GenAI is Finally Boring (in a Good Way); Agentic Systems are the Next Big Thing

Since the launch of ChatGPT, Generative AI (GenAI) has upended much of how we think about content creation and interaction with customers. But like any "hot" technology, the "newness" wears off and we settle into a (reasonably) reliable and predictable technology. With GenAI, and that's not a bad thing, for two reasons.

First, the hot and new properties of GenAI have introduced a genuine appetite for change. In a little over one year, organizations have gone from "let's see how this works" to "we need to make this work for us ASAP." And second, GenAI has enabled the next new wave, which will enable a much larger change in how technology works: agentic systems.

What are  Agentic Systems? An Introduction

Agentic systems are advanced forms of AI that autonomously execute multi-step tasks, adapt based on real-time feedback, and make decisions without human intervention. They're not the often-touted "Artificial General Intelligence," but they are much more than a simple chatbot.

Agentic systems are not simply reactive; they are designed to:

• Perform Multi-step Tasks: Manage intricate workflows that involve numerous interdependent steps.
• Adapt and Iterate: Modify their strategies based on the outcomes of their actions to optimize future performance.
• Integrate Continuous Feedback: Solicit human feedback when necessary and incorporate it into their decision-making processes to enhance collaboration.

Agentic systems represent a shift from basic task automation to comprehensive, decision-making capabilities that can drive entire workflows. 

The Potential Impact of Agentic Systems 

By handling repetitive, high-stakes tasks, agentic systems can squeeze out human error. Human teams can spend less time generating boilerplate and fixing their prior (natural)  mistakes and focus on work that leverages their unique strengths.

Here's how this can work.  Think about a system embedded within a software development pipeline that autonomously manages the entire workflow:

Initial Code Creation 

In this first phase, the agent translates high-level software requirements into an initial set of code. This can include multiple files and multiple formats. In this scenario, we're assuming a generative AI system such as a Large Language Model (LLM) does the code generation, formatting the response in such a way that the agent can break it down into its component pieces and save and format them properly. We already see this on a smaller scale with tools such as Github Copilot.

Test Creation and Execution 

Now it's time to test the code. The LLM uses the code generated in the previous step, along with the initial requirements, to create unit tests and integration tests with tools such as Python's unittest module or Testify for Go, which the agent then saves and runs against the code created in the previous step.

Iterative Refinement

Just like with a human coder, it's unlikely that all the code will work perfectly and satisfy all requirements the first time around. The agent will run tests, note any failures, then request improvements on the code, which will then be implemented. Those improvements will also lead to new tests, which the agent will execute against the new codebase. This step can run a set number of times, or until no more errors are detected. 

Human Feedback Integration: 

The agentic system is not a replacement for human coders or stakeholders. Specific "breakpoints" can be implemented for coders and/or stakeholders to view progress or approve moving on to the next step in the process. In addition, the agent can refer problems the LLM can't solve back to human coders.

Auxiliary Task Management: 

The agentic system can also automate other tasks, such as GitOps management using Github Actions, assignment of code reviews, summarizing changes for reviewers, and working with static code analysis tools such as SonarQube or Snyk Code to recommend security or performance changes the LLM didn't recognize on its own.

Agentic systems can learn from each cycle of work, adapting their approach to optimize outcomes. This dynamic capability allows businesses to remain agile in rapidly changing environments.

‍

How "Disruptive" Will Agentic Systems Be?

Perhaps even more than GenAI, agentic systems have the potential to affect a large number of different labor markets.

Automation has always eliminated lower level roles. Already Robotic Process Automation (RPA) has done away with many of the more "drudgery" type of tasks such as moving files and filling in forms. Agentic systems extend that risk to more complex, skilled work. However, the majority of cases may see job transformation rather than elimination such as:

• Shift in Work Focus: As agentic systems automate routine tasks, workers can focus on more creative tasks that machines can't replicate (yet).
• New Skill Demands: Emphasize problem-solving, creativity, and AI collaboration skills.

