed by GPT-3.5, which accounts for 25%.
Anthropic’s Claude 3 model performed well in this survey, with its usage rate significantly increased compared to before. Mistral also appeared on the list of used models.
Tools and Databases
The number of companies using vector databases (such as MongoDB and Chroma) for model customization has increased significantly, with 63.6% of respondents reporting using vector databases.
The larger the company, the more likely it is to use a vector database or RAG (search-augmented generation), with 33% of companies with more than 5,000 employees using these tools.
The most popular vector database is MongoDB, which has the highest satisfaction rate. Chroma is a close second, with high user satisfaction as well.
Hardware Resources
Only 13.2% of companies own or operate their own GPUs, with most renting GPU resources from cloud providers such as Amazon Sagemaker and Databricks. Databricks performed best in terms of user satisfaction.
Among companies that invested in GPUs, 53.7% said that this investment brought a positive ROI (return on investment), especially among large companies with more than 5,000 employees, 40% of companies reported a positive ROI.
AI Development Tools
59.1% of companies use some form of AI development tool, either homemade or commercial. The most commonly used single tool is HuggingFace, accounting for 15.8% and with the highest user satisfaction.
Among the multi-tool combinations, LangChain (21.3%) and HuggingFace (20.1%) are the most commonly used tools.
40.9% of respondents do not use additional AI development tools, indicating that there is still a lot of room for development in this area.
Challenges and Prospects
Main obstacles
The main barriers to developing AI applications include data security, insufficient resources, and organizational challenges (such as obtaining senior management support). These barriers are not related to AI technology itself, but to the enterprise environment and resource management.
38.9% of respondents mentioned accuracy and “illusionment” of model outputs as major issues, and 33.5% mentioned data access and security issues.
Satisfaction
Overall, respondents are highly satisfied with their AI technology stack, with more than 72% knowing where improvements are needed and willing to make adjustments. The most satisfied industries include healthcare (45%), education (43%), consulting (42%), and nonprofits (40%).
Future Outlook
Most respondents believe that there is a high probability that general artificial intelligence (AGI) will be achieved within the next decade, with 59.9% believing that the probability is high or very high, and 2.7% believing that AGI already exists.
