Patrik
Wise word of the week
Plans are made to be changed.
Automate everything.
Journal
I’m about 3-4 weeks late on my development schedule. Feedback happened!
Instead of launching and letting everyone in, we decided to ramp up our product slowly. Very slowly. We are inviting batches of users from our waiting list when we are ready. There are two main reasons for this.
Technical reason: we are not fully scalable yet, so every additional user is some work we need to do. This must be fixed and we are working hard to automate everything so that we can decouple the product from the number of users. Some activities like user support will never go away, but at least the complete usage life-cycle should require no interaction from us. Right now, some parts of the user management are still manual, in particular the steps taken near the end of the subscription. Marc is currently sending the warning and reminder emails manually, I need to fix that.
Product reason: at the moment, feedback is the most valuable information we get. We have a good idea of where we want the product to be, but the users keep us grounded and ensure we are developing the right product. But there is also a lot of more tactical feedback, like missing features, frictions, and bugs. Currently we invite 20 people, collect the feedback (some arrives almost the same day as they sign up!), improve the product, and iterate.
The last batch of users was really good and we got an enormous amount of feedback, including two bugs in the simulation. As I really want the numbers to be as correct as possible, I spent the last weeks fixing those bugs instead of adding the features that I promised to launch by the end of April. But I’m happy that today after checking one more thing, I’m able to launch the last fix and go back to feature development.
How did I use AI this week?
The last week saw a significant change in the way I use Cursor.
I see four ways of using it, in growing order of task complexity.
- Autocompletion. This is very local, works reasonably well when doing changes, because Cursor will try to apply the same change to the code. But during “development from scratch” is not great (understatement), and IMHO works only when doing pseudo-coding by writing a comment of what needs to be implemented. When developing, I find it very distracting and often it adds errors in a subtle way. I’ve disabled it completely.
- Quick Edit. Can be useful for very local changes that don’t require much context. Mostly to avoid typing boilerplate or obvious code. I use it occasionally.
- Agent. The chat. Allows more complex tasks than the quick edit, the agent creates an internal plan and executes it. There is a dialogue with the AI that allows you to amend the results. So far I was using this mode the most. And spending time amending the results.
- Plan. In this mode the plan is explicit and I can discuss the plan with Cursor before applying it. I’m now using this mode heavily, basically instead of using the Agent mode.
Why did I switch from Agent to Plan mode? Plan mode has the great advantage that I can review the changes before they are applied to the code.
This way I can discuss the higher level design decisions and tune the plan instead of tweaking the code. It’s actually easier to work at a higher abstraction level than at the code level.
Does this look like waterfall instead of agile? Yes and no. Yes, I’m planning rather than just coding, but unlike the waterfall approach where planning takes a long time, here planning is fast and almost interactive. It’s easier to spot issues in a plan than reviewing dozens of code changes.
So, right now planning mode has almost completely replaced agent mode for me.
Marc
The build
The past two weeks were rather calm for me regarding new features for FI Planner. Longer weekends, shorter weeks. The only thing I could ship was the “What’s new” menu item, pointing toward our release log… what an achievement!
At least, now you have a way to quickly jump to the latest updates of FI Planner.
Yet, now that Patrik nailed down the fix (and big refactoring) for AHV/AVS, we’ll restart the onboarding process of users on the waitlist. Stay tuned! (and feel free to reply to this email if you absolutely wanna be part of the next batch of invitations).
The edge
I was talking to a friend about this earlier this week: even if I knew about it, my brain went into wishful thinking mode that I’d find more than the 24h in a day to move forward with the FI Planner frontend roadmap I have in mind.
Except that, sickness happens. Family and kids “emergencies” need to be taken care of when they happen, not in “let’s move this ticket to next quarter”… It’s always hard to admit for someone who fits into the “over-achiever” bucket, but so be it. Humbling experience.
I can’t wait to be FI, which may happen near the end of the year / beginning of next year. Until then, I will keep prioritizing ruthlessly.
Note to a friend
Make sure to align with your cofounder on your ambitions before you kickstart any new venture.
If we hadn’t done this with Patrik at first, and he expected me to work nights and weekends, we’d have a problem these days. Or I would be in a bad mental state due to guilt and exhaustion.
Tool of the week
Adobe Acrobat Reader Pro… I never thought I would write about this tool and company that I dislike for various reasons…
Here is the story (relating to my programs’ slides):
- iLovePDF is great for compressing and OCR.
- Except that their OCR tool is not keeping web links…
- Which brought me to the good old (and expensive!) Acrobat Reader Pro version (thankfully there is a free trial), which has a nice feature of “Compress a PDF” which automatically adds the OCR layer without breaking the links in it