Last Updated: April 12, 2026
Something fundamental is shifting in how businesses consume technology, and most companies haven't noticed yet. The SaaS model that dominated the last two decades is hitting a wall. What's replacing it isn't another software category. It's an entirely different paradigm: hiring AI agents as digital employees.
This isn't theoretical. I've spent nine years in consulting, six of those at EY as a Manager in IT Advisory, helping organisations like IAG, Genesis Care, CBA, and Westpac navigate technology transformations. I've seen plenty of "next big things" come and go. This one is different. Not because the technology is flashy, but because the economics are irreversible.
The agencies that will dominate the next decade aren't SaaS resellers or traditional IT consultancies. They are AI automation agencies: teams who know how to build, deploy, and manage AI agents that run on a client's own infrastructure, trained on their own knowledge, growing more valuable every single day.
Here's why.
Why Is the Traditional SaaS Model Dying?
The SaaS revolution was a massive upgrade from on-premise software. No more servers to maintain, no more version upgrades, no more upfront licence fees. But SaaS introduced its own set of problems that are now impossible to ignore.
You rent everything and own nothing. Every month, you pay for software that you will never own. Your data lives in someone else's database. Your workflows conform to their product roadmap. Your business processes bend to fit their interface. When they raise prices, you pay or you migrate. When they shut down a feature you depend on, you adapt or you leave.
One size fits nobody perfectly. SaaS products are built for the widest possible market. That means compromises. You use 20% of the features and lack the 10% you actually need. Customisation is limited to whatever the vendor decides to support. Your competitive advantage gets flattened into the same toolset your competitors use.
Your data subsidises their AI. Here's the part nobody talks about: SaaS companies are training their AI on your data, your workflows, and your patterns. You're paying them to build intelligence that they will then sell back to you, and to your competitors. Your proprietary processes become their training data.
The subscription treadmill never stops. The average mid-market company now spends $500,000 to $2 million annually on SaaS subscriptions, according to recent industry analysis. That number only goes up. Every new department need is another subscription. Every integration gap is another tool. The stack grows, the costs compound, and the value per dollar diminishes.
The SaaS model made sense when software was the product. But software isn't the product anymore. Intelligence is the product. And intelligence can't be effectively delivered as a one-size-fits-all subscription.
What Is the Agentic AI Employee Model?
The agentic employee model flips the SaaS equation on its head. Instead of renting generic software, you hire specialised AI agents that work as digital employees. These agents run on your infrastructure, learn your processes, and accumulate knowledge specific to your business.
They are custom, not generic. An AI agent built for your accounts receivable process understands your specific invoice formats, your approval chains, your escalation rules. It doesn't force you into someone else's workflow. It adapts to yours.
They learn and improve. Unlike software that only gets better when the vendor ships an update, AI agents improve every day through interaction. Every conversation, every document processed, every edge case handled makes the agent more valuable. It's an appreciating asset.
They work 24/7 without complaint. No sick days, no handover gaps, no inconsistencies between shifts. The agent that handles your customer onboarding at 2am on a Sunday is the same agent that handles it at 10am on a Tuesday.
They integrate with everything. Because they run on your infrastructure, AI agents can connect directly to your databases, your CRM, your ERP, your email. No API limits from a third-party vendor. No data flowing through external servers.
This is not science fiction. The tools to build and deploy these agents exist today, and they are increasingly open source.
Why Will AI Automation Agencies Dominate This Shift?
The transition from SaaS to agentic AI isn't something most companies can navigate alone. It requires expertise across multiple domains: infrastructure management, model selection, agent design, security, compliance, and ongoing optimisation. That's where AI automation agencies come in.
They bridge the gap between AI capability and business application. The open source AI ecosystem is exploding with powerful tools, but most businesses lack the expertise to evaluate, deploy, and maintain them. AI automation agencies translate raw capability into business value.
They provide the "last mile" that open source needs. Open source tools are powerful but rarely plug-and-play. Someone needs to configure the infrastructure, set up the agent frameworks, implement security protocols, and tune performance. Agencies provide this critical integration layer.
They build institutional knowledge. Every deployment teaches the agency something new. Every industry vertical adds patterns and playbooks. Over time, agencies accumulate a deep library of deployment patterns that make each new client faster to onboard and more reliable to run.
They offer ongoing optimisation. AI agents aren't set-and-forget. Models improve, costs change, new tools emerge. Agencies provide the continuous optimisation that keeps agent infrastructure running at peak efficiency.
In my TOGAF-certified enterprise architecture practice, I learned that the biggest technology failures aren't technical. They're organisational. Companies buy tools they can't deploy, deploy tools they can't maintain, and maintain tools they can't optimise. AI automation agencies solve this by owning the full lifecycle.
How Is the Open Source AI Ecosystem Enabling This?
