Partnership announcements are easy. Anyone can put two logos on a slide and call it strategic. At LEIT Data we have always taken a harder line: we only partner with a vendor when we can see, from experience, that the technology changes the economics of delivery for our clients. Not the demo. Not the pitch deck. The economics.
That is why I am pleased to announce that LEIT Data has become one of the first boutique partners of Factory in the UK and EMEA. We are at the beginning of that journey: Factory’s agentic software development platform is already in use on our latest engagements, and this post explains why we made that call, what we saw before we made it, and why I believe it matters far beyond our own delivery teams.
The Snowflake Lesson
Some history. Before co-founding LEIT Data, I worked for the first UK partner of Snowflake. This was back when Snowflake was a name most data leaders in the UK had not heard, and the safe money was on the incumbents.
What I saw there was something genuinely different. It was not just that the technology was good. It was that the value it added could be directly attributed to the cost of consuming it. You could point at a workload, point at what it cost to run, and point at the business outcome it produced. The bill and the value arrived in the same conversation. That alignment is rare in enterprise technology, and it is a big part of why Snowflake rewrote the rules of the data platform market.
I have spent the years since looking for that same pattern in other technologies. Most do not have it. They add cost in one place and promise value somewhere else, and the two never quite meet in the accounts. When I looked properly at Factory, I saw the same pattern I saw all those years ago: a platform where the spend and the value can be tied together, line by line, outcome by outcome. That is the synergy that made this partnership an obvious decision rather than a leap of faith.
The Problem With Individual AI Licences
Let me describe a situation that every CTO and CFO reading this will recognise, because it is playing out in thousands of organisations right now.
The individual AI coding tools are genuinely impressive. I have used Cursor. I have used Claude and Anthropic’s models directly. As an individual, they are fantastic. A single developer with a good AI assistant is measurably more productive, and I would never argue otherwise.
Now try to scale that across a team.
Give every engineer a thousand dollars a month of AI spend and watch what happens. Every one of them is reinventing the wheel. Every one of them is solving the same problems in slightly different ways, in their own sessions, in their own style, with none of that logic reused by the person sitting next to them. There is no coordination. There is no shared context. There is no organisational memory of what worked.
Worse, you cannot assess it. You cannot answer basic management questions. Who is using it well? Who is burning tokens on dead ends? Are we getting the best value from the spend, or just the most activity? Nobody knows, because the tools were designed for individuals, and individuals do not need to answer those questions.
And there is a third problem that gets less airtime: model lock-in. Go down one route and you are using one vendor’s models. Go down another and you are using a different vendor’s models. Either way, you are betting your engineering productivity on a single model family, with a single pricing curve and a single set of strengths and weaknesses.
What We Found in Factory
My experience, and our experience across client engagements, is that the best results do not come from any single model. They come from a combination of models working together, each used for what it is genuinely good at, combined with open-weight and local models where they provide a more cost-effective route to the same outcome.
That is the first thing Factory gets right. It is not a bet on one model. It is a platform that lets you use the frontier models where their strengths justify the cost, and route work to cheaper models where they do not. The model becomes an implementation detail rather than a strategic dependency.
The second thing is the one that matters most to me, because it is the one that decides whether AI engineering survives contact with a real organisation: governance and guardrails that hold at scale.
With Factory, the agents do not operate as a hundred uncoordinated individual assistants. They work inside an organised framework: shared context, reusable logic, defined standards, and visibility into what is being done and what it is costing. The wheel gets invented once and reused. The organisation can see how the capability is being used, and whether the value coming out justifies the spend going in.
This is the difference between AI as a collection of personal productivity hacks and AI as an engineering capability. The first is nice to have. The second is something you can build a delivery model on.
Tokens Are a Cost of Sale, Not an Overhead
This brings me to the point I care about most, because it is the point your CFO cares about most.
Right now, in most organisations, AI token spend lands in the accounts as a large, undifferentiated operational expense. It sits there looking like a cost centre with a vague story attached to it. Finance sees the outflow. They do not see what it delivered. That is a bad place for any spend to be, and it is why so many AI initiatives are one difficult quarter away from being cut.
It should not be an overhead. In a consultancy like ours, and frankly in any product or engineering organisation, AI spend is a cost of sale. It is part of the cost of delivering a specific outcome for a specific client or a specific product. It belongs on the same line as the work it produced.
That is only possible if you can attribute it. You need to know which tokens were spent on which engagement, producing which deliverable, under which controls. This is exactly what the Factory model gives us. The spend is visible, governed and attributable. Our CFO does not see a large unexplained OPEX number and a request for trust. He sees a cost of delivery, linked to what it delivered, measured against the revenue it supported.
I have said for years, in a different context, that Snowflake is expensive if you leave the doors and windows open. The same is true of AI tokens. Ungoverned, unattributed, uncoordinated AI spend is expensive. Governed, attributed, coordinated AI spend is one of the best investments an engineering organisation can make. The difference is not the technology. It is the operating model around the technology.
What This Means for Data and Engineering Leaders
If you are a CTO, CDO or head of engineering working out how to scale AI-assisted development, here is the practical guidance I would give from where we now sit.
First, stop measuring adoption and start measuring attribution. The number of licences you have bought is not a metric. The value delivered per unit of spend, per team, per engagement, is a metric. If your current tooling cannot tell you that, you have found your first requirement.
Second, treat reuse as a first-class outcome. If ten engineers solve the same problem ten times in ten private sessions, you have paid for the same answer ten times and kept none of them. Shared context and reusable agent logic are not nice extras. They are where the compounding return lives.
Third, resist single-model thinking. Model capabilities and prices are moving too fast to marry one vendor. Build on a platform that lets you route work to the right model, including open-weight and local models where they make economic sense.
Fourth, put the governance in before you scale, not after the first incident. Guardrails, review gates and spend visibility are far easier to establish when ten people are using the platform than when two hundred are.
The LEIT DATA Perspective
This is where the partnership becomes practical rather than philosophical. At LEIT Data we are at the start of our journey with Factory, already using it on our latest client engagements and building it into how we deliver: agentic engineering with governance and cost attribution built in from the start, not retrofitted when the bill arrives.
For our clients, that means AI-accelerated delivery where the value and the cost stay in the same conversation. For us, it means a route to scaling this way of working across an organisation without losing the guardrails that keep quality, security and spend under control. That combination is what we were looking for, and it is why we chose to start early rather than wait.
Looking Ahead
The next twelve to eighteen months will separate the organisations that bought AI tools from the organisations that built AI capability. The tools phase was about individual productivity. The capability phase is about coordinated, governed, attributable delivery, where every token spent can be traced to an outcome.
If you are working out how to make that transition, or if your CFO has started asking awkward questions about the AI line in the budget, come and talk to us at LEIT Data. We are on this journey ourselves, we are starting it alongside our first clients, and we are happy to share openly what the early numbers look like when the guardrails are done properly.
Originally shared on LinkedIn as part of the #MeanDataStreets series: https://www.linkedin.com/in/chris-tabb-mean-data-streets