Build vs. buy, rewritten: notes from inside a retailer's switch to AI-built software
AI has made writing software fast and cheap. A national retailer is betting on it, and finding that the bottleneck has moved.
A national retailer with more than 80 stores is doing something most retailers wouldn't risk. It is replacing the off-the-shelf software that runs its stores, point-of-sale, and warehouses with bespoke software. AI has rewritten the economics of build vs. buy.
I'm advising on the transformation, and writing about it in flight.
The odd thing is, it all feels familiar. I've lived this build vs. buy decision before, twenty years ago, at a different retailer. But the case is radically different this time.
In the 1980s and 1990s, if you wanted software to run your business, you coded it. You got a tailor-made suit that fitted your business. The major wrinkle was that the business never stood still. In retail (my domain), new countries, new languages, new formats, new categories and changing supply chains meant constant changes to the code. Even a large engineering team couldn't keep pace. Writing code was the bottleneck.
That spawned the rise of commercial off-the-shelf (COTS) packages: enterprise resource planning (ERP) platforms, big and small. They did most of what a business needed, but not all of it: an off-the-shelf suit that never quite fitted. Some of the costs were explicit: licence fees every year, and upgrades on the vendor's timetable. Others were implicit: a business shaped around the software, and far too many points of friction: processes that cost too much to change, features that cost too much to build. The price was conformity. Businesses bent themselves to fit the same packages and, in doing so, got a little more average. That cost never shows on an income statement.
AI coding agents write code orders of magnitude faster, and better than the best human engineers. As Bill Gates put it on The Ezra Klein Show, they are now "superhuman at writing code",* and "most applications can be developed very inexpensively." Tailor-made software now costs less than off-the-shelf software, fits the business far better today, and keeps fitting as the business changes. All three matter.
Starting Decisions
Why. The retailer's 30-year-old off-the-shelf package was no longer fit for purpose. For five years it had searched for a replacement, but nothing off the shelf fitted its customer promise, and it wasn't willing to move until something did. That search gave it a serendipitous head start: detailed requirements, already written. So when AI coding agents arrived, a big chunk of the work was already done.
How. My advice was that the retailer could now own everything. Every line of code, every tool and every document. All owned from day one: its own code repositories, own cloud hosting and own databases. And, in time, its own team with the know-how to run it all. A third-party AI software factory would help get it going, then hand over to the retailer's team as it grew confident enough to take the lead. In short, the retailer would become a software business (inside a retailer).
Who. Not just any AI software factory. I found one built by ex-retailers who had hand-coded the systems that ran their own stores and warehouses, and then pivoted into AI software. They brought a ready-made production line for enterprise-grade software: solution architecture, data architecture and security architecture designed up front, and established tools and processes for the build.
When. We fed those requirements into the factory's production line, and it blasted through the initial build-out. We plan to switch the first store onto the new system by month 7, when about 70% is built: enough to run a store, then the rollout begins while the build finishes.
This is not a project with an end date. It's a product that keeps evolving with the business. As the retailer's own team proves itself, the factory's share of the work tapers, at the retailer's pace. The transformation is temporary, by design. What's handed over is the lead, not the relationship: the factory stays on as technical lead, because best practice in AI engineering is changing constantly. The graph below shows both clocks: the build, and the handover.
The Transformation
Writing code doesn't make you a software business. Eliyahu Goldratt's theory of constraints says a system moves only as fast as its bottleneck allows: the one constraint that sets the pace for the whole system. Remove it, and the next one shows itself. AI coding has removed the bottleneck, writing code, and exposed the next one: the people and the process around the code.
The software factory is built to produce software, and the retailer has never been a software business. Each side has blind spots the other can't see. Someone has to stand at the join and shape a way of working that fits both. That's my job.
That's the transformation.
In-Flight Learnings
We're four months in. About 60% of the software is built, but only a fraction has been reviewed and accepted by the business. The business is falling behind its software. That's what we're wrestling with now. What we're after is high-cadence, high-volume review: hundreds of small changes a day, screen by screen, menu by menu, smoothing the rough edges: a price applied the wrong way, a button in the wrong place, one step too many. We're not there yet.
One thing is already settled. No human writes code any more, and no human reviews it. AI agents write it, and other agents test it. This build is the proof. The agents are burning through it so fast that they've exposed the next constraint, just as Goldratt predicts. It shows up as two interlocking challenges:
1. The bottleneck has inverted. Code used to trickle out, and that pace set the cycle of roadmaps, quarterly plans and sprints. Now it comes like a fire hose, faster than the business can review it, test it and decide what it wants next. Old cycles have compressed out of shape.
2. Roles are merging. Business owners can now write specifications and build working prototypes themselves, using AI, so the business owner, business analyst, product owner, reviewer and tester are folding into new, multi-skilled roles: people who know the business, shape the product, judge whether what the agents build actually works for the business, and write the next round of requirements. (One name we're toying with is product engineer. What do you think?)
Writing code is fast and cheap now. Becoming a software business isn't.
What I Think (For Now)
I think owning its core software is where every retailer should be heading: the systems that run how it trades and set it apart. Software as a service isn't dead, but it only makes sense for non-differentiating systems: the general ledger, HR, payroll. Or for specialist capability you couldn't build yourself. What's dying is the idea that one giant ERP platform should run the whole business. Back then, bespoke software couldn't keep pace with the business. Soon it will be the ERP that can't: every change waits on the vendor's roadmap, while software the business owns can change in days. That's the full circle. Almost. This time the code isn't the constraint. Building the muscles to use it is.
The core is where a business competes, and it has to become the business's own. The businesses that keep renting their core will struggle to survive against the ones that write it. They'll still be average, and soon they'll be slow as well. But the real win isn't the lower cost, or freedom from the vendor's roadmap. It's the tailor-made suit, re-cut every time the business changes shape: an engine that keeps tailoring the software around the business. The proof of the pudding is in the eating, and we're still at the first fitting. More on that in a later post.
For now, the job in hand is the next constraint: how we work. The old delivery model, and its cycle of roadmaps, quarterly plans and sprints, is compressing, and the business is moving closer to the code. We need to shape new ways of working, and skill a team to run them. There's no framework or playbook for this. It's dynamic by definition, and we're figuring it out on the fly.
It's a jagged frontier, moving faster than anyone can predict, and we're on it. Whether that makes us the early bird that catches the worm or the pioneer who takes the arrows, we'll find out. I'm betting on the worm. If you're on the same journey, wrestling with the same questions, I'd love to compare notes.
*Bill Gates, The Ezra Klein Show, 29 September 2026.
Colin Fleming is a Retail Technology & Transformation Advisor and Fractional Executive, based in Cape Town.