The open source AI community is experiencing growth that makes the early days of Linux look gradual. Every week, new frameworks, models, and tools are released that bring frontier AI capabilities within reach of any organisation willing to deploy them.
Here's what's happening and why it matters.
The cost of intelligence is plummeting. Open source models like DeepSeek and GLM (Zhipu AI) are delivering performance that rivals proprietary frontier models at a fraction of the cost. DeepSeek's models have demonstrated that you don't need a billion-dollar training budget to compete. GLM's models from China are proving that the best open source AI isn't confined to Silicon Valley.
The agent frameworks are maturing fast. OpenClaw provides the foundational agent framework that runs entirely on your own infrastructure. Multica turns coding agents into real teammates with managed agent platforms. Hermes Agent offers another open source alternative to OpenClaw. LangChain Deep Agents handles complex agent orchestration. DeerFlow, released by ByteDance, brings superagent capabilities into the open source world.
Optimisation is getting dramatically cheaper. GEPA, an open source prompt optimisation tool, uses reflective text evolution to beat reinforcement learning approaches at 90x lower cost. It integrates directly into DSPy, the framework for programming language models. This means you can optimise your agents without a PhD in machine learning.
Enterprise-grade options are emerging. NVIDIA NemoClaw provides enterprise agent framework capabilities, while IronClaw offers a security-focused alternative. These sit alongside community tools like GBrain, the personal knowledge brain for AI agents popularised by Garry Tan's setup. Grok Imagine brings open source image generation into the mix.
The closed-source world is responding, not leading. Anthropic's Claude Managed Agents represents the closed-source approach: powerful, but you're renting intelligence on someone else's terms. The contrast with open source is stark. With OpenClaw or Hermes Agent, you own the deployment. With Claude Managed Agents, you're another tenant.
The net effect: the technology moat is disappearing. The agencies that win won't be the ones with the best proprietary tools. They'll be the ones who can most effectively deploy, secure, and optimise the incredible open source tools now available to everyone.
Why Is Owning Your Infrastructure Non-Negotiable?
This is the point where most AI consultants get uncomfortable, because it's easier to sell a cloud subscription than to build infrastructure. But the data is clear: for any company serious about AI, owning your infrastructure is the only rational choice.
Data sovereignty. Your client data, your trade secrets, your customer conversations, your proprietary processes. All of it stays on your servers. Not in a SaaS vendor's database. Not training someone else's model. Yours.
Full control over agent behaviour. When an AI agent runs on your infrastructure, you control its memory, its personality, its decision boundaries, its update schedule. You're not waiting for a vendor to fix a bug or add a feature. You own the full stack.
Cost optimisation by intelligence tier. Not every task needs a frontier model. Simple classification can run on a lightweight open source model for fractions of a cent. Complex reasoning can invoke a more powerful model. When you own the infrastructure, you choose the right intelligence level for each task. SaaS forces you into one pricing tier regardless of actual usage.
No vendor lock-in. Open source means you can migrate, modify, and extend without asking permission. Your agent infrastructure isn't held hostage by a vendor's roadmap or pricing decisions.
Compliance and auditability. For regulated industries (healthcare, financial services, government), running AI on your own infrastructure isn't just preferable. It's often legally required. You can't demonstrate compliance when your data flows through third-party systems you don't control.
In enterprise architecture, we call this the "control plane." If you don't own the control plane, you don't own the system. Full stop.
How Do You Teach AI Agents Using Domain Knowledge?
Here's the insight that changes everything: your company already has incredible domain knowledge trapped in documentation, processes, and the heads of your experienced staff. The bottleneck has never been the intelligence. It's been the mechanism to capture and deploy it.
AI automation agencies unlock this through a structured process.
Set up secure agent environments. Before any training happens, the infrastructure needs to be right. Secure networking, appropriate compute resources, model serving, memory management, and monitoring. This is where agencies with deployment expertise add immediate value.
Capture knowledge in plain English. You don't need to write code to train an AI agent. Modern frameworks let you teach agents through natural language instructions, example conversations, and document ingestion. Your accounts manager explains how she handles disputed invoices. The agent learns. Your senior engineer describes the debugging process for a specific system. The agent learns.
Layer knowledge progressively. Start with the core processes. Get the agent handling routine tasks reliably. Then add edge cases, exceptions, and escalation rules. Each layer makes the agent more capable and more valuable.
Optimise continuously. Tools like GEPA and DSPy allow agencies to systematically improve agent performance over time. This isn't guesswork. It's measured, data-driven optimisation that compounds.
The result is an agent that doesn't just follow a script. It understands your business the way a well-trained employee does, because it was trained by your well-trained employees.
Why Do AI Agents Appreciate in Value Over Time?
This is the economic argument that makes the agentic model irresistible, and it's the exact opposite of how SaaS works.
