4 Steps to Building Smarter AI Agents

Salesforce's new framework details how CIOs can succeed in the agentic era through this four-step process. Image credit: Imagen 3/Ken Yeung
, , , , ,
"The AI Economy," a newsletter exploring AI's impact on business, work, society and tech.
Welcome to "The AI Economy," a weekly newsletter by Ken Yeung on how AI is influencing business, work, society, and technology. Subscribe now to stay ahead with expert insights and curated updates—delivered straight to your inbox.

IN THIS ISSUE: Dive into Salesforce’s new framework for CIOs, offering advice on how their organizations can effectively harness the power of AI agents. Explore the four steps necessary to develop mature, intelligent, effective bots, and discover how companies can align their AI strategies with business goals.

Plus, catch up on the latest from Google Cloud Next 2025 in this week’s most important AI news roundup. 

The Prompt

Have you heard about the latest company to launch an AI agent?

These bots are everywhere right now—and with all the buzz, some are starting to wonder if agents are just overhyped. The term itself has become a bit of a catch-all, with definitions varying wildly depending on who you ask. The reality is that what agents can do is determined by how integrated, scaled, and governed the technology is within an organization. 

As AI agents become increasingly central to how businesses operate, Salesforce is stepping in to define the path forward with its new Agent Maturity Model framework—a guide for organizations looking to scale their use of autonomous and assistive agents. The company has identified a four-step roadmap that IT departments can use to help identify where they are today, what’s needed to progress further, and how to align their AI deployment with their firm’s long-term business objectives.

“While agents can be deployed quickly, scaling them effectively across the business requires a thoughtful, phased approach,” Shibani Ahuja, Salesforce’s senior vice president of enterprise IT strategy, remarked in a blog post. “Understanding the progression of AI agent capabilities is crucial for long-term success, and this framework provides a clear roadmap to help organizations move towards higher levels of AI maturity.

The four steps of Salesforce's Agentic Maturity Model. Image credit: Salesforce
The four steps of Salesforce’s Agentic Maturity Model. Image credit: Salesforce

Level 0: Fixed Rules and Repetitive Tasks

This is the lowest level in the Agent Maturity Model and likely the form most people experience when interacting with an agent. Salesforce defines these bots as handling “automation of repetitive tasks using predefined rules, with no reasoning or learning capabilities.” In other words, these agents mainly focus on retrieving information—no further action is needed. 

To progress beyond this level, Salesforce advises organizations to examine use cases in which chatbots and copilots are disadvantaged by rigid decision trees. They should conduct time studies to understand how much time would be saved through agents automating tasks and recommending actions. Then, after choosing the use cases that match the firm’s risk tolerance and establishing a mitigation strategy, properly connect the data sources to the agent.

Level 1: Information Retrieval Agent

These are bots that not only find and return information but also recommend the next steps. An agent that surfaces data from a knowledge base to suggest how to answer a customer support ticket fits this bill.

For organizations looking to evolve their bots, Salesforce suggests moving from recommendations to actions—don’t leave it to humans to decide on what’s next; make the agent think proactively. In addition, IT teams should “harmonize additional data” that’s “cleansed, transformed, and pre-processed.” Governance frameworks should be put in place, along with policies for collecting user feedback and measuring agent performance.

Level 2: Simple Orchestration, Single Domain

At this level, organizations have more intelligent agents capable of completing low-complexity tasks but using siloed data. After you’ve trusted it enough, it’s time to give it some more responsibilities but limited to a single specific area or system, such as sales, support, or scheduling. An example is a meeting scheduling agent that also automates follow-up emails using an internal calendar and email system.

To evolve agents, Salesforce recommends teams determine whether one or more multi-functional agents are needed to achieve company goals. Factor in latency impacts when choosing multiple bots. Then, give these agents the same access to data as their human counterparts while limiting their permissions to only what’s necessary to complete their tasks. Ensure responsibilities are clearly separated to maintain security and control.

Level 3: Complex Orchestration, Multiple Domain

Agents in this penultimate level are like those in level two but can operate across different functional areas and handle complex, cross-functional tasks. Salesforce cites an agent managing an organization’s sales pipeline while pulling data from the Customer Relationship Management system, customer service tickets, and financial reports to compile a holistic customer view as an example.

To advance to the final step, it’s advised that teams focus on real-time, cross-domain agent collaboration use cases requiring optimized workflows using dynamic agent teams. Additionally, a universal agent communication layer, such as an API, is needed, along with dynamic agent discovery for regular maintenance. Companies should utilize a scalable architecture to support any-to-any agent interactions, fine-grained access controls, security policies, and a robust governance framework with auditability, error handling, and transparency.

