GPTs and the Shift to AI-First Products

OpenAI's Dev Day on November 6th announced something I've been expecting: GPTs. Customizable AI assistants for specific tasks. It's a subtle but significant shift in how AI product design works. By November 2023, it's clear that the traditional product building paradigm—ship a product with features, update monthly—is being challenged by AI-first thinking.

Let me be clear about what GPTs actually are. They're not revolutionary technology. They're configurations on top of the GPT-4 API with file uploads and custom instructions. But the implications for product design are real.

What Changed

The old product model: you define features, ship them, users learn them. The feature set is relatively stable. Updates are releases.

The AI-first model: you ship a system with an AI assistant. The AI learns from interaction. The feature set is emergent. Users define capabilities through conversation.

This is fundamentally different. In the old model, you're designing the UI. In the AI-first model, you're designing the system that responds to user instructions. The user becomes a co-designer of the product through their interactions with the AI.

GPTs represent OpenAI's bet on this model. You create a GPT by providing instructions, connecting it to data, setting constraints. Users interact with it. Over time, you refine the instructions based on what works.

What This Means for Product Strategy

Traditional product management is becoming harder. Here's why:

The product roadmap looks different. Instead of "Q1: add reporting, Q2: add approvals, Q3: add mobile," the roadmap is "improve AI reasoning, expand data access, refine instructions." The PM is still managing, but the thing being managed is different.

Feature release cycles change. With AI, you don't need to ship a new version for every improvement. You update the AI's instructions, and the behavior changes. Rollout is instant and reversible.

The feedback loop is tighter. With AI, users engage differently. They ask the AI to do things. If it fails, they try differently. The AI learns from this. Your product team can see these interactions and iterate quickly.

But there's a cost: you lose control. In a traditional app, behavior is predictable and bounded by what you built. An AI system is emergent. You can't perfectly predict what the AI will do in every scenario. That's powerful for flexibility. It's terrifying for stability.

The Implication for Enterprise Product

For enterprise, this shift is important but nuanced. Enterprise wants predictability. They don't want AI doing unpredictable things in their ERP system. But they also want flexibility. They want to customize behavior without engineering a new feature.

This is where I think the real opportunity is. Companies that can build AI-first products that are both flexible and bounded. Systems that use AI for intelligence but maintain operational guardrails.

Imagine an ERP system where the workflow routing AI learns from company-specific patterns. "Orders from this customer are usually high-priority. Orders from this department tend to get stuck in approval. New vendors need extra scrutiny." The AI learns these patterns. It improves routing. But it never breaks the formal approval hierarchy. Flexibility within structure.

That's hard to build. But it's valuable. And GPTs are a step toward making it possible.

What This Means for Teams Like Arvension

For us, this changes product thinking. We're not just building features. We're building systems that improve over time. That requires different skills:

• Understanding how to write system prompts that guide AI behavior
• Ability to iterate based on how users interact with the AI
• Comfort with emergent behavior within bounded constraints
• Strong measurement of "is this actually better?" not just "is it different?"

It also means hiring differently. We need people who understand AI, not just software engineers. Data people who can structure information for AI. Product people who can think about AI interactions, not just UI flows.

The Uncomfortable Transition

Here's the hard part: the shift from traditional to AI-first product design is a transition for everyone. The people who are great at designing traditional features might not be great at designing AI interactions. The feedback loops are different. The intuition doesn't transfer perfectly.

I think there will be a period—maybe a year or two—where AI-first products are seen as less reliable than traditional products. Because they are, initially. The AI hallucinates. It makes weird decisions. Users distrust it. Over time, as we get better at training and constraining the AI, that changes. But the transition is uncomfortable.

Companies that navigate this transition well—that maintain enterprise stability while adopting AI-first thinking—will have an advantage. Companies that swing too far toward AI and lose operational discipline will fail.

The Realistic Take

GPTs aren't the end of software development. They're not going to replace engineers or PMs. But they do represent a real shift in how we think about product design.

The next wave of enterprise software won't be "software that helps you do your job." It'll be "AI systems that help you do your job better, faster, and with fewer steps." The software becomes an intelligence layer, not just a toolset.

That's different. It requires different thinking. Different skills. Different organizational structure.

For companies building enterprise software in November 2023, the question is: how do we incorporate AI thinking into our product strategy without losing the discipline and predictability enterprise demands?

The companies that answer that question well will win. The ones that just bolt AI onto their existing product will find that users want something more fundamental.

The shift is happening. GPTs are one signal. But the broader shift—from feature-based product thinking to AI-first thinking—is already underway. And by the end of 2023, I expect we'll see more products designed this way. Some will succeed. Most will fail. But the ones that succeed will define what enterprise software looks like for the next few years.