The agentic AI starter kit

Santo Orlando at Insight outlines the five fundamental steps to get started with agentic AI

 

Most of us are familiar with generative AI by now. In fact, plenty of organisations have already embraced the technology for tasks such as content creation, code generation and even customer support. But generative AI is just the beginning. The technology merely scratches the surface of what artificial intelligence can truly do.

 

Enter agentic AI. 

 

Unlike generative AI, which relies on prompts and human input, agentic AI can act autonomously to achieve complex goals without significant human intervention. As a result, nearly 45% of business leaders think agentic AI will outpace generative AI in terms of impact, and more than 90% expect to adopt it even faster than they did with generative AI. 

 

However, in spite of its promise, our collective understanding of agentic AI - and how to implement it - is still very much in its infancy. 

 

So, where do you start? Here are five foundational steps to help you kick off your agentic AI journey.

 

 

Agentic vs. generative AI

If generative AI is like having a personal assistant, helping you one-on-one to speed up your tasks, then agentic AI is more like having a team of smart, independent coworkers who can take initiative and get things done across your business - without needing constant oversight. 

 

One powerful example of this in action is in sales. With agentic AI, organisations are able to receive real-time insights during discovery calls. The AI ‘agents’ allow sales reps to respond with timely, relevant information, helping them build trust, operate faster and close deals more effectively.

 

By collecting and analysing data from across teams, agents can uncover patterns, translate complex metrics into actionable strategies and even highlight opportunities that might otherwise be missed. In some early implementations, sales teams have reported saving five to ten hours per rep each month,  adding up to thousands of hours redirected toward deeper customer engagement.

 

The one-to-one relationship we’ve grown accustomed to with generative AI has evolved into the one-to-many dynamic of agentic AI, which is capable of handling tasks for multiple users and automating entire business processes. Even more impressively, agents can make decisions, manipulate data and take actions on their own - a capability that can seem daunting without a clear understanding of how it works.

 

That’s why businesses need to start small, and here are a few practical steps to get going quickly - and wisely - with agentic AI. 

 

 

Step 1: Prepare your data

Agentic AI is the logical progression for organisations already exploring generative tools. However, the data needs to be in an optimal condition - clean, organised and secure - before autonomous agents can be deployed effectively.

 

As such, eliminating redundant, outdated and trivial (ROT) data is vital. Without removing ROT, agents may rely on obsolete information, leading to inaccurate or misleading outputs. For example, this could happen if a company deploys an HR chatbot that’s connected to outdated data sources. If an employee were to ask about their 2025 benefits, the chatbot might pull information from as far back as 2017, resulting in confusion and misinformation.

 

Proper file labelling, standardised document practices and use of version histories in place of multiple saved versions help to ensure agents access only the most relevant and accurate information.

 

 

Step 2: Begin with low-risk cases 

Agents work on a transactional basis, charging for each operation, which can quickly add up. As such, it’s wise to experiment with simple, low-stakes applications first. This approach allows for quicker deployment and demonstrates immediate value to the business without significant costs or risks.

 

One example could be using an agent to assess sentiment in social media responses following a product launch. This can offer real-time feedback on public perception and inform messaging strategies. Other low-risk use cases include generating reactive press releases and monitoring competitor websites. Additionally, prioritising automation of routine tasks, especially those involving platforms like Salesforce, SharePoint, or Microsoft 365, allows teams to maximise impact without costly system overhauls. 

 

Overall, organisations need to be willing to fail fast and expect failure. It won’t be perfect from the start. However, an experimental pilot approach helps to efficiently refine AI agents, reducing the risk of costly mistakes and making sure that only effective solutions are scaled up.

 

 

Step 3: Create a centre of excellence 

Establishing a dedicated, cross-functional team to explore agentic AI use cases helps prevent siloed adoption and supports enterprise-wide visibility. This team should span as much of the organisation as possible and include representatives from departments such as marketing, finance and technical solutions.

 

Collaborative workshops can then act as a forum to identify key processes that would benefit from autonomous capabilities and help businesses align potential applications with specific departmental objectives and broader business goals.

 

 

Step 4: Education is key 

Many companies underestimated the importance of training and governance with generative AI - and agentic AI is no different. Organisations need to establish clear governance to define how AI agents should and shouldn’t be used, covering not just technical implications, but HR, compliance and risk concerns as well.

 

Equally, businesses and those employed must understand agentic AI’s full functionality to get the most out of it. Like with almost all technical training, AI education cannot be viewed as a one-time ‘tick-box’ exercise. Ongoing learning is necessary to keep pace with new capabilities and best practices.

 

For example, consider what’s already emerging, like security agents that automate high-volume threat protection and identity management tasks; sales agents that find leads, reach out to customers and set up meetings; and reasoning agents that transform vast amounts of data into strategic business insights.   

 

 

Step 5: Measure ROI 

Enthusiasm around agentic AI is high. But before organisations dive in headfirst, it’s important they first define success. Technology can’t be the solution if there is uncertainty surrounding the goal. Successful deployment requires a clear definition of the problem organisations are looking to solve and knowledge of how to align the solution with measurable business value. Without this, initiatives risk stalling at the experimental stage.

 

Key performance indicators should also be identified early. These may include increased productivity, time savings, cost reduction or improved decision-making. Establishing these benchmarks and taking a data-driven approach ensures that AI initiatives align with business goals and demonstrate tangible benefits to stakeholders.

 

 

The path to greater value with AI

The journey to adopting agentic AI is not just about integrating advanced technology but transforming how organisations approach common challenges with broad impacts. The path to greater value lies in thoughtful, measured adoption and the willingness to iterate and learn along the way. Ultimately, it’s the small, first steps and simple use cases that will get you further with AI.  

 

---

 

Santo Orlando is Practice Director - App, Data and AI Services at Insight

 

Main image courtesy of iStockPhoto.com and MF3d