The reality is that workers will need greater digital literacy, adaptability, and problem-solving skills to remain relevant. Organizations that want to hold onto these workers should provide continuous training and promote a "growth" mindset to help employees transition into roles that partner with agentic systems rather than fighting against them.

Still, some jobs will inevitably be lost. That makes it essential for those affected to upskill and prepare for new roles. Programs that teach AI fundamentals, critical thinking, and interpersonal skills will be valuable.

Preparing for an Agentic World: Actionable Steps

The process of preparing to take advantage of agentic systems is straightforward. It involves:

• Assessing your current position
• Building out your infrastructure
• Enhancing your data pipelines and integrate AI models
• Upskilling for advanced automation
• Implementing pilot projects
• Establishing a culture of continuous improvement
• Building partnerships with other organizations
• Creating an ethical and compliance framework

Let's look at each of these steps.

Assess Your Current Tech Stack and Identify Opportunities

Before beginning to investigate how you can integrate agentic systems into your organization, you'll want to assess your current infrastructure to identify which parts of the development and operations pipeline can benefit most from automation. For example:

• Audit Current Tools: Review existing CI/CD tools such as Jenkins, GitLab CI/CD, or GitHub Actions to determine how agentic systems can enhance or possibly replace specific workflows.
• Workflow Mapping: Use tools such as Terraform or Ansible for infrastructure as code (IaC) to document current deployment processes. This helps visualize where agentic systems might be best suited to take over repetitive tasks or manage complex multi-step procedures.
• Process Identification: Start by identifying repetitive or high-value tasks, such as automated testing and environment provisioning, where tools like Selenium, Puppeteer, or Cypress can be first complemented, then perhaps replaced, by agentic systems capable of autonomous decision-making.

Building out a Scalable and Resilient Infrastructure

Adopting agentic systems requires infrastructure that can handle dynamic workloads and real-time data processing.

• Leverage Containerization: Use Docker and/or Kubernetes to build a scalable environment capable of running agentic system components in isolated and efficient containers.
• Serverless Frameworks: Incorporate serverless technologies like AWS Lambda or Azure Functions for event-driven processes that complement agentic automation, enabling lightweight, scalable execution of functions.
• Data Management Tools: Ensure data quality with tools like Apache Kafka or RabbitMQ for reliable message queuing and data streaming. These will support real-time data feeds that agentic systems rely on for immediate response and adaptation.

Enhancing Data Pipelines and AI Model Integration

High-functioning agentic systems require access to well-managed and high-quality data pipelines to make effective decisions and adapt to new inputs. Some things to consider:

• Data Engineering: Use tools such as Apache Spark or Databricks to efficiently process large data sets. These tools can handle the real-time data analytics that agentic systems use for learning and optimization.
• AI Model Deployment: Integrate machine learning frameworks like TensorFlow Serving or MLflow to deploy and manage AI models that feed into agentic systems. This ensures that models are accessible, scalable, and continuously updated for accuracy.
• Version Control for Data: Implement data versioning with tools like DVC (Data Version Control) to maintain consistency and traceability of the data fed into agentic systems.

Upskilling for Advanced Automation and AI Interaction

To get the most out of agentic systems, developers and operators must enhance their skills in AI operations and advanced automation.

• AI Operations Training: Learn how to manage and monitor AI models using Kubeflow or Azure Machine Learning for MLOps in order to bridge the gap between model development and production use.
• Advanced Scripting and Programming: Strengthen proficiency in Python, Go, or Rust for building components of agentic systems that can interface with existing operations tools.
• Ethical AI Practices: Learn about AI ethics and data privacy, focusing on how frameworks like Hugging Face’s Transformers or Google’s TensorFlow Privacy can be used to build systems that comply with data protection standards in real time rather than after the fact. 