A policy for layered human/AI supervision, a methodology for managing the AI agent lifecycle sustainably, and a system for measuring ROI are also essential.

Level 4: Multi-Agent Orchestration

At this stage, multiple agents collaborate across different systems to complete tasks and are supervised by another agent. Bots will work together interdependently and handle workflows across various domains. Imagine multiple agents teaming up to autonomously process orders, manage inventory, and route customer feedback through different departments in real-time.

Although there’s no other evolution, Salesforce offers these tips on how organizations can ensure agents generate the maximum value: Strengthen security and governance protocols so it supports seamless interactions across the entire ecosystem. Identify new business models created by these agent collaborations. Develop metrics to measure multi-agent system value. Lastly, when monitoring ROI, focus on the broader business impact, tracking outcomes such as revenue growth, cost savings, and enhanced customer retention.

It may seem daunting to create these sophisticated agents, but new technology innovations are helping to streamline their development. Companies like Google, Microsoft, IBM, OpenAI, and LangChain have released Agent Development Kits (ADK) so developers can add AI to their applications. In addition, multiple groups are working on interoperability standards, such as AGNTCY and Google’s new Agent2Agent (A2A).

A critical distinction between these levels isn’t just the intelligence or system integrations behind the agents but the development of robust governance and security policies—particularly as these bots begin to process data from multiple systems. As these agents evolve, they will take on an increasingly central role in automating core tasks. To successfully navigate this shift, organizations must ensure these processes are carried out securely while also preparing the workforce to embrace these AI agents as part of the team.

While this framework can apply to any AI platform, Salesforce hopes to reap the benefits of the Agent Maturity Model. After all, Salesforce has criticized technology providers, claiming their AI solutions don’t offer value. And if its Agentforce platform is to succeed, Salesforce needs to further educate customers on what an agent can do. The company may also hope it’ll encourage Trailhead participants to dive deeper into AI courses, seeing there’s much more to learn about the technology. It’s already facing headwinds when trying to sell its Agentforce platform, so maybe frameworks like this will spark some creativity inside organizations to embrace the agentic era. Sharing this knowledge is intended to keep Salesforce top of mind and viewed as a market leader.

Quote This

“Using AI effectively is now a fundamental expectation of everyone at Shopify. It’s a tool of all trades today and will only grow in importance. Frankly, I don’t think it’s feasible to opt out of learning the skill of applying AI in your craft; you are welcome to try, but I want to be honest: I cannot see this working out today, and definitely not tomorrow. Stagnation is almost certain, and stagnation is slow-motion failure. If you’re not climbing, you’re sliding.”

— Shopify CEO Tobi Lutke, in a company memo, stated that “reflexive AI usage is now a baseline expectation.” He later instructed employees looking for more headcount and resources to “demonstrate why they cannot get what they want done using AI.”

Don’t Miss out on Future Issues of ‘The AI Economy’

The AI Economy is expanding! While you’ve been getting weekly insights on LinkedIn, I’m gearing up to bring you even more—deep dives into AI breakthroughs, more interviews with industry leaders and entrepreneurs, and in-depth looks at the startups shaping the future. To ensure you don’t miss a thing, subscribe now on Substack, where we’ll be rolling out more frequent updates. 

Don’t worry; the weekly newsletter will still be published on LinkedIn, but other stories will be available on Substack.

Subscribe to The AI Economy

This Week’s AI News

🏭 AI Trends and Industry Impact

🤖 AI Models and Technologies

✏️ Generative AI and Content Creation

💰 Funding and Investments

☁️ Enterprise AI Solutions

⚙️ Hardware, Robotics, and Autonomous Systems

🔬 Science and Breakthroughs

💼 Business, Marketing, Media, and Consumer Applications

💥 Disruption, Misinformation, and Risks

🔎 Opinions, Analysis, and Editorials

End Output

Thanks for reading. Be sure to subscribe so you don’t miss any future issues of this newsletter.

Did you miss any AI articles this week? Fret not; I’m curating the big stories in my Flipboard Magazine, “The AI Economy.”

Follow my Flipboard Magazine for all the latest AI news I curate for "The AI Economy" newsletter.
Follow my Flipboard Magazine for all the latest AI news I curate for “The AI Economy” newsletter.

Connect with me on LinkedIn and check out my blog to read more insights and thoughts on business and technology. 

Do you have a story you think would be a great fit for “The AI Economy”? Awesome! Shoot me a message – I’m all ears!

Until next time, stay curious!

Subscribe to “The AI Economy”

Exploring AI’s impact on business, work, society, and technology.

Subscribe

Leave a Reply

Discover more from Ken Yeung

Subscribe now to keep reading and get access to the full archive.

Continue reading