How businesses are accelerating time to agentic AI value

A recent survey of 1,050 CIOs revealed that 93% of IT leaders will implement AI agents in the next two years, with IT leaders working to implement the technology by focusing on removing data silos.

The average number of apps used by respondents was 897, with 45% reporting using 1,000 applications or more, hindering IT teams’ ability to build a unified experience. 

Also: The end of data silos? How SAP is redefining enterprise AI with Joule and Databricks

Only 29% of enterprise apps are integrated and share information across the business. To prepare for the expanded use of AI, enterprise CIOs allocate 20% of their budgets to data infrastructure and management, four times more than their spend on AI (5%). 

So, what are AI agents? According to ARK Invest, AI agents are poised to accelerate the adoption of digital applications and create an epochal shift in human-computer interaction because they:

• Understand intent through natural language
• Plan using reasoning and appropriate context
• Take action using tools to accomplish the intent
• Improve through iteration and continuous learning

According to ARK, AI will supercharge knowledge work. Through 2030, ARK expects the amount of software deployed per knowledge worker to grow considerably as businesses invest in productivity solutions. Depending on adoption rates, global spend on software could accelerate from an annual rate of 14% over the last 10 years to annual rates of 18% to 48%. 

So, how can businesses accelerate the time to value from agentic AI? According to technology research firm Valoir, agentic AI promises to deliver exponential benefits from AI by automating complex tasks and interactions without human intervention. 

However, creating agentic AI that can handle complex tasks with acceptable performance is a challenge. Valoir found using a platform optimized for agentic AI development, such as Salesforce Agentforce, enables organizations to deliver autonomous AI agents an average of 16 times faster than other approaches while increasing accuracy by 75%. 

Also: Crawl, then walk, before you run with AI agents, experts recommend

Valoir has defined seven phases of agentic development (the complexity of agentic tasks and volume, sources, and hygiene of data varied by customer, as did the size and level of data):

1. Model setup
2. Data and application integration
3. Prompt engineering
4. AI guardrails and security
5. User interface and workflow/application development 
6. Tuning
7. Data accuracy

One key finding from Valoir regarding model setup was the variations between a Do it Yourself (DIY) approach and a deeply integrated platform with embedded agentic AI capabilities. 

Also: AI data centers are becoming ‘mind-blowingly large’

Valoir found that most organizations taking a DIY approach use pre-built models, typically requiring three to 12 months to set up. In contrast, Agentforce’s models are pre-integrated and pre-tuned, requiring little to no set up time, on average 7.5 times faster versus pre-built models.

Valoir also found that organizations using open-source alternatives spent at least a month selecting a RAG approach. Processes included integrating document ingestion, retrieval, and storage tools, integrating the RAG with generative models, and an additional two to three months to train the retriever and model with domain-specific data. Agentforce data and app integration was completed in weeks, or three and a half times faster. 

The most significant comparison of DIY vs using a deeply integrated AI platform was for AI guardrails, trust, and security. Trust was the key factor enabling organizations to move from generative to agentic AI use cases. Development teams with significant development and data science expertise would need more than 12 months to develop the equivalent trust layer.

Also: AI agents might be the new workforce, but they still need a manager

Data accuracy is a key factor in time to value, the time needed to build and train AI agents to deliver acceptable levels of correct response. Depending on task complexity, the accuracy percentage varied based on DIY approach versus using a deeply integrated platform. 

For simple tasks, the accuracy rates were 50% for DIY versus 95% for Agentforce. In complex tasks, such as sales coaching, the accuracy was 40% for DIY versus 95% for Agentforce. Overall, the platform approach can increase agent accuracy by 75%. 

Valoir concluded that the average total months spent on DIY projects was 75.5 while the average time needed to bring an Agentforce project to productive accuracy was 4.8 months, making the platform approach 16 times faster. To learn more about Valoir’s agent AI research, go here.