Implementing Pilot Projects for Incremental Adoption

You don't have to dive into this pool all at once. Start by integrating agentic systems in controlled, incremental steps to gather data, identify challenges, and adapt strategies.

• Pilot Automation Projects: Use tools such as Airflow to orchestrate data workflows that you can later automate with agentic systems. Start by automating code testing and branching logic and see how the system handles these tasks.
• Monitoring and Feedback Mechanisms: Implement observability tools like Prometheus and Grafana for monitoring the performance of agentic workflows. These tools help gather real-time metrics and identify performance bottlenecks or areas needing improvement.
• Feedback Loops: Integrate feedback systems with Slack API or Microsoft Teams Bots for developers and stakeholders to provide real-time input on agentic system performance and behavior.

Establishing a Culture of Adaptability and Continuous Improvement

Integrating agentic systems isn't a one-time shift. It's an ongoing process that requires you to have a culture focused on learning and adaptation. Think about:

• Adopting Continuous Integration of AI: Build on CI/CD practices by incorporating CI/CD for ML models (CI/CD/ML) using tools like GitLab or Jenkins X. This gives you seamless updates and integration of new agentic capabilities into production environments.
• Fostering Collaborative Learning: Organize internal hackathons or workshops focused on agentic system use cases and development. Use platforms like Jupyter Notebooks for collaborative experimentation and documentation.
• Supporting Open-Source Contributions: Encourage team participation in open-source projects related to agentic systems, such as OpenAI Codex or AutoML frameworks, to support a culture of continuous learning and exposure to cutting-edge technologies.

Partnering with Industry Leaders and Building Alliances

Collaboration with industry leaders can accelerate the adoption and enhancement of agentic systems.

• Technology Partnerships: Partner with cloud providers like AWS, Google Cloud, or Microsoft Azure for access to their latest agentic automation tools and training resources, or with providers such as CloudGeometry to help build out your agentic systems goals.
• Consortia Participation: Join AI and technology consortia like AI Ethics Consortiums or ML Commons to stay informed on best practices, ethical guidelines, and the latest research in agentic systems.
• Leverage SaaS Solutions: Explore SaaS platforms that provide agentic capabilities, such as DataRobot for automated machine learning or UiPath for robotic process automation enhanced by AI.

Ensuring Ethical and Compliant Deployment

Ethical and regulatory concerns are critical in the deployment of agentic systems.

• Implement Transparent AI Practices: Use tools such as Explainable AI (XAI) frameworks from IBM Watson or Google Cloud AI to make sure that decision-making processes are understandable and transparent.
• Data Security and Privacy: Strengthen data protection practices using encryption and privacy tools like AWS KMS (Key Management Service) or Azure Information Protection to safeguard data handled by agentic systems.
• Compliance with Standards: Make sure that systems comply with relevant standards like GDPR or CCPA, using compliance management tools like OneTrust or TrustArc to monitor adherence.

What it all comes down to is that experienced developers and operators preparing for agentic systems need to focus on planning, technical upskilling, and a culture of continuous improvement. Many of the tools you need to make this happen are already out there, so by leveraging existing tools and best practices, investing in scalable infrastructure, and ensuring ethical compliance, you can make a seamless transition into agentic systems as they stand today, and be ready as they get better.

‍

Conclusion: Prepare for a Future enabled by Agentic Systems

The evolution from GenAI to agentic systems marks a new era in AI, one that extends beyond simple content generation to fully autonomous process management. While agentic systems will disrupt the job market, they will also transform the nature of work, emphasizing strategic, creative, and complex problem-solving skills. To avoid getting left behind, you must take action now—assessing capabilities, building infrastructure, and upskilling teams—to effectively integrate agentic systems in the short and long term. This dual focus will let you take a balanced approach that leverages the technology while still empowering human workers.

Feeling overwhelmed by risks and the potential?  Call CloudGeometry and we'll help you find your footing.

‍