What it changed
- One engineer taking over a codebase that dozens of developers had worked on, and bringing it under control.
- A full audit of the requirements, performance, cost, security, deployments and code, with the security holes and AWS bad practices fixed first.
- Next, under way: staff describe a change in plain words, AI turns it into a specification and a tested change, and it ships safely. Anything complex comes to me.
The situation
This organisation isn't a tech company. Over the years it had ended up running its core programmes on three or four separate systems, each on a different stack, each old, unreliable and complicated. Administrators spent their days working around them, and human mistakes and wasted effort were the cost of doing business.
So it decided to modernise: consolidate everything into a single portal. It hired an outsourced development firm, a local management team in front of a large remote team of developers, the model where the margin comes from the gap between what the client pays and what the developers cost. Dozens of developers committed to the code. Deadlines and budgets slipped, communication broke down, and the organisation lost patience.
What I found
When I opened the code and the AWS accounts, the problems went beyond the delays:
- Security holes, and AWS set up against its own best practices.
- The wrong tools for the job, chosen without regard to cost or to what the workload needed.
- No reliable way to deploy or maintain it: code written by many hands, with nothing holding it together.
What I'm doing
The work started with a full analysis: the original requirements against what was built, then performance, cost, security, the deployment procedures and how maintainable the code is. The security problems come first, then the infrastructure and the delivery pipeline, so every later change is safe to make. One engineer with AI agents does what the large team didn't.
What comes next: changes without a development team
Most of what this organisation needs changed is small: a page, a field, a rule, a new report. Waiting on developers for each one is slow and expensive. So the next phase, already under way, is to let the people who know the business make those changes themselves, by talking to AI:
- A staff member describes what they need in plain words.
- AI turns the business need into a precise specification, then into the change, with tests.
- The change goes through the same checks as any other before it reaches production, so a mistake can't slip through.
- When a request is too complex or too risky for that path, it comes to me, and I build it.
Why it matters to you
A stalled project is rarely rescued by adding people. It's rescued by someone who can read the whole thing, decide what to keep, and fix what matters first. And once the platform is sound, AI changes the economics of running it: most routine changes no longer need a development team, only a process that makes them safe.