This is a basic rundown for those who arenโt familiar with the tech as a whole, so if youโre already ahead in the knowhow, my last few segments around this techโs impact might be more valuable than this first half!
AI Agents, what are they? The Googleable Stuff ๐๐:
A fast and loose definition of an AI Agent:
Basically an Agent is a virtual robot (yes, you read that right) that has a prescribed task and performs it using the tools itโs connected to, and intaking information based on its sensors. (Source)
Thanks Wikipedia! Hereโs what the overall architecture of a Model-based Agent looks like.
Originally from Stuart J. Russell and Peter Norvig, Artificial Intelligence: A Modern Approach p.51These Agents exist within an environment not too different from a hypervisor with virtual machines. Can I plug Agents into other Agents to have larger chains of processes? Yes, and many people are already experimenting with these.
The use case for AI Agents:
Having a digital personal assistant. This is great for tasks that are complex but repetitive, that involve multiple systems, and have a relatively constrained area of focus, and that are software boundโฆ for now. So far, the Cloudโs the limit.. Which is to say anything is possible with enough integrations!
Comparing Agents with some Existing Technologies:
AI Agents vs. Robotic Process Automation (RPA):
RPA: Automates rule-based, repetitive tasks within software environments, can play nicely with some amounts of APIs and webhooks, great for basics of data entry or invoice processing.
AI Agents: They go further. Automated context-driven decision-making means each agent has a LOT more flexibility when adapting to changes, and operating across diverse environments.
AI Agents vs. Legacy Middleware Systems:
Legacy Systems: Often static, requiring extensive coding and configuration for each new feature or task.
AI Agents: Act as dynamic middleware, wrapping around legacy systems to streamline operations without requiring direct modifications.
AI Agents vs. Chatbots:
Chatbots: Limited to conversational tasks and guided flows, often restricted to specific domains.
AI Agents: Can take on broader, more complex tasks involving data aggregation, decision-making, and integration with external systems.
OPPORTUNITY: Having a chatbot as the intake/linchpin for a larger chain of AI Agents is where I see some real fun start to take shape.
Why Are AI Agents Such A Shakeup For SaaS? ๐ตโ๐ซ
Imagine an employee that handles one particular microtask at any point in time. Once chained hand-in-hand (and negotiating a few guardrails) your micro-tasked โemployeesโ can tackle a larger, harder task that is otherwise needing a human brain to do. AKA, you now have a small, headless workforce to (hopefully) complement your human one.
This headless workforce can perform tasks that would otherwise have to be explicitly described by developers in, and defined like traditional software via code, and integrations, but now, youโve got some leeway to go further, more easily.
This means that software-as-a-service takes on a new identity:
One where SaaS becomes primarily user oriented to simply capture user inputs. The rest is handled by software, rather than having the user page through many screens to control every aspect of the task at hand, this is much more ambiguous, and flexible than what raw code can reasonably handle.
So to interact with this kind of service, there are at times, still some steps that require UIs, and such SaaS is not completely dead, just the ingestion & output viewing screen of what is otherwise a complex, multi-task service operated by software.
This has the potential to be the ultimate middleware between legacy enterprise systems and newer means of development.
e.g. Wrap a chain of agents in an API. ping the API when you need interactions made with older systems without needing more than humanly understandable documentation about the legacy system.
Envisioning an AI Agentic Future ๐ฎ
Dipping into futurology for a moment, an Agentic future does represent a shrinking demand in administrative tasks being handled by humans.
An โAgent overseerโ job could very well become part of the new normal. Knowing the work that needs doing, and having the means of confirming that the work is done well. Being able to perform quality Assurance + improvements to the systems in-house is also a critical
Ai Voice assistants can go much further by integrating with a few other things.
This is also opening the potential for AI Personal Assistants to handle admin work rather than
Experimentally, this might represent a new (energy intensive) way to code. By building a jobs roster, and descriptions, one can put a handful of bots to work.
This can mean that SaaS takes on a whole new pace of speed in turning around fast features. Not energy efficient ones, but flexible features nonetheless.
A Somewhat Rocky Road Ahead For Adoption โฐ๏ธ
There are some avoidable issues with AI Agents that I see coming in my humble technical opinion, so Iโve listed off a few here that are very likely to be part of the conversation. This isnโt to say that itโs doom and gloom, but rather that the conversation is multi-faceted, and as actors in the space, we should be able to discuss the less glamorous side of things as well.
Scalability Issues ๐
While AI Agents can perform tasks effectively, scaling their usage across large organizations can present challenges. Think along the lines of compute costs, integration complexity with existing systems, and ensuring consistent performance as the network of agents grows.Energy Consumption โก๐
The AI models powering agents are very resource-intensive, thereโs no way to deny that. A key piece to the sustainable staying power of this tech will be in finding wins in processing optimization, model optimization and in hardware optimization.โGarbage In-Garbage Outโ continues to hold true ๐๏ธ
Some data of doubtful quality/bias will definitely mar agentsโ ability to make optimal decisions. Itโs really no surprise to you, dear readers, that good data management practices are still expected with this tech.Quick tips on data management from across the internet:
User Trust and Acceptance ๐ค
While many may be hesitant to trust AI Agents for critical tasks, it will be the implementer's responsibility to provide whatever amount of transparency they can on their agentโs behaviors- read: ACCESSIBLE TRANSPARENCY (Enough with the barely readable system logs; your users are mostly non-developers/non-technical! We are in 2025 darn it!)
Transparency carries into Debugging and Oversight!
When chained processes fail, diagnosing the root cause in an interconnected system of AI Agents is likely to be difficult without clear, accessible, means of supervision.so having tooling to do this kind of monitoring is a must.
The elephant in the room ๐ - Ethical Concerns For Human Jobs
As is partially in motion with some industries, radical motions towardsan over-reliance on AI Agents will lead to job displacement and widen the digital divide. As is best practice for most AI use in todayโs world, I preach that developers aim for an accelerating amount of augmentation of human capacity!
Wrapping this Up
Now you know just enough to talk about AI Agents over a coffee. Thereโs still lots to say on the topic, whether itโs discussing the technical, social, or other impacts this technology can have onย entire industries at large.
While the future around AI Agents hasnโt been completely mapped out, this is still something that many are considering tech worth experimenting with. More to come in the AI Agent direction!
(Ripped straight from my last Thought Byte) When new foundational technologies come out, don't get roped up into the hype and blindly shove it into everything. Instead, use this "weird" time to re-evaluate how you deliver value to your customers.
After understanding its limitations, ask yourself the following: How can new technologies like AI Agents help you create a whole new paradigms for user interactions?
Let's move beyond the "shiny new wrench" mentality and start thinking on a grander scale. Consider the potential to disrupt new technologies, and how they can be harnessed to create truly groundbreaking user experiences.
The future continues to be weird, yes, but it's also brimming with possibilities. So buckle up for 2025, and get ready for another wild ride!
Yee Haw ๐ค
Sam from The AI Product Report
Want to talk about this longer? Need more customized help on the matter? Emailย me [email protected] donโt be shy, letโs talk.
I know there are a LOT of other AI-Product topics to cover like feedback loops and ethics, so let me know if thatโs something you want to see discussed!
Hereโs an anonymous channel for you to send me your thoughts if the comments section isnโt for you! ๐