Sand Saga Insider: Summer 2026

Patrik Harag, 2026-10-05

Hello everyone! The summer was a bit quieter than usual, with fewer updates and a couple of skipped releases. I took a more relaxed approach to development for a while, but became more active again in September. This quarter was mainly about Machinery 2.0 and more sound effects.

Additions

Elevator and Machinery 2.0
For some time, I had a list of things that I wanted to change about machinery, and the latest release was a good opportunity to finally tackle some of them.

I noticed that machinery in workshop creations often ended up looking rather messy. For example, players would try to connect belts or grinders into a row, but getting them properly aligned was difficult. It was clear that some kind of grid or alignment system was needed. In the end, I decided to use snapping to already existing machinery instead. This is a more complicated solution, but it gives players more freedom while still helping them build organized structures. The snapping is also intelligent and tries to connect compatible entities, which can provide useful guidance while building.

Pipes were another part of this work. I was not very happy with how their interaction with other elements looked, especially when the available space was already crowded. I eventually decided to move pipes visually into the background. The supply pipe also gained a flow setting.

The biggest addition, however, is the Elevator. It was a missing component that allows factories to be built vertically as well as horizontally. I spent quite a bit of time working on both its design and implementation, but I think it was worth it.

The elements inside an Elevator are fully simulated. Elevators can be extended to arbitrary lengths, and conveyor belts can be snapped to both their input and output. Launchers can also be attached to the output.

There are still some ideas for future machinery. A filter and piston are two of the more obvious candidates.

Sound Effects
I continued adding new sound effects throughout the quarter. The most noticeable additions are probably the new sounds related to burning, including thermite and the extinguisher, and bird sounds.

Sound design is still an ongoing process. There are many more things that could use their own sounds, and existing ones can still be improved, so this will likely remain a long-term part of development.

Sand Saga in the Age of Early Agentic Development

Until recently, I had not really used AI for developing Sand Saga, apart from occasionally using small pieces of code generated with ChatGPT or similar tools.

I wanted to share some of my experience with using coding agents and, perhaps, also dispel the idea that software is now somehow developing itself. The short version is: development can be faster, but making something excellent or unique still takes a lot of work.

In general, if you give an agent enough context and rules – such as AGENTS.md, specifications, skills, and knowledge of the project's internal architecture – then writing a precise specification in natural language can actually be faster than implementing the same thing directly in a programming language.

The resulting code can also look surprisingly similar to code you would have written yourself, especially when the project's rules and conventions are well defined. However, I still consider myself the owner of all APIs, including internal ones. I review generated code, and quite often I ask the agent to rewrite parts of it.

This is the low-level approach I currently use for more difficult domains. When the domain is simpler, you can go further. The specification does not need to be nearly as precise, and agents can sometimes implement entire features from a relatively high-level description. If maintainability is not particularly important, such as for a prototype or a small tool, there is also much less reason to carefully review every detail.

Sand Saga is a different case. It is a difficult domain, and the quality of the code and APIs matters. Current coding agents can help a lot with the implementation, but they do not really understand the game's physics, gameplay, and aesthetics. They cannot see how the game actually behaves, and therefore they often miss important details that are obvious when you understand the system as a whole.

In short, if you give an agent a general request to add something new to the game, the result will be an AI slop accompanied by tests that do not actually test much.

I had several things on my backlog that were relatively easy to specify, but very tedious and time-consuming to implement. I had been postponing them for quite a while. The agent does not mind tedious work, and this allowed me to clean up parts of the engine and also helped with some of the work involved in Machinery 2.0.

On the other hand, when I tried to design and implement the Elevator before I had really figured out how it should work, I got exactly the result I expected: unusable slop.

Conclusion

That is all for this quarter. September was considerably more productive, and Machinery 2.0 finally brought several long-planned changes together. The next release will probably focus more on performance optimizations.