I still remember the first time I tried to use AI to write some simple code. I was trying to automate a super basic task for a personal project, and after wrestling with the syntax for ages, I figured AI could speed things up. I typed in my request, full of hope, and back came… absolute gibberish. It was like talking to a very enthusiastic but ultimately clueless parrot. Fast forward to today, and that same parrot has somehow become a brilliant software engineer, capable of building entire apps from just a few sentences. It's incredible how much has changed, and it's forcing all of us in product to rethink what's possible.

The Product Manager's New AI Playbook

Artificial intelligence has come a long way. It used to be that tools needed explicit instructions, but now we have fully autonomous agents making their own decisions. What began as a way to summarize text or generate SQL from plain English has now evolved into something new: agentic AI.

This shift isn't just about better models. It's about how we, as builders and users, adapt and push AI's boundaries. Each phase has opened new possibilities, not only because of technological improvements but also because ambitious founders experimented, iterated, and proved what was possible. Every major AI evolution has led to successful startups pioneering the next big thing.

AI as a Tool: The Early Days

The first wave of AI was all about direct inputs producing direct outputs. You had to prompt it precisely to get the result you wanted. This was particularly useful for structured tasks like summarization, translation, and generating SQL queries from plain English.

Early tools, for instance, turned human language into functional SQL, making databases far more accessible. Similarly, startups like TextGen built AI-powered content generation tools that automated many creative tasks. At this stage, AI wasn't truly "thinking"—it was responding. The intelligence still relied heavily on the user’s instructions.

AI as an Assistant: Getting More Integrated

The next phase saw AI move beyond one-off interactions into more integrated, interactive assistants. Instead of simply producing an output, AI became part of a workflow, continuously helping users with tasks.

For example, CodePal embedded AI inside development environments, making coding more intuitive. These assistants still required significant human direction. They reduced friction, but they didn’t independently make decisions.

They helped, but they still relied heavily on user direction.

AI as an Agent: From Assistance to Action

Then came AI that could not only assist but act. AI agents execute tasks end-to-end, taking automation to another level.

A breakthrough example is AlphaCode, an AI capable of coding and debugging on its own. This was a major shift—AI wasn't just enhancing human capabilities; it was beginning to replace manual effort entirely.

The Rise of Agentic AI

Now, AI is moving beyond executing predefined tasks to dynamically figuring out what needs to be done next. Instead of following a fixed script, agentic AI can adapt in real time.

Anthropic’s Claude is one of the leaders in this space, aiming to build AI that not only completes requests but can also determine what’s necessary to achieve a broader goal. This marks the beginning of AI managing complex processes with minimal human intervention.

The Race to Find New Use Cases

With the rise of agentic AI, there’s intense competition among founders to discover the next breakthrough use cases. Investors are placing massive bets, sometimes funding AI startups at sky-high valuations—even before they’ve proven product-market fit.

This is both an incredible opportunity and a dangerous trap. Some of these AI startups will fail. A lot of capital will be lost. But for the ones that succeed, the upside is enormous.

Take AppForge.ai as an example. This text-to-functioning-app editor rocketed to $17M in ARR in just six months. The appeal is clear: you simply type the app you want, and AppForge generates the necessary code and deploys it for you. This kind of AI-native experience would have been unimaginable just a few years ago.

What Enabled This Shift?

Several key factors have driven AI’s rapid evolution:

• Better models: More capable AI like GPT-4, Claude, and multimodal systems that handle images, text, and actions.

• Better infrastructure: Vector databases, AI memory, and improved orchestration frameworks have made complex AI applications more viable.

• Successful experiments: Startups testing the limits of AI have demonstrated what’s possible, inspiring others to follow suit.

Why This Matters for Product Managers

Each AI evolution doesn’t replace the last—it builds on it. AI assistants didn’t kill AI tools, and AI agents haven’t made assistants obsolete. Every step forward has unlocked new, higher-value use cases.

We’re just at the beginning of agentic AI. The most exciting part isn’t what AI can do today—it’s what people will build with it next. The startups that master this new paradigm will define the next era of AI-powered work. Soon, instead of asking AI to perform a task, we’ll just define a goal—and AI will figure out how to achieve it.

The race is on for product managers to harness these new capabilities. Will you be leading the charge, or playing catch-up?

Related Articles

• The AI Product Manager: Redefining the Role in the Age of Intelligent Automation

• Product Managers, Reinvented: Embracing Your Role as AI Orchestrators

• The AI Advantage: Why Your Competitors Will Be Using Intelligent Product Tools