The End of Augmentation? How Agentic AI is Redefining Our Role

When AI Stops Assisting and Starts Deciding

Is the familiar mantra that AI will augment, not replace, our jobs still a reliable comfort? For a long time, we at Project Flux held that belief close. It was a sensible and reassuring way to frame the seismic shifts happening around us. But as the pace of technological change accelerates, we feel a growing unease with that simple dichotomy.

In the 100th episode of the Project Flux podcast, the conversation we had has led us to question this long held assumption, and to wonder if we are on the cusp of a more profound transformation than we ever imagined.

The speed of change is more than just incremental. It is exponential. And that changes everything.

Our Evolving Perspective

From Curation to Commentary

When we first started Project Flux, our goal was to report on the burgeoning world of AI, to be a signpost for interesting developments. We were learning alongside our audience, and our newsletter reflected that: a curated collection of links and articles.

Over time, as our understanding deepened, we began to inject more of our own perspective, to question and analyse the trends we were seeing. This shift was in part a response to feedback from you, our readers and listeners, who wanted to know what we thought.

We have always welcomed this dialogue, and the contrarian viewpoints that challenge our own thinking. It is in this spirit of open inquiry that we have arrived at a new, more nuanced understanding of AI's impact.

The Model Release Fatigue

We used to cover every single model release as a major event. Then we realised there were simply too many. The real story lies in the fundamental shifts in capability, not incremental improvements.

And lately, the most significant shift has been something far more profound than better models. It is the emergence of AI systems that can think, decide, and act entirely on their own.

The Rise of Agentic AI

Beyond the Chatbot

For us, the most significant shift in recent months has been the emergence of what we are calling 'agentic AI'. These are not just chatbots that respond to prompts; they are systems that can control their own workflows, connect with other tools, and operate outside the confines of a chat window.

They exist in your terminal, on your computer, and can, in essence, do anything a human can do with a keyboard and mouse. This is a profound leap in capability, and it is what underpins the feeling that we are moving beyond simple augmentation.

Packaging Workflows into Skills

We have been experimenting with tools like Anthropic's Claude skills, which allow you to package up any repeated workflow into a set of instructions that the AI can execute with the press of a button.

Consider a task that might have taken a few hours. Now it can be completed in moments.

This goes beyond making existing processes more efficient to fundamentally changing what is possible. It is about replacing the need for human labour entirely, at least for specific, well-defined tasks.

The implications are staggering. When we started seeing what Claude could do, and then when Anthropic released co-work as a front end for Claude, we realised something had shifted.

These tools had become truly agentic. They could connect to third-party services, control workflows, and operate independently.

The Terminal Revolution

Model Context Protocols (MCPs) took this further, moving these agents out of the chat window and into your terminal. Now, terminal-based agents can do almost anything you can do on your computer.

The keyboard and mouse are merely an abstraction; the AI does it directly. This represents a fundamental shift in how AI systems operate:

•They are no longer confined to conversational interfaces

•They can integrate seamlessly with existing tools and workflows

•They can operate independently without human intervention

•They can connect to multiple third-party services simultaneously

The SaaSpocalypse and the Future of Software

The Disruption of Traditional Models

The implications of this are far reaching. We are already seeing the beginnings of what some are calling the 'SaaSpocalypse', as the stock market begins to price in the disruptive potential of agentic AI.

Why pay for a costly software licence and the trained personnel to operate it, when a skill or a plugin can achieve the same result for a fraction of the cost?

We are even seeing the emergence of companies that can spin up custom-built, single use applications on demand. The app itself becomes disposable; the value lies in the data and the insights it generates.

Self-Improving Systems

The latest version of OpenAI's Codex represents a dramatic leap from the clunky, bug-ridden AI tools of just a year ago. We have been amazed by its ability to debug its own code as it works.

In a fascinating, and perhaps slightly unnerving, development, OpenAI even used Codex to help build itself. When the tools start to improve themselves, we are in truly uncharted territory.

The Overthinking Problem

But there are downsides. We have found that GPT 5.3, when used in normal situations, will start to overthink. It will do something, then go back and do it again, second-guessing itself.

This drives up token costs and can hit rate limits. It is a reminder that even as these tools become more powerful, they are not perfect. They have quirks and limitations that we are still learning to navigate.

What This Means for Project Delivery

The Skills We Need Now

We believe that the project delivery industry is in the midst of a profound transformation, and the rise of agentic AI will only accelerate this. The skills that have served us well in the past may not be the ones we need in the future.

The ability to craft a well-defined prompt, to understand the capabilities and limitations of these new tools, and to think critically about the outputs they generate will become paramount.

A Time for Adaptation, Not Fear

This is a time for curiosity and adaptation as we navigate the transformation ahead. We must be willing to let go of old certainties and embrace the new possibilities that are emerging.

The conversation has shifted to focus on how AI will change our jobs and how we can best prepare for that change.

Why You Should Listen to the Full Episode

Our conversation on the Project Flux podcast goes far deeper into the rise of agentic AI, the 'SaaSpocalypse', and the future of software.

We explore the recursive self-improvement of AI models, discuss the challenges of token costs and rate limits, and examine what it all means for the future of work. We also discuss the emergence of companies that can spin up custom-built applications on demand, and the implications of AI tools that can debug themselves.

There are nuances and insights that simply cannot be captured in a blog post, and we encourage you to listen to the full episode to hear the complete picture of where we believe the industry is heading.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.