No. 01 · Conceptual

The Two-Billion-Dollar Ship of Theseus

Alberta's technical estate has outgrown the human capacity to manage it. AI provides a meaningful path forward to modernization at speed.

The technical estate overseen by the Government of Alberta has grown beyond human capacity to manage. The signs of this are present here in Alberta and shared anecdotally by other governments across Canada. This so-called "technical debt", the decades of application code, infrastructure, business rules, data, vendor dependencies, contracts, and documentation which have accrued over 40 years are overburdening government and inhibiting transformation initiatives. Despite best efforts, the vessel of government cannot keep up with the rate of change of technologies, with ever-increasing cost, and the ever-declining skills and knowledge needed to maintain these legacy systems.

Presently, our government has several hundred million dollars in Application Master Services Agreements with vendors which exist to keep these systems modern. However, these contracts are failing to deliver the outcomes we need. Application health is getting worse faster than it can be patched, and the latest AI models are uncovering cyber vulnerabilities faster than ever. In the last 12 months, the number of publicly known vulnerabilities in the underlying technologies that Alberta uses rose by 78.2%. Our own deep assessment of the estate uncovered a further ~33% increase above that baseline, so documented cyber concerns have more than doubled. Our speed at remediating them has not.


## §01 A generational problem

In Fall 2024, the senior leadership team with the Department estimated that the cost of modernizing these systems was roughly two billion dollars. The methodology used some simple heuristics; the cost is likely much higher. Regardless, the number serves as a reasonable starting point for a conversation. Within the Ministry of Technology and Innovation (TI), our budget to modernize such systems fluctuates between $80M to $120M annually. This means that if we did nothing else, and costs did not increase, and all technologies and their supports stayed static, and we dedicated our full time to fixing this problem, we could modernize these systems in roughly 20 years. That's a lot of 'if's.

The reality is demand for IT services has never been higher; the public have a strong appetite to gain access to modern, secure, and purposeful public supports. Ministries wish to modernize their programs, reduce red tape, and deliver results for Albertans. Vendor application support is ending on numerous critical systems in the coming 12 to 48 months, and big investments need to be made to address changes in legislation and regulation. As an example, the launch of the Alberta Disability Assistance Program and Care First insurance model are introducing significant program changes which will impact hundreds of thousands of Albertans. The systems supporting these programs have cost. Currently there are more than forty such legislative or regulatory initiatives underway. 

In the pre-AI era, we were not delivering new platforms and solutions fast enough. When we benchmarked our delivery velocity in 2024, we estimated it would take 130 years to lifecycle our existing systems. It is safe to say that no government, population, or vendor can tolerate or support technology projects that span into the centuries. Government continually responds to the needs of the people and its programs, policies, regulations, and laws are continually evolving. We needed to work differently.

To modernize at today's pace 130 years. At the current speed of remediation, the runway to clear the existing technical debt runs to roughly 130 years. No government, population, or industry can tolerate projects that span into the centuries.


## §02 The ship of government

In pursuing these changes, TI was implementing a 'Ship of Theseus' approach. The philosopher Plutarch used this famous ship of Theseus as a thought experiment, imagining over a similar timespan replacing every plank, deck board, mast, sail, and piece of rigging, and asked the question as to whether after this replacement the famous ship would still exist. Alberta was attempting to replace government digital systems in a similar fashion, substituting small pieces, patches here and there, occasional replacements, expecting over time for the vessel of government to stay watertight. 

But governments cannot stop, and services cannot be suspended, to enable technicians, developers, project managers, to rebuild our technologies. And we cannot afford to patch this ship over generations. We need new methods to rebuild a new vessel.

"We must rebuild this ship as it is sailing, and sometimes that means a lot of bailing out of water as we fix a breach in the hull." · Paper 1 · The Two-Billion-Dollar Ship of Theseus

Often the greatest inhibitor to transformation is the technical debt of the systems currently in place. Where Ministries wish to move fast, the systems we've built hold us back because they are rigid, complicated, and often inscrutable. Despite these seemingly insurmountable challenges, TI has developed a strategy to modernize government not on the scale of centuries but in a few short years.

These white papers present a range of approaches, technologies, and methods. Some of which are conceptual ways of humans working, some of which are technical documents which describe our methods, tools, and technologies in great detail. Most of these approaches employ artificial intelligence systems, which is the only way we can overcome this issue.


## §03 Why AI

For the layperson, what does the current era of generative artificial intelligence (just called AI henceforth) offer Alberta that previous methods and humans cannot? 
In the domain of technology, AI already offers human-level intelligence with superhuman speed. It is also scalable, allowing numerous parallel instances (known as agents) to break down complex problems using their inherent pattern matching skills augmented with common software tools. These agents also are persistent to the point of relentless, possess a greater attention span (known in AI as a 'context window') than many humans. Further, they speak and comprehend nearly every human and digital language ever created. When they don't understand, they predict, or 'infer', based on logical patterns that we cannot see. 
Where the average human is capable of naming a few dozen attributes of an object, person, or system, AI systems classify thousands of attributes for every word concept, document, pattern, image or part thereof that they see through a process called 'vectorization'. This depth of understanding means that they have intuitions we do not consciously possess, and in the technical space this enables them to 'see' the shape of concepts and patterns rapidly, often in seconds, where a human would laboriously work to prepare similar insights over months. And while humans can read between 100 to 400 words per minute, AI can process millions of words in the same time, finding either a needle in a haystack or the pattern in the noise across vast knowledge and code bases.

