If you talk to any experienced developer, they will all agree that the thought of an AI writing 'greenfield' code as well as they can is laughable. Many of us either avoid it, use it just for problem-solving, or adopt it with a "ok but I can't look at it" approach, but it's full of approaches that you'd just never write yourself.
And at the same time, it can be incredibly helpful, even for CRO experiences.
Jump ahead a few lines
Github Copilot does an excellent job of drip-feeding you code. It's small enough that in isolation it's fairly insignificant, but it saves you a few minutes every few minutes and this adds up hugely over time.
Further, there are things that are just painful to write at times, like creating a modal from scratch, or writing validation rules in your form submission to keep data as clean as possible. But AI doesn't care, and you're far from the first person to write code to solve these problems.
With really simple prompting and autocompletions while you type, the time-to-value of coding has tightened greatly. Not to a point where someone experienced isn't needed anymore, as we've seen through countless API key leaks over the past year and an estimated 11% of all vibe-coded apps leaking keys. But to a point where someone smart can do these jobs really quickly.
Help me to help you
Another interesting aspect of vibe-coding has been how you can "train the AI" with relevant bits of code. Some input their massive monorepos and have every line of code ever written. Others use Vector Databases to find a few similar blocks of code to what you're doing to help guide the AI.
So whether it's code that you specifically have written, or that anyone has in your "account", the potential is there to retrieve it in a smart way and feed it in as extra juice for your next AI suggestion.
So, "make me a pop-up" isn't just a random pop-up, but it's using pop-ups you've built in the past to figure out an general aesthetic.
The more context AI has, the better it gets. So feeding it your past code as suggested above, means it syatys to reflect your patterns and style, not just generic output.
So again, this means the time-to-value of your code shortens. Freeing up more time elsewhere and for more opportunities.
When is a good time for AI to be present?
We believe in a few places.
One is directly, as you're building. You type, it suggests. You ask, it advises. This is very Copilot-esque functionality that developers should be familiar with in the modern age.
The other is for it to act as a guardian-angel, watching over you and course-correcting when it sees you making mistakes. And for us, this is where domain-specific knowledge and intelligence lives.
If you had a Lead Dev sat watching you code, before you click save they'd be pointing out that your code has errors, and even suggesting how to fix it. They'd be proactively scanning for problems and flagging them to you. If you ask to Peer Review, they'd have both a playbook and experience to come back with meaningful changes to make - beyond "indent this an extra space" or "maybe turn this into a variable".
This would allow junior developers to be fuelled by our domain-specific knowledge, helping with the ongoing retention crisis for developers in CRO as a whole, and giving businesses the opportunity to employ and bring in more junior developers.
This gives AI a purpose, which is the overarching theme of this series of blogs. There are practical problems that you can start to solve which is genuinely valuable to you and your organisation. AI-assisted code can be incredibly value for CRO and help to speed up development, processes and results. Giving even more back to you.
Back to the problem
So when you've got developers or semi-capable part-time coders who are just trying to get through things without errors, or problems that are taking place that you need to be advised of and guided through, these are all problems AI can solve today. And for us, that's AI with Purpose - far beyond the promise of going from prompt to fully working, lead-dev-approved code.
