How AI Transforms Software

originally posted 02/15/2025

Most current AI integration in software simply bolts on a chat interface — an obvious but underwhelming first step. AI is going to fundamentally transform how we interact with ALL applications. Here is a cursory overview of key changes I see coming:

For Users:

1. True Real-time Collaboration

Users won’t just be “talking to AI” through chat interfaces. Instead, they’ll work alongside AI in real-time, collaboratively editing and creating together — just as if a colleague was there editing the same document with you.

2. Smarter Automation That Knows Its Limits

We’ll see automation that:

- Confirms important actions before executing them

- Handles complex workflows while keeping humans in the loop

- Knows when to ask for help rather than making wrong assumptions

3. Multi-Modal Interaction

Users will move beyond text-only communication with AI:

- Just as human conversations benefit from voice, gesture, and visual cues, AI interactions will become richer and more natural

- My work with Spiky.ai illustrates how analyzing video alongside text dramatically improves understanding of user intent

What We Need to Build Under the Hood:

1. LLMs as Interface Layer, Not Complete Solution

We need to:

- Use LLMs primarily as intelligent interfaces to specialized tools

- Stop trying to teach LLMs everything and instead teach them when to use existing tools. instead of desperately trying to get LLMs to do good math, show them how to use a calculator.

2. Rethinking Workflow Automation

As developers, we need to recognize that:

- Traditional engineering approaches often fail to capture real-world complexity

- We need systems that handle exceptions and edge cases gracefully

- Automation should be viewed as a continuous spectrum, not an all-or-nothing proposition

3. Sophisticated Feedback Systems

We must build:

- Nuanced data collection that learns from actual user behavior, not just explicit feedback

- Systems that can tune results both for individual users and across the entire user base

- Ways to understand the quality of AI outputs implicitly without to constantly improve systems vs relying on direct user feedback. Thats fuzzy implicit feedback is what Ai is supposed to be good at interpreting.

The Path Forward

We need to focus on designing for continuous collaboration between humans and AI while maintaining user trust and control. We will need to rethink all our apps from the ground up. Perhaps to some that is frustrating and daunting. personally I am excited and exhilarated by this new dynamic!