AI agents: From experimentation to value creation
Practical steps for business leaders to drive real value with agents today
The shift to AI agents
The agent era of AI is upon us. But agents are not a new concept. The idea of machines that can perceive, reason and act on their environment covers a lot of history, from the Rand Corporation’s General Problem Solver (GPS) in the ‘50s to antiviruses starting in the ‘70s to the chess-playing Deep Blue in the ‘90s to the ubiquitous Siri (2011) and Alexa (2014) voice assistants.
The difference now is that the reasoning part of the equation is readily available to everyone. In other words, the cost prohibitive barrier to success at scale has been eliminated. That said, the emergence of AI agents as practical business tools isn't the result of a single technological advancement—it’s the convergence of several critical factors, including advanced language models with sophisticated reasoning capabilities, improved APIs that make it easier to connect agents with existing systems, and maturing digital and data infrastructures that provide the framework for deriving value.
This convergence has made it both feasible and affordable for every organization to incorporate AI agents into their business strategy, opening up windows of opportunity that weren’t available even a few years ago.
What exactly is an AI agent?
There are blurred lines between how people discuss basic generative AI solutions, LLM workflow patterns, and agentic systems. There is no “right” or “wrong” to these variations, but to add clarity to this discussion, we distinguish agents from other AI solutions based on the existence of five core components:
Autonomy
AI agents dynamically plan and execute multiple steps, selecting tools or making decisions without explicit human prompts at every turn.Use of tools
AI agents can leverage tools to do research, test theories, perceive the world around them and complete tasks (e.g., integrating with CRM software or using APIs).Complex reasoning
AI agents can make logical assumptions based on a variety of inputs, going well beyond rules-based decision-making.Adaptability
Agents respond to evolving conditions—retrieving information, interpreting instructions, and adjusting their approach to learn and improve over time.Modularity
Think of agents as experts, not generalists. Agents are designed for specific tasks and have the related expertise, tools, and intelligence for that task. Teams of agents may combine their expertise to complete even more complex tasks.
While conventional AI excels at specific tasks—like image recognition or text generation—AI agents operate with a degree of autonomy that fundamentally changes how work gets done. These agents can maintain ongoing operations, adapt to changing conditions, and handle complex sequences of decisions that previously required human oversight at every step. When well-designed, agents can handle tasks a simple workflow can’t.
Where we’re seeing value today
The AI agent market is rapidly evolving with both large and small platform vendors emerging almost weekly. We tend to divide vendors and custom solutions into two categories: horizontal and vertical agents. Horizontal focuses on broad intelligence that can provide outcomes across many general topics, such as the broad enterprise solutions from Anthropic and OpenAI. Vertical solutions are trained to provide deep expertise in a focused area like an industry solution or specific role.
Below are a few areas where our clients are seeing immediate value from agents.
Internal operations
Agentic systems shine in complex, back-office scenarios where the path forward isn’t always predictable. Instead of strictly scripted automations, an agent can dynamically decide which processes or data sources to invoke based on the current input.
Coding agents are a prime example. In some development teams, a central "orchestrator" agent manages high-level objectives—such as improving system performance—by delegating specific tasks to "worker" agents, each focusing on a particular aspect of the project. This approach is especially effective when the scope of work is unpredictable, such as varying numbers of files or the extent of required updates. However, for routine, well-defined tasks like transferring data between systems, simpler automation solutions may suffice.
Companies that do adopt agentic systems in internal operations often see lower error rates and time savings. Agents self-monitor their progress, escalating exceptions only when human approval is necessary. This way, employees can focus on high-level strategy while routine tasks happen seamlessly behind the scenes.
Customer experience
In customer interactions, generative AI is becoming a cornerstone of CX innovation. In 2024, customer service chatbots emerged as a top priority, with 53% of enterprises identifying them as their primary generative AI use case. When we consider that the broader AI landscape saw an astonishing sixfold surge in overall investment over the previous year, it's clear that organizations see distinct value in bringing AI to their customer service operations.
Agentic systems are evolving chatbots to handle multi-part questions, dynamically retrieve information from multiple systems, and act to resolve issues. Jaja Finance’s “Airi,” developed in partnership with Slalom, has dramatically transformed Jaja's customer service by automating routine inquiries, cutting response times by 90%, and freeing up 30-40% more time for service teams to focus on complex, high-value tasks. By relieving human representatives of repetitive tasks, Airi enables them to concentrate on nuanced issues that require empathy, critical thinking, and deep problem-solving skills.
