This Week, I Put AI to Work on Real Deliverables
My biggest takeaway from a week of working with AI wasn't that my productivity "suddenly doubled." It was that all the small, fiddly tasks that usually stall me could finally be pushed forward one step at a time.
None of this is flashy, but all of it is real:
- Payment integration
- Production login
- Email setup
- Scheduled jobs
- User guides
- UX polish
They all come back to one question: can AI actually help a developer move a product forward?
Payment integration isn't just adding a button
This week, most of my work on MakanCloud went into the YeahPay integration.
Phase one was already live, but we hit a real-world problem right away. The payment gateway was selected in the admin panel, yet the option didn't appear on the checkout page or the scan-to-order page. You can't just start changing code on a hunch here. First you have to work out whether the setup is incomplete, a display condition isn't being met, or the business logic is supposed to work that way.
This is where AI helped: it broke the debugging into clear steps. Check the gateway config first. Then confirm the merchant info is complete. Then verify the display logic on the payment page. Finally, generate a real QR code and reconcile a small test order.
For phase two, I didn't let the AI jump straight into writing code. First I had it put the API tests, the console steps and the points needing human sign-off into a doc split into sections. That way each section can run on its own, and if something gets interrupted, I can pick up right where I left off.
Production login bugs: the scariest ones "look fine"
In another product, clicking the login link in the email dropped users on the homepage instead of their profile page.
Bugs like this are annoying because everything can work fine locally and only break in production. The AI and I eventually got the login redirect working, and then kept going on the email templates, domain config and one-time-code login.
One reminder from this: when you're dealing with consoles, tokens, domains and email services, AI can help you find your way around and explain what's going on. But sensitive credentials should only be used temporarily and cleaned up right after. "Letting the AI handle it" is no excuse to relax your security.
Taking a tool from "I don't get it" to "I can start using it"
I also spent quite a bit of time on my personal investment research system.
The first question was simple: how do you actually use this thing? Is it only for experienced traders, or can regular users get something out of it too?
So I had AI help me with a few things:
- Figure out the use cases
- Set up a watchlist
- Configure scheduled jobs
- Put together a user guide
- Adjust the default setup, which leans toward China A-shares, so it works better for US stock research
There was also a small but important change: strategy analysis no longer runs on fixed examples. You now enter your own stock or crypto ticker.
This isn't investment advice, and I'm not saying the tool can make decisions for you. The real value is that it took me from "I have no idea where to start" to "I know what to check each day, how to trigger an analysis, and how to use the results."
How I've learned to work with AI
After this week, I see AI less and less as an all-knowing developer and more as a collaborator that needs managing.
Here's what works:
- Break tasks down, write a plan first, then work through it in stages
- Anything that touches production has to be verified
- Anything involving accounts or keys needs a human sign-off
- After each task, leave a handoff note, a commit record and clear next steps
The real problem AI solves for me isn't thinking less. It's not losing the thread when things get complicated.
If you're building your own product, try using AI for things like debugging order, launch checklists, organizing docs, test steps and handoff prompts. None of it is glamorous, but it's genuinely useful.
I'll keep sharing what working with AI actually looks like. And I'd love to hear from you in the comments: what's one specific problem you'd most like AI to help you solve right now?