Planning for 2026. Three priorities. Everything else is off the list, or moved to later. This post is our record of the decisions, written publicly so we can be held accountable to them.
Priority one: double down on Alumia. Alumia shipped in mid-2025. Early reception was better than we expected, and we now have a concrete, real-world product with real customers and a real revenue line. The single highest-leverage thing we can do in 2026 is make that product dramatically better for the people already using it. Not broaden it. Not add surface area. Make the current surface area much more reliable, more observable, and more delightful to use. This is the unglamorous priority on the list. It will not generate exciting launch announcements. It is the one most likely to produce real value by the end of the year.
Practical initiatives inside this priority: a substantial investment in observability and debugging tools for agent runs, which is the number-one thing our design partners have asked for; a push to reduce cold-start latency, which directly affects perceived quality; a better system for handling agent failures gracefully so that when an agent gets stuck, the user experience is one of easy recovery rather than confused silence; and a tighter integration with our open-source pieces so that advanced users can customize more of the platform without leaving it.
Priority two: keep investing in open-source infrastructure. Our libraries are used by thousands of developers now. We have barely scratched the surface of what they could be. The current generation is functional but verbose; the APIs are more complex than they need to be; documentation, while present, could be better. We are allocating meaningful engineering time in 2026 to polish — cleaning up the APIs, improving docs, cutting the number of concepts a new user has to learn to get productive, and filling in the obvious gaps (better error messages, more consistent conventions, more examples).
There are also three specific new packages we want to publish this year, each tied to problems we are already solving internally. One is a simpler memory primitive that trades some generality for much better defaults. One is an evals library for grading agent runs in a reproducible way. One is a small, focused project for agent-to-agent communication, for cases where you have multiple specialized agents working on the same problem. All three will be shipped under MIT, as always.
Priority three: deepen the first university partnership before starting new ones. The UTCN competition runs for the first time this year. Whatever we learn from running that event — about logistics, about the quality of what students build, about what parts of the format work and what do not — we want to apply before we try to do anything bigger. The easy mistake to make is to run one event well and immediately try to scale to ten universities. We have seen other people do exactly that and regret it. We are going to run this one event, run it twice, and only then think about what comes next.
Within this priority: we are funding one small research project at UTCN this year, not three. We are not trying to run events at multiple schools. We are not building a portal or an app or any institutional machinery for this. One competition, one funded project, done well, iterated on the basis of what worked.
What is explicitly not on the list.
Emebo and Pawk stay in research mode. We announced both in 2025, and we are happy with the pace they are moving. Neither is close to a product. Neither is a priority for the business in 2026. Both continue with small dedicated research teams. We will post substantive updates when there are substantive updates to post, and not before.
No new product launches. Alumia is the product. No second product this year.
No expansion into new geographies or go-to-market motions. We do not have a sales team; we are not hiring one. Our growth comes through open source and word of mouth, and we think that is the right engine for our stage. Trying to layer on a traditional enterprise motion this year would fracture the team and distract from priority one.
No new university partnerships. See priority three.
No podcast. We have been asked a few times. It is a different kind of work and we are not going to do it well while also doing everything on the list. Maybe next year.
No media tour. Same reasoning.
The other thing worth noting. Our team is still small. The three priorities are big enough to absorb the entire team, if we are honest. Anything we add on top of them cuts into the priorities themselves, because the work has to come from somewhere. Saying no to things is the harder discipline than saying yes, and we are trying to get better at it.
We will write a quarterly check-in on these priorities. What is going well. What is not. What we have had to adjust. Real numbers where we have them. This is how we want to operate — public about our goals and public about our progress, including when we are behind schedule or have to revise a commitment.
That is the plan for 2026. It is deliberately narrow. If it works, we get to write a very satisfying year-end post. If it does not, we get to write an honest one. Either is better than where many companies end up, which is with a plan too broad to execute and a year too cluttered to describe in a single page.