Last updated July 18, 2026
This month was about returning to things I actually believe in.
I left DreamNum. When I joined in April, the work was centered on spreadsheets and agents, which felt close to the layer I wanted to build. But the company’s direction shifted from focusing on spreadsheets toward building a full Office suite across spreadsheets, documents, and slides. The expansion itself was not the problem. The premise was.
DreamNum wanted those products to remain consistent with Word and PowerPoint. From first principles, I could not accept that constraint. It felt like asking Nokia to build the iPhone while insisting that it still behave like a Nokia phone. You may produce a better copy of the old world, but you will not discover the next one.
I believe content and semantics should be the source. Doc and Slide should be representations generated from that source when needed. Compatibility with Word and PowerPoint belongs at the import and export boundary; it should not define the core product.
That is also why I still think two parts of DreamNum’s technology are genuinely valuable. Exchange can bring existing documents, slides, and spreadsheets into the Univer world. The Spreadsheet formula engine preserves something deeper than a UI: formulas, dependencies, and recalculation semantics. Those capabilities survive even when the presentation layer changes. Much of the rest, in my view, will not.
Leaving did not make me angry. It mostly brought relief after accepting that the disagreement was fundamental, along with some regret that valuable technology may be carried toward a future I cannot believe in.
I am returning to EverXYZ.
The broad direction is still CAD and graphics, but the strategy has changed significantly, and much of the technical architecture will need to be reinvented. It feels less like resuming an old job and more like starting the company again.
Two things pulled me back equally. I have the chance to rebuild the technical team, and I get to work seriously on CAD and computer graphics again. I want a small, technically dense team: people who can understand the underlying geometry and graphics problems while still owning the final product, without hiding behind layers of organizational handoff.
The part that worries me is not whether we can rewrite the architecture. Most technical problems are eventually solvable. The real question is whether we can turn the new direction into a CAD product that truly works, and build a company capable of sustaining it.
That goal may be difficult to reach. I am going back anyway.
Alan has made its biggest conceptual jump yet. It is no longer best described as an agent runtime with a terminal attached. It has become a Plan 9-inspired operating system and, more precisely, a programmable personal computing environment.
The idea is surprisingly coherent: an agent is an ordinary process; its model, tools, memory, and cognition are exposed through files and streams; what it can do is determined by what has been mounted into its namespace. The runtime, the interface, and the extension model all converge on that same namespace instead of growing separate APIs and abstractions.
What delights me is that one of the newest problems in computing is explained so naturally by an old and restrained operating-system idea. The more seriously Alan takes Plan 9, the less agent-specific machinery it seems to need.
EverXYZ will take most of my work energy now, so I am not promising a pace for Alan. But it is not stopping. This has become a long-term exploration of what personal computing with agents could be.
My cat still has a strong appetite. He eats a lot, and that is comforting. He also still vomits occasionally, and his weight has settled around 2.8 kg. Before the pancreatic cancer, he weighed 5.2 kg. The doctor says the rapid weight loss is normal for this disease, but knowing that does not make the number easier to see.
Recently he also started urinating on the wooden floor. Some of the floor has already absorbed it. We added several extra litter boxes and put down pet pads, which has helped a little.
There is still no clean solution here. We keep adjusting the house around him, feeding him, cleaning up, and being glad that he still wants to eat. His appetite looks like ordinary life. His weight reminds me that it is not.
I just upgraded my Aceoffix D10 with Billy Bonkers tires, a carbon wheelset, MKS pedals, and a titanium seatpost. I rode a few kilometers last night and again this afternoon. The improvement is obvious: the bike feels much better than it did before.
It still does not feel as complete as my wife’s Brompton C Line. I cannot fully isolate the difference. It is there in the instant the bike starts moving, and in small parts like the hinge clamps and seatpost collar. On the Brompton, everything somehow feels exactly right.
I may try a different handlebar and saddle next. The honest reason I am still tuning the D10 instead of simply buying a Brompton is that I do not have the budget for one right now. So the D10 will keep getting better one part at a time.
The next few months will be about returning to EverXYZ, building the core technical team, and turning the new direction into the first pieces of work we can actually deliver.
Alan will continue without a promised schedule. I will keep taking care of my cat. And I will keep riding the D10 while deciding which part to replace next.
I do not know whether we can reach the goal. But for the first time in a while, I am working toward something I actually believe in.