Product
Your Product Has an Ecosystem Problem. My Aquarium Showed Me Where.
A fish tank tells you something is wrong long before the fish do. Most products have no equivalent, and that is where they quietly fail.
system design lessonsproduct design ecosystemaquarium system design
Overfeed a fish tank and the fish look fine for a while. The water has already told you otherwise, if you own a test kit and bother to use it.
I keep planted freshwater aquariums. They are my break from a day of code, interfaces and user experience work. Somewhere along the way I noticed that the habits a tank forces on you are the same habits most product teams skip. Watching a tank is cheaper than watching a dashboard, and it teaches the same thing faster: systems rarely fail because of one big mistake. They fail because a small imbalance went unwatched.
Why does a healthy-looking tank still fail?
Because looks are the last signal to change. Water parameters tell you whether the system is healthy long before the fish show any distress. By the time a fish looks unwell, the problem is already old.
Software is the same. Usage patterns, response times and error rates are your water test kit. If you only react when users complain, you are reading the fish. I wrote a bit about a version of this in my rate limiter asked the wrong question, where the number I was watching was not the number that mattered.
If I had to pick one habit to copy from the tank, it is this: set a baseline while things are calm, so that you can tell what “off” looks like later. You cannot spot an anomaly in a system you never measured when it was healthy.
What does feeding a tank teach you about inputs and outputs?
A tank runs on a cycle. Nutrients go in, waste comes out, and the filter sits between the two. Too much food with too little filtration ends in disaster.
Products have the same shape. Inputs are user requests and data volumes. Outputs are whatever your system does with them. When inputs grow and processing does not, you get a bottleneck, and a bottleneck can take the whole operation down. It is easy to celebrate the input side (signups, traffic, uploads) and forget that someone has to clean up after all of it.
Here is the question I now ask when a growth number looks good: what is the filter for this? If nobody can name it, I get nervous.
Does the foundation really matter that much?
In an aquascape it starts with the substrate. It is a hidden layer, but it feeds the plant roots and does heavy filtration work. Nobody admires the substrate. Everybody depends on it.
Early tech choices behave the same way. Rushed decisions in the first weeks can leave you with a system that collapses under the weight of its own success. Not every product needs a perfect architecture on day one. My guess is that the real damage comes from foundations nobody chose on purpose.
Why do you need room to grow?
Experienced aquascapers plant with the future in mind. Fast-growing plants need space, or they overwhelm the tank. A layout that looks full on day one is a layout with a problem in three months.
For products, leave space for features and users you do not have yet. Designs that are too rigid turn into constraints. Some flexibility built in early is what lets the next thing fit without a rebuild. A simple test: can the system handle ten times today’s load, and what would have to change? If you cannot answer, that is already useful information.
What is the calm surface hiding?
A serene tank is a lot of equipment: filters, heaters, pumps, calibrated lighting, and a maintenance routine. The calm is the result of work you do not see.
Good product design hides complexity the same way. A smooth interface comes from careful engineering, testing and constant refinement. It never comes from a simple solution to a hard problem. The risk is that the calm surface convinces people the machinery is optional. Automated monitoring and resilient systems are the maintenance crew, and they are the first thing cut when a roadmap gets tight.
Which parts of the tank depend on each other?
Every organism has its own needs: water parameters, compatible tankmates, a balanced diet. You cannot treat one in isolation. Add the wrong tankmate and the whole system changes.
Products live inside larger systems too. They connect to other tools, fit into user workflows, and shift when the market or user preferences shift. When I review a feature now, I try to list what it touches before I list what it does. This is the part where the title of this post comes from. Most “product problems” I have seen were really relationship problems between parts.
How do the tank and the product compare?
| In the tank | In the product | What I would check |
|---|---|---|
| Water test kit | Metrics and alerts | Do we have a baseline from a calm week? |
| Food in, waste out | Requests in, processing out | What is the filter for this growth? |
| Substrate | Architecture and data foundations | Which early choices were never really chosen? |
| Plant growth | Feature and user growth | Can it take 10x? What breaks first? |
| Pumps and filters | Monitoring, automation, deploy pipelines | Who maintains the invisible parts? |
| Tankmates | Integrations and workflows | What does this change touch? |
What would I do differently on Monday?
- Pick the three numbers that would move first if the system were drifting, and write down what normal looks like for each.
- For every growth lever, name the thing that absorbs the extra load.
- Review one early technical decision that nobody remembers making.
- Ask what the system would need at ten times the load, even if you never get there.
- Spend ten minutes just watching real usage. I do not have data on how much this helps, but in my tanks, noticing a new leaf or a shrimp behaving oddly is usually how I catch things early.
Patience matters here too. A good tank takes water changes, chemistry adjustments and a few unexpected hiccups. Products need the same slow loop of feedback and adjustment. Nothing about it is quick, and I have not found a shortcut.