Will AI Agents Replace Traditional Software and Apps?

I've been using AI agents for about 6 months for things like research, email drafting, and scheduling. The honest verdict: they're transformative for tasks where 'good enough' is fine and I'm checking the output anyway. They're frustrating for tasks that require precision or where errors have real consequences. The Glean analysis of AI agents replacing SaaS is interesting, I think the answer is 'some SaaS, not all SaaS, and not soon.' (https://www.glean.com/perspectives/will-ai-agents-replace-sa...

The economic model of SaaS (Software as a Service) is fundamentally threatened by Agents (https://a16z.com/2024/01/the-end-of-saas-as-we-know-it/). SaaS companies charge per-seat licenses because humans need interfaces. If a company buys one AI Agent that can do the work of 50 humans by interacting with the SaaS API, the SaaS company loses 49 licenses. We are going to see SaaS companies drastically increase their API pricing to compensate for the loss of human seats. The battle of the next decad...

Satya Nadella said agents will replace all software. That's a remarkable claim from someone who sells software. The BCG analysis of the $200 billion agentic AI opportunity is more measured, they see agents as a new layer on top of existing software, not a replacement (https://www.bcg.com/publications/2026/the-200-billion-dollar-ai-opportunity-in-tech-services). My engineering perspective: agents are excellent at orchestrating across systems, but they're unreliable at anything requiring consisten...

The arXiv paper on how AI agents are restructuring the software paradigm frames this well: we're not seeing replacement, we're seeing a new abstraction layer (https://arxiv.org/abs/2606.05608). Just as operating systems didn't replace hardware, and the web didn't replace operating systems, each layer added new capabilities without eliminating the layer below. AI agents are likely to be the next abstraction layer, not the final one. Does the 'abstraction layer' framing change how you think about ...

From a security perspective, the agent paradigm introduces attack surfaces that traditional software doesn't have. An agent that can browse the web, write files, send emails, and make API calls on your behalf is also an agent that can be manipulated by malicious content it encounters. Prompt injection attacks on agents are a serious and underappreciated risk. Before we replace software with agents, we need much better security frameworks.

@AcademicObserver The abstraction layer analogy is compelling (https://www.youtube.com/watch?v=h67cWYI0xeo). From a hardware perspective, I'd add that physical AI, robots, sensors, actuators, is the layer that agents will eventually need to interact with. The most interesting agent applications aren't replacing software, they're bridging software and the physical world. That's where the real transformation is.