So what else have I learnt from using AI?

TL:DR in the same way that programme teams have a mix of skills to become the sum of all parts, to get the best out of Ai, use more than one reference. Using the maturity (experience) of various tools benefits your vision, same as having cross functional teams.

So, how’s things going?

Those that saw my first post (https://medium.com/@dsearle-ovate/so-what-have-i-learnt-from-using-ai-908b0a63fe2d ) will have seen how I created an app, that brought my 2020 vision (sorry!) to life.

So, you’ve deployed it and you’re making millions?

Short answer, no!

As I mentioned in my last post, my next steps were to get the app production ready. This is where the problems started.

Cursor took my instructions and started updating the code, I set the context “remember x, check y” in my instruction, however the tool seemed to be getting in to an infinite loop of pointing to localhost, switching to api base url or “forgetting” to update all parameters.

I was using my included credits and not making any great strides or progress. Fix one bug, create another.

I’d be lying if I didn’t say that I was slightly disheartened.

“It’s not you, it’s me” was what I was thinking. I decided the best thing for our relationship was to take a break and create some space.

I’m went off and did some soul searching (well google searching) to see what I was doing wrong, these searches only pointed me to other tools, including Cursor…..

I missed the fun that we had together and knew I had to make things work between us…… I looked for advice elsewhere…

Enter ChatGPT

So, what benefit has that had?

My first instruction to ChatGPT was to ask about context and references points for AI coding tools. The screen starting populating with ideas for prompts.

I moved to more focussed questions around the deployment needs and making these apps production ready.

Again, a clearly defined checklist was created and allowed me to go back to Cursor and start addressing some of these bug loops.

I now had a delivery model / team in place…..

• A Product Manager (Me)
• An Architect (ChatGPT)
• An Engineer (Cursor AI)

This approach was making some good ground and was allowing me to set the correct context to apply.

What it also outlined to me, is that I’d got caught up in the “isn’t this clever”, adding in other complexities (I.e. an admin webapp, with cross platform API dependencies, when I already had admin capability on the iOS App)

Making the realisation that I had added some quite complex technical dependencies as well as increasing the probability of bug loops, I decided to pause development of the admin web app and focus purely on the iOS app, with its different tiers and dynamic views based on permissions (including Admin).

“Hmm, if I’m going to admin this and create content, I need more real estate to pull this together! I’ll deploy it on phone and tablet!!”

“Make the iOS app mobile responsive, so it can be used on phone and tablet”

Before asking Cursor to make responsive, I used the XCode simulator to deploy to an iPad. As expected, the view was just the mobile one in the middle of an iPad screen.

Once I asked for the app to be mobile responsive, the tool kicked in to gear, reviewing all areas of the code. I kept an eye on the usage dashboard and could see my included credits disappearing with each refactor.

After about 5 minutes, the updates completed.

“Framework now in place, would you like me to now apply to all the screens”

Hmm… this isn’t what I was expecting….. what else have I missed?

I went back to ChatGPT asking various questions around making app production ready. It soon dawned on me that I was going to need to reevaluate what tech stack was deployed and what was needed to make the app work when deployed.

I’d become wrapped up in the “function” of the interface, I’d neglected to think about what set-up and costs I would need to make this (or any other app) work for users.

Yet another example of the need to build a “team” to get the best out of AI tools, I now conclude my team needs to include:

• A Product Manager (Me)
• An Architect (ChatGPT)
• An Engineer (Cursor AI)
• A Service Ops (various under evaluation)

“So, you’ve failed”

No, not at all, again looking at typical programme structures, we have a vision, a goal and a need to prototype designs, proof of concepts (spikes), before centring on tech stack to drive things forward.

The main difference, this process has taken just over one week, and the only cost to date is my time and my first month Pro subscription to Cursor.

“Fail fast” is a term banded in agile teams regularly, the cost involved in failing fast is the key to “delivering incremental pieces of value”, aligned to a need, user experience or company goal.

As I mentioned in my first article (https://medium.com/@dsearle-ovate/so-what-have-i-learnt-from-using-ai-908b0a63fe2d ) the use case for these coding tools I see as being an imperative for Product/UX teams. Being able to “code” a branded, functioning prototype, that can be made available to user testers for feedback loops before entering into full development sprint cycles, for me, this has the potential to reduce waste in the early concept / ideation phases of projects.

Yes, there will need to be some boundaries and failsafes applied to these tools use I.e. intellectual property and protection of market movements, however even if you were not to brand the prototype and kept features at the highest level, then realistically all you are calling upon is code approaches and frameworks that are already in the public domain. Rather than a blocker, this becomes a due diligence checklist to ensure that IP and data are secure as part of these prototype/ideation phases (don’t connect to live data stores!)

So, what’s next?

In short, I am reviewing my learnings, my gaps and looking at the easiest path forward, creating my checklist and “virtual” amigos checks to ensure that I avoid bug loops and unnecessary refactoring (and a loss of coding credits)

As I concluded last time, humans need to learn with AI, commands need to be concise, clear but also have ongoing context and a view of the primary vision and goal.