PODCAST: Commentary of some recent developments on AI Agents as of July 13, 2025
AI agents can be applied across various enterprise functions, ranging from simple to highly complex tasks, serving as a powerful tool to streamline operations and enhance efficiency. The decision to implement AI agents should begin with identifying processes within a business that could benefit from extra assistance.
Here’s an analysis of enterprise applications for AI agents:
General Enterprise Use Cases:
- AI agents can autonomously perform tasks across procurement, sales, and customer interactions.
- They can streamline simple tasks, such as sending messages.
- AI agents are envisioned as “digital employees” that can automate and manage many low-touch or low-complexity interactions within an organization’s workflows.
Specific Applications:
- Software Development and IT Operations (Google Code Assist Agents): Google’s Gemini Code Assist introduces agent capabilities designed to help with tasks across the software lifecycle. These include:
- Generating new software by creating applications from product specifications written in Google Docs.
- Migrating code, which involves transforming code from one language to another or translating code between languages and frameworks. While effective for single functions, switching entire frameworks might be challenging.
- Implementing new features from GitHub, where the AI reads “issues” (bug reports or feature requests) and writes code to implement them, such as adding two-factor authentication.
- Performing code reviews, identifying potential faults or areas for improvement in code.
- Generating tests, including both positive and negative test cases, to confirm code functionality and increase reliability.
- Performing AI model testing by creating validation tests to assess how an AI model behaves with certain inputs and outputs, particularly useful for content safety and AI subsystem performance.
- Creating documentation using a built-in wiki-builder, providing a strong starting point for programmers.
- Microsoft Copilot Features (Agentic Capabilities): Microsoft’s Copilot has introduced several features that demonstrate agentic capabilities:
- Actions: Allows Copilot to take action on a user’s behalf, such as booking event trips or reservations through partnerships with travel and dining platforms like Booking.com, Expedia, and OpenTable.
- Copilot Vision: Provides real-time assistance by viewing and understanding the context of what a user is doing online, available on mobile (using the phone’s camera) and Windows (for screen assistance like changing settings or organizing files).
- Podcasts: Enables Copilot to create AI-powered podcasts based on user-provided content (like a research study) or information pulled from the web, allowing conversational elaboration on topics.
- Deep Research: Similar to capabilities in Gemini and ChatGPT, Copilot Research can conduct complex research by searching numerous online sources and compiling results into an organized document, saving significant time.
- Pages: Functions as a co-editor within a text editor interface, collaborating with users to analyze, revise, polish, and organize text.
- Shopping: Acts as a personal shopping assistant, helping users find products, providing buying advice through comparisons and research, alerting about price drops, and enabling direct purchases.
- Copilot Search in Bing: Blends traditional search with generative AI to provide conversational answers pulled from the web in real-time, with footnotes for source verification.
Spectrum of Complexity: AI agent applications exist on a spectrum, with some being simple and “grounded in one or two files,” while others are more advanced. It’s recommended to start with use cases that matter and can have a greater impact, even if they are on the simpler end.
Considerations for Implementation: Before implementing AI agents, a critical step is assessing the hygiene of a company’s data infrastructure. Poorly organized or “garbage” data can lead to “garbage outcomes” from even the best algorithms. Thinking of deploying an AI agent as “hiring an employee” can help ensure proper diligence in training and deployment, avoiding common mistakes of expecting immediate positive transformation without careful deliberation.
The sources distinguish an AI agent from a chatbot primarily by their ability to perform tasks autonomously and reason independently, rather than just providing information or predefined responses.
Here are the key distinctions:
- AI Chatbot Capabilities:
- An AI chatbot provides users with solutions to questions by referring to its training data.
- It can only perform tasks it was predetermined to tackle.
- A chatbot is fundamentally a “dialogue system” for conversation.
- For example, if you ask a chatbot what to order for dinner, it can generate a list or, in more advanced cases, place the order.
- AI Agent Capabilities:
- An AI agent is a software tool that can perform tasks autonomously for you.
- Unlike chatbots, AI agents can execute tasks on your behalf and identify when they are needed, rather than just feeding you an answer from a predetermined subset of information.
- They use a combination of human-delivered instructions, environmental triggers, and, crucially, their own reasoning capabilities to deduce what needs to be done and when.
- AI agents are “goal-seeking” and “have the smarts to think about what is needed to perform the task” and “figure out what tools are needed”.
- They can use the context from information around them to make their own inferences and perform tasks, which requires a higher level of reasoning.
- As Srini Iragavarapu from AWS explains, AI agents are not only having conversations but are “actually doing things on your behalf as well”.
- The amount of assistance an AI agent can provide for individuals and businesses is significantly higher than that of a chatbot.
- To illustrate, while a chatbot might list dinner orders, it would not have the analytical capability to check your smart home inventory (like what’s in your fridge), analyze dietary preferences from your health app, or coordinate with your calendar to know when you’ll be home to receive an order. An AI agent could potentially integrate these contexts to make more informed decisions.
- AI agents leverage various AI capabilities, including large language models (LLMs), reasoning, natural language processing, and long context windows, to perform tasks more intelligently and adaptively. They understand the context of an interaction, customer need, or business transaction, and then apply automation to manage the associated workflow.
In essence, while a chatbot talks and provides information based on its training, an AI agent acts and can infer, reason, and execute multi-step tasks independently based on broader context and goals.
Adobe Firefly offers several practical use cases, primarily focused on generating and enhancing visual content for marketing and business owners, while ensuring commercial safety.
Here are some key applications:
- Text to Image Generation: You can create images from a simple written prompt. For instance, if you’re launching a product and need a lifestyle image with a specific mood or color palette, you can describe it to Firefly instead of hiring a photographer or searching stock sites.
- Generative Fill: This feature allows you to remove or add objects to an image. An example provided is removing an ex from a photo.
- Text Effects: Firefly enables you to apply various styles and textures to text using a prompt, such as metallic, neon, stone, or watercolor looks, to make plain text visually striking without extensive manual layering.
- Image to Video Conversion: A still image can be transformed into a short animated clip, which is useful for marketing placements like Instagram Reels, YouTube Shorts, or TikTok, repurposing static ads into dynamic content.
- Text to Video Generation: You can start with a prompt, and Firefly will generate a video clip based on your description, suitable for explainer content or product introductions.
- Video Translation: This powerful feature allows you to translate an existing video into another language, such as English to Spanish or French, potentially opening up new markets and revenue streams.
Firefly integrates into Adobe Creative Cloud products like Photoshop, Illustrator, and Premiere Pro, allowing users to generate images, text effects, vector art, and video enhancements, thereby speeding up the creative process. A significant advantage of Firefly is that all its generated content is based on licensed, commercially safe data, drawn from Adobe Stock and public domain material where copyright has expired, addressing concerns about copyright infringement and legal issues.