Agents as Persistent Systems of Record
Filip Kozera, CEO of Wordware, on the rise of vibe doing
The real battleground in AI agents is shifting from doing a task once to carrying forward a living model of a user’s work. Manus proved there is strong demand for an agent that can browse, fill forms, run code, and produce outputs fast, but knowledge work creates value in the memory layer itself, where emails, meetings, files, and preferences have to stay organized and usable across weeks, not just a single run.
-
In coding, Git is the system of record. The agent opens a job, changes code, gets approved, and the session can be discarded. In knowledge work, there is no equivalent main branch. The useful artifact is the evolving workspace, which means the agent needs persistent state, not just task execution.
-
Wordware’s product direction is built around that persistence problem. Sauna keeps a file system, identity files, semantic memory, and always on background monitoring across connected tools, then surfaces work in a delegation board with to do, working, and review states. That is much closer to a digital chief of staff than a one shot browser agent.
-
Manus’s rapid growth shows how strong the market is for all in one agents, reaching an estimated $90M annualized revenue by August 2025 after launching in March 2025. That traction matters because it validates the category, but it also shows why the next layer of competition will be about retention through memory and workflow continuity, not novelty alone.
From here, the winners in knowledge work will look less like disposable task runners and more like systems of record with agency. The product that best remembers priorities, relationships, approvals, and unfinished work across tools will have the strongest lock in, because each completed task makes the next task easier and more personalized.