Harnessing AI in end to end web development

Developers are now becoming prompt engineers.

Historically, we’d right code ourselves, filling in any gaps by looking through documentation, external resources and past experiences. The thought processes and development steps are the same, but thanks to AI, the workflow has shifted.

A typical project kick-off now would be to prompt AI to generate the initial structure and boilerplate, then refine and adapt it to fit the needs of the brief. To do this effectively, it means working with AI through a series of prompts to hone in on our desired outcome - ensuring context is retained and the end result fits our requirements.

Where we currently are, AI works as a really great collaborator rather than a replacement. Experience is still needed to evaluate what it produces, shape it, and integrate it properly into a project. A useful way to think about it is as a combination of a lightning-fast search engine and a junior developer - it gets you most of the way there, quickly, but still requires guidance and oversight.

We regularly use build analysis tools across our projects during development and as part of ongoing performance optimisation.

These tools provide a detailed visual breakdown of the entire application, including third-party dependencies, making it easier to spot bottlenecks and performance issues. They’re powerful, but interpreting the output can be time-consuming, and knowing what can be safely changed isn’t always straightforward.

This is where AI has started to make a real difference. By feeding the analysis output into AI, we can quickly turn complex visual data into clear, actionable insights, complete with explanations of causes and how to address them. It doesn’t remove the need for careful decision-making, but it significantly speeds up the path to meaningful improvements.

Working with significantly large datasets and databases comes with the territory, whether exporting data and reformatting or running complex queries. Being able to understand the data is a difficult and very time-consuming skill to master.

AI is supporting us in this area to build complex database queries. The kind of queries that used to take a day or two to get right, now can only take a few minutes.

Streamlining and optimising testing is a major part of how we utilise AI. Our testing processes cover multiple layers: 

• Backend unit testing

• Frontend acceptance testing

• Production periodic testing of broken links, performance issues, CWV, etc.

AI is used here to support the building and running of these tests. We can configure AI to crawl websites and detect issues, build its own tests, automate alerting when tests fail and even suggest fixes for the issues identified.

In addition to analysing a site and suggesting event tracking options, AI can construct the events and custom variables within Google Tag Manager, speeding up the process to achieve a more robust solution that delivers custom events to Google Analytics.

AI also helps when putting together custom reports in GA4, again speeding up the process and removing trial and error that can happen when navigating Google’s constantly evolving platform.

We’re using it to speed up the way we build and optimise websites, not replace the thinking behind them.

How we’re harnessing AI (in a nutshell): 

• Quickly generates components, APIs, tests, configs

• Identify obvious bugs

• Suggest cleaner patterns

• Modernise code

• Understand unfamiliar codebases quickly

• Supporting in the build of small snippets and specific targeted areas

While we embrace the speed of AI, security is our primary filter. We use locked-down versions of AI to ensure data privacy, using the tech to scan for vulnerabilities and stress-test code. It’s about building faster, but also building more securely.