Hi again! :) Two weeks sure flies by, I’m glad to hit pause for a moment and catch up on all that’s happened.

It’s been a fun couple of weeks. Plenty of time outside, and about as much progress as I could have hoped for with Herd while SIMBS hungrily eats up time in preparation for our AGM on March 3rd. Runs, Broadway, biking and first aid — what more could I ask for!

Last newsletter, I spoke about intentions for the year. Goals are important for me too though — they help me realize those intentions. This year, I set a couple goals that truly intimidate me:

Why might these be so intimidating? Well, I’ve been battling a chronic knee injury for the past 6 years. It’s prevented me from running more than 4km without severe pain; as a result, it’s robbed hiking and multi-day backpacking from my mid/late-20s. And launching this product is something I’ve thought about for years, and this feels like my one shot.

While goal setting is no stranger, these goals feel different. I’m in full control and I’ll have to live with the consequences, so it’s easy to be fully invested. Goals with a clear levers to pull, and which have consequences, are motivating like nothing else.

Progress has been lumpy. Sometimes really fast, sometimes slow. This is pretty natural in the early stages of a product in my experience. I build out an aspect of the product quickly, then step back to evaluate, re-orient, and clean up what I’ve just done in a flurry. However, getting used to the new pace developing with AI has made the fast parts even faster (and the cleanup from the mess, even bigger).

While we have questions to answer about how AI should be used in various aspects of our lives, when used right, programming with it is not just faster, it’s (usually) more fun!

I’ve found progress quick when working in tandem with an AI agent, like working with a capable peer. I can take the mock-ups I shared with you all in my last newsletter, use UX Pilot to build high fidelity mock-ups, spend some time tweaking them, then collaborate with an agent to build it together. As I build out a feature, the agent pulls in best practices when working with new libraries, and tends to do much of the grunt work I don’t care for. When it’s working well, it’s re-ignited a spark that made coding fun!

But it’s only good because I’ve learned from letting things going off the rails. A number of times I’ve found myself repairing and reworking code more than writing it. This tends to happen when I pass off entire features or modules to a coding agent like Claude Code or Codex. It will build out the bulk of the work, but usually make a few weird (or just inconsistent) choices along the way. The longer the agent runs, the more the bad work multiplies as it strays from what I want it to do.

It’s unclear why exactly this happens. It could be because my codebase is still nascent, the public dataset of mobile apps is much smaller (than other coding projects), or my agent context (eg. CLAUDE.md) isn’t yet robust enough. Or maybe my opinions are too strong! Either way, these tools are good but not great yet. At each point, there’s still a fair amount of manual work in fixing outputs that aren’t consistent, lack important context, or include a number of silly mistakes (for some reason in mobile apps, they seem particularly bad at routing and organizing modules). So frequent course correction is necessary.

While the media tells my discipline again and again that “our jobs are cooked” and “knowledge work is dead”, I don’t foresee a future where we won’t need some amount of people. If models plateau, we’ll continue to need people to direct and supervise work. If models keep getting better, there’s two things I don’t think AI can replicate: vision, and taste.

Here’s a preview into what this app could really look like. Details will obviously change, as could the style. But this feels a bit more real, doesn’t it!

There’s a few features in the mix that are still up in the air. I’m not sure if it’s within scope to include:

Thanks to some of you who reached out after my last newsletter, I had the opportunity to do a little user research!

After speaking to you, I found a couple interesting patterns:

And there were some key challenges planners faced:

The story that I kept hearing was: “I’m the planner in my group, and I have a hard time getting commitment from my friends. As we get closer to the event, the details I post keep getting lost in the chat or people don’t seem to know what’s going on.”

This is exactly what Herd will be built for. Helping planners create plans in the short to mid term. And using features to engage the participants so everyone knows what’s happening.

With any luck, it’ll be so easy to herd your friends, everyone can do it a little more :)

How do you currently get around these types of planning challenges? Are there any features in iMessage/Whatsapp that’s helped, or any other products worth noting?

Hit reply and let me know!