SaaS is a depreciating relationship. On day one of a SaaS subscription, you get maximum value: shiny new software with all its promised features. From there, value erodes. The product evolves in directions you didn't ask for. Your customisation work becomes obsolete with each update. Your institutional knowledge of the product becomes a liability when they redesign the interface. You're paying the same (or more) for diminishing returns.
AI agents are appreciating assets. On day one, an AI agent is a capable but inexperienced digital employee. It knows the basics and can handle routine tasks. But every interaction makes it better. Every document it processes adds to its knowledge. Every edge case it handles becomes part of its training. Every process it refines becomes institutional memory that doesn't walk out the door when an employee resigns.
The compound effect is extraordinary. After six months, your AI agent has processed thousands of transactions, handled hundreds of edge cases, and absorbed the domain knowledge of your best employees. After a year, it's arguably your most experienced "team member" in its specific domain. After two years, it's a proprietary asset that would cost a competitor enormous time and money to replicate.
This changes the build-vs-buy calculus entirely. With SaaS, you're renting a depreciating asset. With an AI agent, you're building an appreciating one. The longer you invest, the greater the return. This is why owning the infrastructure matters so much: you can't build a valuable asset on rented land.
How Does Flowtivity Approach This?
At Flowtivity, we've built our entire practice around this shift. Not because we predicted it from day one, but because our consulting experience led us here naturally.
Our background is enterprise consulting. Nine years of it, including six at EY where I managed scrum teams of 15+ and delivered over $15 million in measurable business benefits for clients like IAG, Genesis Care, CBA, and Westpac. The TOGAF certification isn't decorative; it shaped how we think about architecture, governance, and long-term system design.
What we noticed was a pattern. Small and medium businesses, the 11-to-200 employee companies that drive the Australian economy, were being left behind by the AI revolution. The enterprise firms had budgets for custom AI. The startups had the technical talent. But the construction companies, the allied health practices, the professional services firms, the local trades businesses? They had all the domain knowledge in the world and no way to deploy it.
So we built a practice around exactly that. We help companies:
- Deploy AI agents on their own infrastructure using open source frameworks like OpenClaw
- Transform domain knowledge into agent capabilities through plain-English training
- Optimise infrastructure costs by matching intelligence levels to task complexity
- Build growing, proprietary AI assets that appreciate in value over time
We're not selling software. We're helping companies build digital employees that become more valuable every quarter.
What Open Source Tools Make This Possible?
The open source ecosystem has reached a tipping point where the tools are not just adequate but genuinely superior for many use cases. Here are the ones we've reviewed and that are shaping this space.
Agent Frameworks
- OpenClaw: The open source agent framework that runs on your own infrastructure. This is the foundation. It gives you full control over agent deployment, memory, and behaviour without vendor dependencies.
- Hermes Agent: An open source alternative to OpenClaw, providing another option for self-hosted agent management.
- LangChain Deep Agents: Open source agent orchestration for complex, multi-step workflows. Excels at chaining agent actions together.
- DeerFlow (ByteDance): An open source superagent platform that demonstrates how sophisticated agent architectures can be when built on open foundations.
- Multica: An open source managed agents platform that turns coding agents into real teammates. Essential for agencies building and maintaining multiple client deployments.
Optimisation and Training
- GEPA: Open source prompt optimisation via reflective text evolution. Beats reinforcement learning at 90x lower cost. This is a game-changer for agencies that need to fine-tune agent performance without massive compute budgets.
- DSPy: The framework for programming language models. GEPA integrates directly into DSPy, creating a powerful optimisation pipeline.
Models
- DeepSeek: Open source AI models that are challenging proprietary frontier models on benchmarks while being dramatically cheaper to run.
- GLM (Zhipu AI): Open source models from China that are competitive with frontier models, proving that open source AI excellence is a global phenomenon.
Enterprise and Security
- NVIDIA NemoClaw: Enterprise agent framework. Useful comparison point for what the enterprise world is building in response to open source.
- IronClaw: Security-focused AI agent alternative. Important for regulated industries that need verifiable security controls.
Productivity and Knowledge
- GBrain: Personal knowledge brain for AI agents. Popularised by Garry Tan's setup, it shows how agents can maintain and leverage deep knowledge bases.
- Grok Imagine: Open source image generation, demonstrating that open source capabilities extend beyond text into multimodal AI.
Closed-Source Comparison
- Claude Managed Agents: Anthropic's approach to managed AI agents. Powerful, but it represents the closed-source, rental model. You get capability but lose ownership and control.
The pattern is clear: open source tools now cover the full spectrum of agent deployment, from framework to optimisation to models to security. You don't need to rent anyone's platform. You can own the entire stack.
What Does This Mean for Australian Businesses?
Australia has a unique combination of characteristics that make it both vulnerable to the SaaS trap and perfectly positioned for the agentic transition.
