AI drives software program productiveness – and challenges – for Motorway
For many years, engineering groups handled code like a classic Ferrari – costly to construct, painstakingly maintained and too treasured to ever throw away. Each line represented a big funding of human capital and time, and has led to a tradition the place code was cherished and its longevity was a marker of success.
However on the AWS Summit in London this week, Ryan Cormack, principal engineer at on-line used automotive market Motorway, consigned that philosophy to the scrapyard. Within the age of agentic synthetic intelligence (AI-)pushed software program improvement, he says, engineering groups can change into extra productive and are capable of construct, revise and preserve code at speeds beforehand unthinkable.
On this article, we take a look at Motorway’s radical shift from handbook coding to an AI-first improvement pipeline powered by AWS Kiro. Cormack talks about how the corporate achieved a 4x improve in engineering output, the challenges that include the power to supply extra code, why the way forward for software program improvement lies in treating code as disposable, and the core advantages of codifying organisational tradition into AI steering information.
The mindset shift: Disposability vs polish
Essentially the most profound change at Motorway is pace of supply but in addition a psychological break from the previous. Traditionally, writing code was a “time-expensive course of”, Cormack says, including: “We needed to have code that was so good that we might cherish it for years to come back, as a result of we had invested a lot time into making it.”
However since beginning to use Kiro – AWS’s agentic AI-capable IDE – that mindset turned a bottleneck. “We shifted away from, ‘We’d like essentially the most well-polished code for each line we write, on a regular basis’, as a result of we are able to rewrite it once more tomorrow at a pace that’s by no means been doable earlier than,” says Cormack.
This has led to a technique of “analysis over manufacturing”. Motorway now generates huge quantities of code – one million traces a month – a lot of which can by no means attain a buyer, says Cormack. As a substitute, it’s used to check and consider a number of alternative ways to resolve an issue earlier than committing to it.
The lesson for different organisations is obvious. Don’t goal for an ideal first move. Use AI to cycle by iterations, then use human experience to refine precisely what you need from the choices the AI helps present.
Managing the ‘quantity disaster’: Rigour over pace
Whereas a 4x improve in output appears like an engineering dream, it creates an actual “evaluate bottleneck”. When you write 400% extra code however preserve 100% handbook evaluate processes, the system collapses. To fight this, Motorway hollowed out the “handbook center” of the event course of and moved human vitality to the ends of the method – specifically, the spec and the evaluate.
“We discover ourselves spending extra time planning code and the entire course of up entrance, and just a little bit extra time reviewing what comes out,” Cormack says. “However we lose all this time within the center the place we beforehand needed to manually write all of the code.”
To make sure AI doesn’t simply produce any code however “Motorway code”, the group utilises “steering information”. These information increase the AI’s system prompts with the corporate’s particular DNA. They’re particular to Kiro and are markdown paperwork that include directions, requirements and preferences to information the AI behaviour and coding model.
They embody, for instance, naming conventions that standardise how software programming interfaces (APIs) are labelled throughout Motorway’s 7,500-dealer community, and design patterns that implement particular software program architectures.
By injecting these guidelines by way of the AI, generated code seems to be and feels prefer it was written by a veteran Motorway engineer.
And AI isn’t simply used for the construct; it’s used for the total lifecycle. “We have to use AI to assist us debug, analyse, perceive, and consider methods as they run,” Cormack provides, noting that brokers now monitor logs and metrics to assist people handle an enormous fleet of companies.
The ‘Kiro’ engine and mannequin agnosticism
A important part of Motorway’s success is that Kiro acts as an agentic loop fairly than only a easy “autocomplete” software.
“Kiro is aware of how our CI pipelines work,” says Cormack. “It is aware of how our infrastructure is code-driven and it is aware of how our inside functions work collectively. It’s capable of assist information us each step of the way in which.
“We’re utilizing Kiro throughout our full software program improvement lifecycle. Our product and UX groups can ship actual prototypes into our clients’ fingers faster than we’ve ever been capable of earlier than. What would take weeks now takes hours.”
His group can leverage its mannequin agnosticism too. Cormack defined they aren’t locked right into a single LLM: “We use Kiro with Claude’s newest Opus 4.7 mannequin, we use it with a number of the open weight fashions, issues like Meta’s Llama fashions … we’re capable of selectively choose the LLM that we all know goes to have the ability to greatest carry out the precise job.”
This flexibility helps to mitigate the danger of hallucinations. Motorway depends on a spec-driven strategy the place the AI should assume by the issue and generate a technical design earlier than writing a single line.
“It can assist us write automated exams which are capable of show that every of those factors has been precisely achieved,” Cormack says. This implies the AI offers its personal proof of labor earlier than a human ever touches it.
Legacy transition from Heroku to AWS
Motorway wasn’t all the time this agile. The corporate was “born within the cloud”, on Heroku, which Cormack acknowledges was “nice for scaling and getting going”. However as the corporate grew, it hit friction factors.
The transition to AWS was pushed by a necessity for “flexibility, adaptability, and scalability”, says Cormack, who views their Kiro-enabled AI-first pipeline as the final word software for such transitions.
If he had been to do issues once more, Cormack says he would “undertake this mannequin of pondering a lot earlier on”. The flexibility to make use of AI to map migration logic and repair dependencies would have saved months of handbook effort throughout the transfer off their legacy platform, he believes.
Classes for the boardroom
For organisations that wish to replicate Motorway’s 250% improve in deployment frequency, Cormack warns towards automating the grind of coding with out additionally automating the rigour of testing.
“When you attempt to construct simply by writing code quicker, it doesn’t remedy the issues,” he says. “I don’t assume our clients essentially need code; they need options and performance.”
The winners of the AI period received’t be those who write essentially the most code, however the ones who construct essentially the most rigorous frameworks to handle its disposability.
As Cormack says: “Kiro’s now writing over one million traces of code for us each single month. So, earlier than we begin any new piece of labor, our engineering group chooses Kiro to assist perceive precisely what it’s that we wish to construct.
“The rigour at first of this course of helps allow the precision we wish in our engineering on the finish. So, every bit of labor that we do begins with a spec, understanding the intent of what it’s that we’re constructing and why.”
