Everyone is talking about AI agents and copilots — their capabilities, use cases, risks, and more. Many say agents are the future of how we engage with computers. I want to give an overview of what agents are from my AI startup perspective.
For context, I’m the co-founder of Sweet, which is applying AI to make tax return preparation easier. An agentic, “virtual CPA” experience that engages the preparers’ clients and processes documents is core to our product.
I’ll start with definitions. Agents are AI-based systems that interact with their environment to complete tasks autonomously. Today’s AI agents and copilots are a step ahead of simple automations like those associated with Robotic Process Automation (RPA). Copilots are usually used in similar ways as agents, but originally the copilot concept came from Github and Microsoft, which have been releasing many Copilots. Copilots generally don’t have the same level of autonomy as agents.
The word copilot itself comes from the world of aviation: the copilot isn’t the pilot, in this case that’s a human. The copilot sit next to the pilot, assisting the human who is flying the plane, acting as a force multiplier. The key features of modern agents and copilots are their level of flexibility, autonomy, complexity, context understanding, integrations, and ability for causal analysis. They can break tasks into subtasks; they can reason, plan, and delegate.
AI Copilot Summit NA is an AI-first event to define the opportunities, impact, and outcomes possible with Microsoft Copilot for mid-market & enterprise companies. Register now to attend AI Copilot Summit in San Diego, CA from March 17-19, 2025.
The opportunities to apply agents are rapidly expanding. To understand where an agent might be able to add value, just look at your daily schedule. Note down every task you do — drafting emails, moving files from one place to another, or buying on Amazon. There very likely are agents out there that can automate the majority of these microtasks you’re performing. As agents gain contextual understanding and deepen their integrations with other systems, they can take on more tasks, just like you would.
I’ve seen businesses successfully apply agents to document processing, software development, healthcare records processing, employee onboarding, prospect outreach, claims processing, QA, vulnerability testing, hiring, debugging, procurement, R&D, and much more.
Trust Levels Improve
There’s been pushback on agents due to concerns about security, trust, and bias. Some of our customers — tax preparation firms — have explicitly asked that their client data not be shared with the model providers. And many in the industry simply don’t trust the results of AI yet — especially in an industry like tax preparation, where accuracy is critical. Overall, however, trust is increasing quickly, especially with younger people.
Bias is a problem of the past; large language models are rapidly improving and overcomig the types of biases we saw when they were first released. Finally, one thing we hear often are references to “humans in the loop.” Based on your customer and industry, you have to identify the most critical areas of “the loop” where a human is necessary, then clearly communicate that to stakeholders. There will always be detractors when it comes to new tech. I expect agents, and people’s comfort, to progress in a similar way as with past technology revolutions.
Agents are also playing a role in startups. AI and automation can help you scale any business faster. My word of caution from personal experience: Start with things that don’t scale. Do it manually, yourself if possible. Figure out what needs to be automated. Then automate. Investing in agents can be expensive so you have to know that you’re automating the right things. Once you know you want to bring agents into the mix, you can experiment with: running custom GPT prompts on datasets, sales agents, Salesforce Einstein, and many other tools. Try searching for “AI agents for…” and include the task you want completed. There’s probably an open-source system or startup that’s addressing it.
Ask Cloud Wars AI Agent about this analysis
There’s been debate about the future of agents, whether they would be mostly horizontal or vertical. In my company’s case, vertical agents have the advantage of industry-specific fine tuning, custom user interfaces that solve specific use cases for a customer profile, and integrations to industry-specific software. However, horizontal agents are winning in use cases like coding assistants, customer support, sales agents, and other more mature areas. Microsoft, UiPath, and many others are also enabling companies to build their own agents now.
I believe agents will be the next major modality in the progression from PC to web to mobile and now to agents. It’s hard to predict how much impact agents will have. In some ways, ChatGPT’s success is a signal of the coming agent revolution. One closing thought to consider: instead of focusing on agent as a noun, focus on the adjective agentic.
As we move into the agent era, think about how agentic your workflow can be, how agentic your workforce is, how your agents are working together with humans, where an agentic experience could be superior. Focusing on these factors will keep you at the forefront of the agentic movement!