The SaaS dependency is deep. Australian businesses adopted SaaS aggressively. The average mid-market company runs 40 to 80 SaaS subscriptions. That's a massive cost base with diminishing returns. Many of these subscriptions could be replaced by purpose-built AI agents that cost less and do more.
The domain knowledge advantage is real. Australian businesses in construction, allied health, professional services, and trades have deep, specialised knowledge. They've been operating in a market with high wages, geographic challenges, and regulatory complexity. That knowledge is incredibly valuable when transferred to AI agents.
The infrastructure is accessible. Cloud computing costs have dropped, and Australian data centres are widely available. Deploying AI agents on your own infrastructure no longer requires a server room. It requires a cloud account and the expertise to configure it properly.
The talent gap creates opportunity. Australia's tech talent shortage is well documented. AI agents can't replace every role, but they can dramatically amplify the productivity of existing teams. A construction company with three project managers and well-trained AI agents can handle the workload of a team twice that size.
The regional advantage matters. Regional businesses, the tradies, the allied health clinics in smaller towns, they've been underserved by technology for years. AI agents level the playing field. A physiotherapy practice in Lismore can deploy the same AI capabilities as a clinic in Sydney CBD. Geography becomes irrelevant.
For Australian businesses, the message is simple: you don't need to become a technology company. You need to hire digital employees who already understand your industry. AI automation agencies make that possible.
Will Every Company Become an AI Company?
Yes. But not in the way most people think.
Becoming an "AI company" doesn't mean hiring data scientists and building models. It means deploying AI agents as digital employees across every business function. Just as every company became an "internet company" in the 2000s (not by building websites, but by using the internet for everything), every company will become an AI company by employing agents for everything.
Customer service agents that know your products, your policies, and your customers' histories. Available 24/7, consistent, and getting better every day.
Operations agents that monitor workflows, flag exceptions, and handle routine processes without human intervention.
Sales agents that qualify leads, draft proposals, and follow up with prospects using your exact tone and methodology.
Finance agents that process invoices, reconcile accounts, and flag anomalies before they become problems.
Knowledge agents that capture institutional knowledge and make it instantly available to every team member.
The companies that get there first won't be the ones with the biggest budgets. They'll be the ones with the best domain knowledge and the right partners to deploy it. The agencies that help them get there, the AI automation agencies, will be the defining service firms of this decade.
The SaaS era gave us efficiency. The agentic era gives us intelligence. Not rented intelligence. Owned intelligence. Growing, compounding, proprietary intelligence that becomes more valuable every single day.
That's not a product update. That's a paradigm shift. And it's already happening.
Frequently Asked Questions
What is an AI automation agency?
An AI automation agency is a consulting firm that specialises in building, deploying, and managing AI agents for businesses. Unlike traditional IT consultancies that implement SaaS products, AI automation agencies build custom AI agents that run on a client's own infrastructure. These agents are trained on the client's specific domain knowledge and processes, making them more like digital employees than software subscriptions. The agency handles infrastructure setup, agent training, ongoing optimisation, and cost management.
How are AI agents different from SaaS products?
SaaS products are one-size-fits-all software subscriptions. You rent access to a tool built for the widest possible market, your data lives on the vendor's servers, and you own nothing. AI agents are custom-built digital employees that run on your infrastructure, learn your specific processes, and accumulate proprietary knowledge over time. The key difference: SaaS is a depreciating expense. An AI agent is an appreciating asset that grows more valuable with every interaction.
Why should AI agents run on my own infrastructure?
Running AI agents on your own infrastructure ensures data sovereignty, full control over agent behaviour, cost optimisation, and zero vendor lock-in. Your proprietary data stays on your servers. You control when and how agents are updated. You can select different AI models based on task complexity, optimising costs. And because the infrastructure is built on open source tools like OpenClaw, you own the entire stack and can modify or migrate at any time without vendor permission.
What open source tools are used to build AI agents?
Key open source tools include OpenClaw (agent framework), Multica (managed agents platform), GEPA (prompt optimisation at 90x lower cost than reinforcement learning), DSPy (language model programming framework), DeepSeek and GLM (competitive open source AI models), LangChain Deep Agents (orchestration), DeerFlow (superagent platform), and Hermes Agent (alternative agent framework). These tools cover the full spectrum from model serving to agent deployment to ongoing optimisation.
Is this approach suitable for small and medium businesses?
Absolutely. In fact, SMBs may benefit most from the agentic approach because they have the highest ratio of domain knowledge to technology resources. A 50-person construction company has decades of project management expertise trapped in spreadsheets and experienced heads. AI agents can capture and deploy that knowledge at a fraction of the cost of enterprise SaaS subscriptions. Open source tools have made the infrastructure costs accessible, and AI automation agencies provide the deployment expertise that SMBs typically lack in-house.
