If you asked ChatGPT to "book me a flight" last year, it would give you a list of websites. Today, the expectation has shifted. We want the AI to actually go to the site, select the flight, and book it.
This is the transition from Generative AI (creating text/images) to Agentic AI (taking action). October 2024 has been a watershed month for this technology, notably with Anthropic releasing "Computer Use" capabilities for Claude 3.5 Sonnet, allowing AI to control a desktop interface just like a human would.
What Defines an AI Agent?
Autonomy
Agents don't just wait for the next prompt. They can plan a sequence of steps to achieve a high-level goal.
Tool Use
They can access APIs, browse the web, query databases, and use software interfaces directly.
Memory
Agents maintain context over long periods, remembering user preferences and past actions.
Recent Breakthroughs (Oct-Nov 2024)
Anthropic's Computer Use
Claude 3.5 Sonnet can now look at a screen, move a cursor, click buttons, and type text. This general-purpose capability means it can use any software, not just ones with APIs. This is a massive leap for legacy system automation.
Salesforce Agentforce
Salesforce has pivoted its entire strategy around "Agentforce," a suite of customizable agents for sales and service. This signals that major enterprise software players are betting the farm on agentic workflows.
OpenAI Swarm
The release of the educational "Swarm" framework highlights the trend towards multi-agent systems, where specialized agents (e.g., a researcher and a writer) collaborate to finish a task.
The Challenges Ahead
Safety & Control
An agent that can click buttons can also delete databases or send incorrect emails. "Human-in-the-loop" (HITL) governance is becoming critical. We need guardrails to ensure agents don't go rogue.
Reliability
Agents can get stuck in loops or misinterpret UI elements. While "Computer Use" is impressive, it's still experimental. For mission-critical flows, API-based agents are still safer than vision-based ones.
Preparing for the Agentic Future
For businesses, the message is clear: Start structuring your data and documenting your workflows. Agents need clear instructions and clean data to function. The companies that prepare their digital infrastructure today will be the ones deploying autonomous workforces tomorrow.