United Airlines partnered with Slalom to create an AI agent that helps provide a more personal touch to travelers experiencing issues. When a flight disruption occurs, the agent weighs multiple real-time factors—weather, staffing, aircraft location—and decides on a messaging strategy to improve the overall experience for the traveler. This kind of adaptability is where agentic systems truly shine, dynamically orchestrating data gathering, validation, and result generation to expedite confident, data-driven decisions.
Complex decision-making
When it comes to high-impact decisions, agentic systems can pull from diverse data sources, weigh pros and cons, and refine recommendations over time. This is particularly critical in dynamic environments where every situation requires a tailored approach.
At Slalom, we use a coordinated approach called "Fleet of Analysts" (FoA), where multiple specialized agents work together. Each agent focuses on a specific area of expertise, analyzing large volumes of data, identifying actionable insights, and supporting strategic decision-making. These agents can operate independently or collaboratively with human oversight, ensuring both precision and adaptability.
For example, one customer leveraged FoA to scale their ability to dynamically price products globally across 80+ countries. This helped them monitor dozens of internal and external sources of information that might influence pricing, analyze that against historical pricing strategies and current economic environments, and create automated pricing decisions.
Another customer applied this approach to enhance their enterprise architecture capability. Teams submitting new project proposals were able to rely on an intelligent agent to conduct architecture reviews. The agent coordinated research across internal guidelines, external regulations, security requirements, and industry best practices to deliver detailed assessments and improvement recommendations. This significantly reduced the workload for enterprise architects, allowing them to focus on higher-value strategic initiatives.
In both cases, the AI agents provided much-needed scalability to expert roles that traditionally spend an inordinate amount of time manually grinding out complex decisions, allowing them to focus on strategic priorities.
United Airlines partnered with Slalom to create an AI agent that helps provide a more personal touch to travelers experiencing issues. When a flight disruption occurs, the agent weighs multiple real-time factors—weather, staffing, aircraft location—and decides on a messaging strategy to improve the overall experience for the traveler. This kind of adaptability is where agentic systems truly shine, dynamically orchestrating data gathering, validation, and result generation to expedite confident, data-driven decisions.
Complex decision-making
When it comes to high-impact decisions, agentic systems can pull from diverse data sources, weigh pros and cons, and refine recommendations over time. This is particularly critical in dynamic environments where every situation requires a tailored approach.
At Slalom, we use a coordinated approach called "Fleet of Analysts" (FoA), where multiple specialized agents work together. Each agent focuses on a specific area of expertise, analyzing large volumes of data, identifying actionable insights, and supporting strategic decision-making. These agents can operate independently or collaboratively with human oversight, ensuring both precision and adaptability.
For example, one customer leveraged FoA to scale their ability to dynamically price products globally across 80+ countries. This helped them monitor dozens of internal and external sources of information that might influence pricing, analyze that against historical pricing strategies and current economic environments, and create automated pricing decisions.
Another customer applied this approach to enhance their enterprise architecture capability. Teams submitting new project proposals were able to rely on an intelligent agent to conduct architecture reviews. The agent coordinated research across internal guidelines, external regulations, security requirements, and industry best practices to deliver detailed assessments and improvement recommendations. This significantly reduced the workload for enterprise architects, allowing them to focus on higher-value strategic initiatives.
In both cases, the AI agents provided much-needed scalability to expert roles that traditionally spend an inordinate amount of time manually grinding out complex decisions, allowing them to focus on strategic priorities.
Making agents work for your organization
Identifying the right opportunities
A successful approach begins by identifying high-value opportunities and evaluating whether simpler solutions—like basic workflows or direct LLM calls—might suffice. Autonomous systems shine in complex, dynamic workflows requiring decision-making and adaptation, but unnecessary complexity can lead to wasted resources.
For example, coding agents can automate tasks like code generation and testing, boosting developer productivity up to 2.5x by eliminating repetitive work without sacrificing quality. Meanwhile, support agents handle unpredictable customer requests seamlessly, while simpler FAQ bots may be more appropriate for routine queries. The key is aligning the tool to the task to maximize ROI.
