Home / Articles / The wall under the reply / Versions / v1.2.0
The wall under the reply: what changed in v1.2.0
From v1.1.0 (2026-10-04, b5a363102) to v1.2.0 (2026-10-04, 025e22b73), paragraph by paragraph.
0 paragraphs added, 0 removed, 2 changed in place, 62 unchanged. About 10 words added and 3 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
← v1.1.0 · all versions · v1.3.0 →
3 unchanged paragraphs
> Where this comes from, and what is behind it. A voice memo on 4 October 2026, about something I have already started asking the agents to do. It follows directly from three published pieces: Custom UIs are not the exception, which already said "send the summary, not the thread" and described the typed blocks the agents pass between themselves; Replicating the agentic inbox, which has the drafts role that would write this; and Memory is not a spectator sport, whose question, what does this agent need in its window right now, is the reader's question too. The history and the current tools were researched on 4 October 2026 at the URLs in the sources, and the gaps in that research are stated where they fall. Dinis also posted the article on LinkedIn the same day.
33 unchanged paragraphs, under In short, The wall, and what it costs, The proposal: end with the state, not the history…
I do not know what the tail should look like, and I distrust anyone who says they do, because the format is personal, the readers differ, and the clients differ. So rather than publish a format, I am publishing an experiment, and I am going to run it on my own correspondence,correspondence via the agent@riskmandate.ai agentic team, which the memo put bluntly: use all my interactions to experiment.
26 unchanged paragraphs, under What exists today, and what does not, The article, condensed by another model, Threads woven here…