2025 will bring major shifts in AI capabilities across several key areas. Let me break down my main predictions.
First, AI agents will keep getting more powerful, but media coverage will shift to call them overhyped. Both perspectives are actually correct – agents are overhyped AND incredibly useful when applied properly.
Anthropic makes an important distinction between two types of AI systems:
Workflows are systems where AI models and tools follow predefined paths. Agents are systems where AI models control their own processes and tool usage independently.
For most business tasks like content creation and lead generation, you want AI workflows rather than fully autonomous agents. I’ve built many AI workflows that execute complex business processes in a controlled, predictable way. Full autonomy only makes sense for tasks that can’t be predefined, like research that needs to adapt to new information.
Perplexity is a good example of a research-focused agent that can effectively use multiple tools to explore topics and adapt its approach. We’ll see more vertical agents like this in 2025, but most businesses will use established ones rather than building their own. The ROI isn’t there for building custom agents since they need extensive testing and can get confused when juggling too many tools.
Second, AI coding tools will keep advancing rapidly. Tools like Cline already let non-coders create custom projects. With upcoming models like Gemini 2.0 Pro’s 2 million token context window and Claude 4’s improved capabilities, we’ll be able to build much more sophisticated systems easily.
Third, video generation will become more practical. Current high-quality options cost around $0.25 per video, making automation expensive. The cheapest option, LTX video at $0.02 per video, sacrifices too much quality. By 2025, I expect high-quality video generation under $0.10 per video, enabling true automation.
Fourth, open source models will continue challenging closed source ones. Models like LLaMA 3.1 405B have nearly caught up to proprietary options. In video, open source models like Hunyuan already match top closed source models in many ways.
Open source isn’t inherently better – AI companies deserve to make money from their work. But strong open source competition pushes the whole industry forward and provides cost-effective options.
We’re in an incredible time for automation. Between advancing AI capabilities and existing automation tools, you can automate almost anything now. And this is just the beginning – the technology will only get better from here.
For more on AI research tools like Perplexity that I mentioned, check out my detailed analysis here: https://adam.holter.com/perplexity-ai-why-its-the-best-research-tool/
