Just over ten months ago, I co-founded a startup called Solace.
Our product, Ananta, was designed to help people capture their memories, stories and voice, and leave behind a digital presence for their family and future generations.
The idea grew out of my Master’s capstone at UNSW and originally went much further. I was interested in whether emerging technology could help us rethink not only how we preserve our ancestral footprint, but eventually how we support people experiencing grief, memory loss and some of the most difficult moments in life.
Last month, we made the decision to close the company and archive the code.
I don’t see the last ten months as a failure. We built something real, explored technologies that are changing almost weekly, challenged many of our assumptions and learnt a lot.
Most importantly for me, I learnt that some of the hardest problems in AI aren’t actually technical. They are human.
1. Just because we can build something doesn’t mean we should
Our original vision included grief support.
The possibilities were incredibly powerful, but so were the risks. We didn’t have a psychologist or grief specialist embedded in the team, and I wasn’t comfortable moving further into that space without the right clinical expertise.
So we changed direction.
It meant stepping away from some of the features I was personally most passionate about, but it was the right decision.
AI makes it increasingly easy to build things that were unimaginable only a few years ago. That doesn’t automatically mean those things should be built, or that the people capable of building them are qualified to decide how they should be used.
Sometimes human-centred AI means having the discipline to say, “not yet” or “not without the right expertise.”
2. AI speeds up building. It doesn’t necessarily speed up thinking
This was probably my biggest learning.
The technology was moving incredibly quickly. Things we couldn’t do one month became possible the next, sometimes forcing us to rethink something we had only recently created.
It created an interesting problem.
Code could move faster than design.
Ideas could become features remarkably quickly, which sounds fantastic until the definition of your MVP starts expanding just as quickly.
Our MVP became a moving target. We kept seeing new possibilities and, before we knew it, we were spending too much time building and not enough time putting what we had in front of people.
As someone who believes strongly in research, prototyping and testing, that was uncomfortable.
I also experimented extensively with AI design tools. On their own, I found the outputs fairly generic. They became much more useful once I had done the design thinking and completed the Figma designs. I could then use AI to turn that work into interactive prototypes very quickly.
But there was a point where one prompt too many could start unravelling earlier decisions. It reminded me a little of using Photoshop before layers back in the 90s. Save lots of versions!
The tools will get better, but I’ve come away believing that faster development should create more space for thinking, not less.
If AI saves us time building, perhaps the opportunity isn’t simply to build more. We can use that time to research more, explore more possibilities, test more assumptions and think more deeply about unintended consequences.
3. Representation in AI isn’t a cosmetic problem
One of Ananta’s more ambitious ideas was creating a speaking digital representation of yourself for future generations.
This exposed limitations I hadn’t fully appreciated until we started working with the technology.
Some likeness tools couldn’t capture people wearing glasses. Hair, particularly women’s hair, could be reinterpreted by the model. A photograph could suddenly acquire a ponytail that wasn’t there. From a single image, the technology might invent teeth, mouth movements and facial mannerisms that didn’t really belong to that person.
Individually these might sound like small imperfections.
But imagine the purpose of the product is to represent you.
I wear glasses. They’re part of how people recognise me. If I’m creating something intended to represent me to my grandchildren one day, I don’t want the technology deciding which parts of me are inconvenient to reproduce.
That changed the question from “does this look impressive?” to “does this represent a person with dignity?”
As AI becomes embedded in more human experiences, I think that distinction is going to become increasingly important.
4. Technical feasibility isn’t the same as product viability
Another important lesson was to model the economics alongside the experience much earlier.
We had around 30 people interested in participating in our pilot. By that stage we had moved well beyond a simple prototype and had built a substantial working product.
But running the experience relied on multiple AI services, and those costs added up quickly.
More importantly, I didn’t want a pilot where people spent a month capturing their stories and memories and then we simply switched their accounts off because the pilot had finished.
That raised another very human question: if somebody entrusts you with something emotionally meaningful, what responsibility do you have for its continuity?
The economics weren’t there for us yet.
AI costs will change. The technology will mature. But technical feasibility, desirability and commercial viability have to evolve together.
5. The human system matters as much as the AI system
Building a startup alongside a full-time career requires a ridiculous amount of energy.
For a while I loved it. I could work all day, come home and work on Solace, spend the weekend designing, and somehow feel more energised rather than less.
That’s the strange thing about working on something you really believe in.
But startups are also human systems. People bring different skills, working styles, pressures, expectations and definitions of success. Those things evolve as the product evolves.
One of my biggest lessons from this experience is that alignment isn’t something a founding team establishes once at the beginning. It needs to be continually revisited.
What are we trying to create?
Who are we creating it for?
What aren’t we prepared to compromise?
And are we still trying to solve the same problem?
Those questions matter just as much as architecture, models or features.
6. Knowing when to stop is also part of innovation
This wasn’t my first experience of closing something I’d invested heavily in.
Years ago I ran my own design studio. I kept it going longer than I should have because I had invested so much into building it.
I wasn’t going to make that mistake again.
As Solace evolved, I found myself becoming less connected to the direction the product needed to take to become commercially viable. At the same time, some of the technology I felt was essential to the experience wasn’t yet at the quality or cost point I was prepared to accept.
We could have kept going.
We could have invested more money.
We could have compromised.
Instead, we stopped.
That doesn’t mean the idea was wrong. Sometimes the timing, technology, economics and people simply haven’t converged yet.
The code is archived. Perhaps one day the technology will catch up with the idea, or the right clinical partnership will emerge and Solace will wake up again.
For now, I’m comfortable leaving it where it is.
What I’m taking forward
The last ten months have changed how I think about AI.
AI teams are understandably full of people thinking about what the technology can do.
I think we need just as many people asking:
Who is this actually for?
How will it make them feel?
What behaviours might it create?
Who might it exclude or misrepresent?
What assumptions are we making?
What happens when it gets something wrong?
What shouldn’t we build?
And what guardrails need to exist before we put it into someone’s hands?
None of those questions are particularly new. Good designers have been asking versions of them since digital design began.
What has changed is the speed and apparent intelligence of the technology.
AI can generate something remarkably convincing in seconds, and that can create an illusion that the thinking is finished.
I think the opposite is true.
AI gives us an extraordinary opportunity to explore more scenarios, challenge our assumptions, simulate possibilities and uncover unintended consequences we might otherwise have missed. But it still requires a thoughtful human to ask the questions, interrogate the answers and decide what matters.
Perhaps the biggest opportunity AI gives us isn’t the ability to think less.
It’s the time and capability to think more.
That’s the lesson from Solace I’m taking with me.