Defining clear metrics—such as error rates, handling time, or user satisfaction—is crucial. Consider metrics tied to strategic outcomes like cost savings or retention to guide your decisions. Tools like our AI-powered accelerator, enhanceIQ, help identify areas where agent-driven systems can deliver the greatest impact.
Essential building blocks
Deploying agentic systems calls for more than just a good LLM—it requires structure in five critical areas:
Business and customer value: Identify real-world outcomes for the agent. Does it cut costs, speed up resolutions, or enable new user experiences?
Strategy alignment and orchestration: Ensure stakeholders understand how an agent’s autonomy aligns with broader goals. Cross-functional teams are crucial to unify efforts.
Security, ethics, and governance: Agents can amplify both good and bad outputs. Implement robust guardrails such as iteration limits and human checkpoints to maintain oversight and avoid negative consequences.
Technology and data: Agents rely on up-to-date knowledge. Solid data engineering, real-time APIs, and logging infrastructure keep them accurate and auditable.
Organization and workforce: Employee readiness matters. Provide training on how to collaborate with an AI agent and clarify roles so staff know when to step in and override the system.
Slalom's AI-powered accelerator, enhanceIQ, shows how different roles can benefit from AI solutions. By uncovering the ways AI and automation can benefit work across teams, you’re empowered to make data-driven decisions that help you build a more adaptable workforce.
Slalom's AI-powered accelerator, enhanceIQ, shows how different roles can benefit from AI solutions. By uncovering the ways AI and automation can benefit work across teams, you’re empowered to make data-driven decisions that help you build a more adaptable workforce.
Essential building blocks
Deploying agentic systems calls for more than just a good LLM—it requires structure in five critical areas:
Business and customer value: Identify real-world outcomes for the agent. Does it cut costs, speed up resolutions, or enable new user experiences?
Strategy alignment and orchestration: Ensure stakeholders understand how an agent’s autonomy aligns with broader goals. Cross-functional teams are crucial to unify efforts.
Security, ethics, and governance: Agents can amplify both good and bad outputs. Implement robust guardrails such as iteration limits and human checkpoints to maintain oversight and avoid negative consequences.
Technology and data: Agents rely on up-to-date knowledge. Solid data engineering, real-time APIs, and logging infrastructure keep them accurate and auditable.
Organization and workforce: Employee readiness matters. Provide training on how to collaborate with an AI agent and clarify roles so staff know when to step in and override the system.
When these elements work in tandem, agentic solutions can transition from pilot to full production smoothly. If any link is weak—like data quality or unclear governance—you’ll struggle to access the agents’ full potential.
A practical approach to getting started
Start with a focused pilot project to automate a specific process, such as a support agent for a product line with complex inquiries or a coding agent targeting known vulnerabilities in select repositories. Early wins demonstrate value, build momentum, and lay the groundwork for broader adoption while managing risk.
Key criteria for selecting pilot projects include …
Clear production outcomes: Go beyond the lab. Focus on delivering tangible value, not just experimental results.
Manageable complexity: Add complexity only when it demonstrably improves outcomes. Simplify or iterate as needed to ensure ROI.
Alignment with systems and processes: Design pilots that integrate with existing systems. Building something that’s both robust and repeatable allows you to scale.
Human integration: Define the role of human oversight and ensure your team has the skills to collaborate effectively with agents.
Visible impact: Prioritize projects that build organizational confidence by engaging sponsors and stakeholders early, establishing governance and long-term support.
When you start seeing positive outcomes, scale thoughtfully. For example, a coding agent could evolve into a full orchestrator-worker model for broader system updates, or a support assistant could expand to handle billing and dynamic routing.
Lastly, make sure to manage expectations and allocate resources appropriately. Communicate with executives leading agent initiatives to set expectations regarding both timelines and results. Ensure sufficient scope for system and team integration while helping the rest of the organization understand that piloting new technologies may take longer than established methods. This approach positions your organization to unlock AI’s full potential while maintaining focus on measurable success.
Take the first step
AI agents represent a significant opportunity for business transformation, but success requires an approach that carefully balances practical value creation with managing complexity and risk. By starting with well-defined use cases and building systematic capabilities, you can capture value from AI agents today while positioning your organization for broader transformation tomorrow.
