AI is changing every day, and the toolsets businesses rely on…
Summary
Many organizations default to chatbots when trying to “do AI,” but this narrow focus overlooks other powerful AI capabilities that can better support real, complex business needs. This post highlights several types of AI systems, including conversational tools for 24/7 information access, agentic systems that automate multi-step tasks, multimodal systems that interpret many kinds of data, and emerging embodied AI used in physical or simulated environments. It emphasizes that organizations often benefit most from hybrid approaches that blend these capabilities, and encourages leaders to choose AI solutions based on specific use cases and goals rather than relying solely on chatbots.
As teams are being pushed by leadership and Boards to “do AI,” many simply invest in a business plan of an enterprise chatbot—like Google Gemini, Microsoft Copilot, or ChatGPT or Claude—and release it in the wild for their teams to use. Consequently, chatbots became the default face of AI solutions. Compounding this mindset, many of these technologies are now embedded in our toolsets, such as how Gemini is now in Google Docs or Copilot is in Microsoft products.
But as new capabilities develop and different interfaces become commercially viable, focusing on chatbots to “do AI” leaves a lot of value behind. Real work is often highly specific, nonlinear, and data-heavy, which might be better served in other interfaces depending on your organizational pain points and challenges. As managers, committees, and leadership teams push to integrate AI into their processes, it’s important to consider not only the business value of chatbots, but other implementation types of AI.
Conversational AI Systems: Chatbots for 24/7 Access
Conversational AI systems interact with humans by communicating with natural language in text forms. The most recognizable form of conversational AI is the trusty chatbot, like the ones directly plugged into your toolsets or available online when you visit a website. These conversational systems are best when you want to provide 24/7 access to information, like enabling your users to self-serve frequently asked questions. Conversational AI systems work by interpreting a user’s request, usually sent through a typed query in a chat box. The system then responds in an output like text (a chatbot), a voice assistant (thanks, Siri), or a phone response system like an IVR.
These systems are especially powerful in use cases where you need to provide or gain access to information at any time in a way that sounds human, not robotic. For example, virtual assistants can answer queries with real-time information on a company’s public website or your employer’s intranet site. Or, organizations may rely on conversational AI systems to gather feedback about a recent experience, like a patient’s visit to the doctor’s office. Organizations use conversational AI systems to provide immediate and always-on support and response, which is becoming ever-more important as consumers demand an immediate, standout digital experience.
Agentic AI Systems: Automate Tasks
2025 has been called “the year of agentic AI,” according to industry experts like Nvidia’s CEO. Agentic AI systems automate tasks by making decisions and taking actions that require several steps. Researchers at Stanford define agents as “goal-directed systems equipped with tool access and multi-step execution capabilities.” In other words, organizations can build agentic systems that tap into their enterprise toolsets and perform multiple actions on behalf of human workers. The same Stanford study says that people are most interested in agentic automation because it frees them for more high-value work, eliminates task repetitiveness and stressfulness, and finds areas for quality improvement.
Organizations can take advantage of pre-built agentic AI systems to automate tasks, such as Hubspot Breeze agents that research and connect with prospects, or Workato Genies to support team-based workflows like recruiting or incident management. Users can also custom-build their own agents, either in their enterprise tools (both Hubspot and Workato allow you to develop your own agents in their platforms) or their own engineering capacities. In either case, agents begin a task based on a user request, a scheduled run, or some other kind of triggered event. The agent then executes that task, possibly by choosing which tools to use and calling on them to do the work. Tools might include a data source (like logs or files), an action system (like a ticketing or email system), and a validation tool, (like testing). Then, you can see some kind of output.
Agentic systems can be built deterministically, meaning that you can specifically tell agents how to accomplish tasks, which tools to call and when, and the method the task is accomplished. Generative agentic systems learn their own approach to performing tasks, meaning that they plan steps, call tools, and check their own work completely on their own. Some organizations may feel that this “black box” method of doing work—not knowing exactly how the work is getting done—is more risk than they’re willing to take and would prefer the transparency of a more deterministic system. (Zirous helps teams understand the different trade-offs and priorities for your system implementation—more about our workshop later).
Multimodal AI Systems: Take Advantage of All Data Types
Much like the name suggests, multimodal AI handles multiple different types of data to create a broad range of outputs. These systems make sense of text, numbers, charts, images, audio recordings, video, and other kinds of data inputs to deliver unique insights, capture the whole story, and deliver actionable recommendations. Multimodal systems shine when you need to combine multiple kinds of data to create richer, deeper understanding of different areas in your business.
Multimodal systems are ideal for use cases that not only need to take in data sources from different kinds of inputs, but also deliver them in multiple modalities. For example, organizations can improve predictive maintenance on their physical products with multimodal systems. The system could monitor live sensor readings like temperature readouts, thermal images, and weather data (which might be in numerals, heatmaps, and imagery), and review maintenance work order history. It ingests this information, then forecasts when a component is likely to fail and recommend a part order. Or, perhaps you want to improve your customer service call center. The multimodal system could identify caller sentiment from audio clips, review associated keystroke patterns from your call center agent, and understand a screenshot or picture of an error. Over time, the system would provide insights about what kinds of issues require a more specialized or adaptive type of service.
Bonus: Embodied AI Systems
Although we are still in the very early days of embodied AI, the concept means putting artificial intelligence into physical systems or virtual environments. As of now, embodied AI is mostly synonymous with technologies like autonomous vehicles or warehouse robots like the ones in Amazon’s fulfillment centers, but can also be embedded in simulated environments or digital twins. For example, Iowa State University researchers are currently exploring ways to create realistic emergency management training content for emergency forces to practice and improve training responses in an online “game” while incorporating generative AI content.
Hybrid AI Systems
Although organizations can tap into these systems individually to achieve their values, these different kinds of AI capabilities don’t always exist in a silo. In practice, systems may blend elements of conversational, agentic, and multimodal depending on your use case and desired goals. For example, your sales user asks an agentic system to draft an outreach email to a new lead. The system taps into Salesforce activity and heatmap trends on the company’s website to understand the user’s journey, and then finds the lead is active online and scrolls through their blog or listens to a podcast excerpt. The system uses all these different modalities of information to draft an introductory email, and sends it to the sales expert for review. The expert approves the email, and the agent schedules it to send to the lead based on ideal open rates and time zone characteristics.
The kind of AI implementation that makes sense for your organization needs to be rooted in your use case and needs. You might have conversational features that give you 24/7 access to information, agentic capabilities that automates certain tasks for you, multimodal experiences that help you take advantage of all the different types of information within your organization, or even embodied AI capabilities to enable safer, more realistic scenarios.
If you’re interested in learning more about different interfaces for AI capabilities, how to uncover your readiness, and the trade-offs and priorities to consider when exploring different systems, reach out to us about our workshop AI & Appetizers. Our 90-minute workshop gives business leaders AI clarity separated from AI hype to build alignment within your organization.
This Post Has 0 Comments