Perhaps nowhere are such skills more valuable than in understanding the complicated systems of government IT. With these capabilities, AI can be deployed to analyze our entire technical estate, a span of code currently exceeding 466 million lines, and read every pattern, dependency, relationship, and function across thousands of applications in less than a day. A year from now, it will likely be able to do this in less than an hour. Later papers in this collection show exactly how this works and this process revealed.

AI is more useful than just admiring the problem. It can fix what it finds.


## §04 The challenges and the fix

Properly engineered, these AI agents offer the potential to both understand and rebuild government systems at speed. Their generative nature, and their deep understanding of technology, terminology, and the technical environment within which they operate, make AI uniquely qualified to analyze and then repair our software and infrastructure. Where we have a fixed and finite number of human staff, TI now has the ability to create a parallel workforce of AI agents in the tens of thousands. Whereas we only typically look at apps once they break, agents can be assigned to do nothing but care for the health of every single app, test and fix every single bug observed, monitor for every known cyber vulnerability and then remediate the application. It is reasonable, and practical, to create 1 to 10 agents which do nothing but obsessively evaluate, monitor, and if need be, repair every application in the estate.

Yet AI is still far from perfect. Despite their obvious exhibitions (or simulations) of intelligence, these AI workers are also prone to distraction, to flights of fancy, to misunderstanding, and they inherently lack any long-term memory. Figuratively, they are genius amnesiacs, capable of answering tough problems only to instantly forget their task, all previous engagements, and you between every interaction. This may be an artifact of the current generation of these models, but it is a big deal presently. Without memory, the continuous chain of thought, planning, and evaluation is easily broken, and progress stalls. At their best, they excel beyond human performance in the technical domain. At their worst, they are unreliable witnesses and can make grievous mistakes or errors in judgement due to incomplete information. The companies that build them have also made them frequently sycophantic and fawning in their praise of even bad ideas, with an overeagerness to please which is at best counterproductive.

On their own, AI agents also lack agency. Their abilities lie only in the words and the media they create. And it is up to us to build the tools to translate those words into action. And despite their general knowledge, which is both impressive and challenging with its unknown breadth and depth, these agents know little about the internal methods of government and nothing of our specific policies and processes. They need instruction, process, clarity, and guidance, just like any newcomer to an organization would require to be successful.

Despite their numerous, obvious, and sometimes infuriating flaws, a human and an AI agent paired together can frequently do the work of 20 people, and often in a fraction of the time. This is a bold claim, but one which we have proven in Alberta time and again over the last 18 months, and which we will lay down the evidence, the approach, and the code which demonstrates the truth of these claims. And despite these challenges, these tools are our only hope in transforming 40 years of code, and billions of dollars, into a modern and secure government.

To do any of this, our human staff in Alberta first need to learn how to direct these agents, overcome their limitations, and build new processes which maximize human and AI creativity and productivity. Working effectively with AI is not easy or obvious. 


## §05 What is to come

Over this series of white papers, Alberta will lay out how we have built agentic AI solutions which will allow us to challenge, and slay, the four-headed hydra of **technical debt**, **cyber security**, **growing demand**, and our **deficit budget**. These papers will walk you through our AI Academy, our agentic harnesses, our new systems and insights, and how we are evolving our management processes to resolve these problems at speed. We are sharing our four approaches to technical debt, each more transformative (and radical) than the last, as well as findings which have emerged through our adventures into AI-driven outcomes.

The Velocity White Papers are intended to serve as a blueprint for other governments to emulate what Alberta is doing. We are still actively doing this work, but we cannot afford to wait in sharing and collaborating. Unchecked, technical debt will drive us into a 'cyber winter', where the threat actors will access for pennies what it will cost us to remediate for thousands. As you will see in the next paper, the cyber issues here are real and solving them is an imperative that all governments must face sooner or later. Linked with increased demand, declining budgets, and the shifting technological landscape, we have an imperative to modernize. We believe that this must happen very soon, and AI is the only path to making it happen. By following these methods herein, we believe that we can reduce both the time and cost of digital transformation.

The number Alberta is pursuing 95%. With the right engineering, training, and strategies, Alberta believes it can reduce the cost and time of implementing these remediations by roughly ninety five percent.

These white papers are how we do it, and how we measure our successes. We will share our progress transparently and continue to publish these white papers openly. We are investing heavily in our people to ensure nobody is left behind in the pursuit of AI adoption, while also reimagining the roles of the future.


We invite you all to join us on this journey. Let's get building.

Tags: thesis, technical-debt, cybersecurity, ai-agents, modernization

Open the interactive version