Will Agents Kill Software? - This Is Not Advice #10

👋 Hi friends,

This newsletter is becoming more like a personal journal for when I have a thought I cannot shake. Last night was one of those times. I was lying in bed, trying to fall asleep, when a question landed in my head and refused to leave. One of those thoughts you have to chase down before you can rest.

Here is where I eventually landed:

If agentic AI, systems that can autonomously act, reason, and chain tasks, really is the inevitable future, then why would software still exist in the form we know it?

If an AI can build exactly the tool it needs, use it, and discard it, why would we need apps with buttons, menus, and dashboards at all? The AI would be the primary user, and we would just talk to it.

The contradiction

Here is the strange part: the same investors loudly proclaiming that agents will change everything are also pouring millions into traditional SaaS products with AI features inside.

Why fund the old model if you believe the new one will replace it?

My working theory

Maybe the answer is that value still sits in the base layer.

SaaS companies already own the customers, the data, the integrations, the compliance.

They have spent years building trust, security, and reliability, things agents would need to rebuild from scratch.

For now, it is faster, safer, and cheaper for an agent to plug into existing infrastructure than to reinvent it every time.

In other words, agents may change the interface to software, but not necessarily the software itself, at least not yet.

Why agents might not code it all immediately

Even if the tech gets there, there are friction points:

• Spinning up bespoke code for every task could be slower and more resource-heavy than reusing a proven API.
• Deep domain expertise, such as tax law, trading algorithms, and compliance, is already embedded in existing systems.
• Regulated industries will still require certified, audited software for a long time.

A possible speculative timeline

Note: I had originally put speculative years on this timeline, but making those predictions feels totally arrogant given how quickly things can change. Honestly, I have no idea, so there is no point making any time-based prediction.

Phase 1: SaaS with AI inside. Humans still the main users.

Phase 2: Agents become the main users of SaaS APIs, humans mostly interact with their agent.

Phase 3: Many SaaS platforms fade into the background as "agent infrastructure", invisible to humans, essential to AI.

The open question

If this plays out, software will not die. It will just shift roles, from something humans touch directly to something AI uses on our behalf.

Will the most valuable stuff come with the best human UX or the most context agents can interact with?