Enterprise AI evolution: from Software 1.0 to Software 3.0

I've previously written about how software is evolving from the era of “1.0” to “2.0”. Reading my first blog will help a lot with context, but to summarise: software development has evolved from deterministic, hand-crafted rules and logic (Software 1.0), to machines learning and crafting these rules automatically (Software 2.0). This shift has unlocked entirely new problem domains - enabling machines to adapt and operate in complex, unstructured environments in ways that traditional programming couldn’t.

The bull case for Software 1.0

While the Software 2.0 wave is exciting, it’s important to note that not all workloads should - or can - be handed over to probabilistic systems.

Imagine replacing your financial system (like Xero or Sage) with an AI model that reconciles invoices and payments via a single prompt. Sounds magical! What you weren’t told though, is that this AI system, being probabilistic, will have a certain error rate: let’s say for example, 1%.

That 1% error, on a $10 million monthly reconciliation, results in $100,000 of misallocated payments per month. Unacceptable for any CFO, and a surefire way to get a human accountant fired - let alone a machine.

Whilst this example is an oversimplification, it’s clear that this is not an appropriate use case for AI, and your tried and tested Sage was actually the right solution in the beginning. This is the bull case for Software 1.0: deterministic systems that handle mission-critical operations with consistency, auditability, and reliability. These systems are not going anywhere. In fact, we see a future where they’ll continue to form the foundation of the enterprise - only now, augmented by new intelligent layers.

AI Agents

Until this point, Software 2.0, or “classical ML” has been narrow: requiring a perfectly curated set of inputs to operate effectively (data scientists call this “feature engineering”). Whilst transformative, this requirement has limited classical ML to well-defined problems, where data is clean, labelled, and structured. These systems performed incredibly well within those boundaries - powering use cases like fraud detection, recommendation engines, and credit scoring - but they struggled in open-ended, messy environments.

That’s now changing.

The rise of foundation models and large language models (LLMs) has pushed Software 2.0 into new territory. We’re moving from narrow pattern recognition engines to general-purpose systems capable of interpreting unstructured data, understanding context, and acting on intent.

This is where agents come in.

Agents: what are they?

No doubt you’ve all heard all the hype about “AI agents”. A simple definition that I like is that they are “systems that dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks”. Agents are LLMs that have been augmented with additional capabilities that allow them to plan, reason, remember and improve from past interactions. They can also use external tools, like web browsers, and call APIs the same way humans do.

Agents are the next evolution of Software 2.0. They operate with truly unstructured inputs, unlike their predecessors, understanding deep nuance in complex environments; (almost) completely out of the box. They are fast becoming the decision engines inside business systems and workflows.

Software 3.0 and the opportunity it brings

The first wave of Software 1.0 SaaS products were software that translated the physical tools humans used to go about conducting their day-to-day work. Think typewriters, paper-based invoicing systems, printers and other equipment that now lives in a digital form on our laptops and mobile phones as digital programs.

The average Fortune 500 company spends about 1-5% of their revenue on this traditional SaaS which equates to about $96 million per year. These SaaS fees displaced the cost of traditional tools in a skeuomorphic transfer from physical to software.

By contrast, the next wave of Agentic-SaaS - what we think of as Software 3.0 - targets an entirely different cost base: the human labour required to operate and interpret these digital programs. Agents aren’t replacing typewriters or filing cabinets - they’re replacing and augmenting the people using them. Think dynamic, unpredictable workflows like outbound sales, customer support, or fraud investigations, where context, nuance, and deep reasoning matter. This opens up a new vector of value creation: software that doesn’t just assist humans, but increasingly stands in for them on the front lines of business operations.

So how big is this opportunity?

Going back to the average Fortune 500 company, human labour cost (payroll, contractors, and benefits) exceeds $5-6 billion, which accounts for 50–60% of their revenue. This is the new TAM for Agents and Software 3.0 - an order of magnitude increase in opportunity.

This shift reframes the role of software in the enterprise. No longer just a tool for process efficiency; software is becoming a thinking, acting participant in the business itself. Software 3.0 opens the aperture from just digitizing work to mirroring cognitive work. Eventually, it will be as easy to “spin up” a new employee as it is to switch on the lights when walking into a room.

How we're implementing Software 3.0

The shift from Software 1.0 to 2.0 - and now to 3.0 - will mean profound shifts in the way we work, and our businesses are run. At PYGIO, we’re not just building this future for our clients - we're applying it internally. We believe our internal teams should be augmenting their work, “hiring” agents into projects to automate the monotonous, low-hanging tasks they do every day.

After all, our business is one of cognitive labour: the sweet-spot for agentic systems. Quite ambitiously, we believe that the next generation of consultancies will scale not by hiring hundreds of thousands of people, but by combining small, high-leverage teams with fleets of intelligent agents - unlocking the capacity of an Accenture with a fraction of the headcount.

In the meantime, we're working on a bunch of exciting partnerships and internal initiatives that will accelerate our work in this space safely, scalably, and with speed. Whether you’re looking to reduce cost, increase output, or build AI-native products, the time to act is now, and we’re excited to be on the forefront of this wave.

More to come soon as we continue sharing our stories from the front lines!