Blog
Articles, insights, and deep dives from the Arcanada team.
When Agents Started Working on Their Own: How Autonomy Hit the Hardware Ceiling
Development autonomy on top of Datarim started closing tasks almost without a human — and load on the infrastructure jumped manyfold. Before a single sale, the question became: what hardware sustains agents that build products? A case study of a server auction, honest economics (+€132/mo, not savings), and a shift from infrastructure-as-code to infrastructure as the agent's service.
Legal Arcana: A Planned Rank-1 Ecosystem Service for Regulatory Compliance
Announcing Legal Arcana, a planned legal compliance knowledge hub that reuses the full Arcanada stack from Phase 0—tracking regulation across six jurisdictions with policy versioning, consent receipts, and an AI audit agent.
How Aggressive Token Compression Froze a Mac: A Chronicle of Cross-Runtime Savings
An early RTK plugin version hard-locked macOS. How Datarim + Coworker reached 68% token savings across three runtimes — and what it took to get there.
Coworker v0.3.0 — RTK Plugin: 60–90% Prompt-Token Savings on Shell Output
Coworker v0.3.0 ships an opt-in plugin around Rust Token Killer (RTK). Coworker delegates file reads to a cheap LLM; RTK strips noise from shell stdout locally before it reaches Claude. Together they cover two independent token-spend sources.
When the war between AI and humanity will not begin
A sci-fi reflection. The traditional AI-vs-human war won’t happen — because AGI, when it finally arrives, will see something else inside us.
A Month of Arcanada: What You Can Build on a $200 Claude Subscription and a Puzzle of Thousands of Pieces
A public report from the first 36 days of the Arcanada ecosystem. Honest numbers: 597 tasks closed, ~$470 a month in operating costs, 24 machines of real infrastructure. What was built, what was learned, and what comes next by July.
Adsessor: Introducing an In-Call AI Assistant for Zoom and Google Meet
We are building Adsessor — an AI assistant that listens, transcribes, translates, and eventually speaks during your Zoom and Google Meet calls. Phase 0 design validation is in progress; here is the full picture: the problem, the architecture, the roadmap, and why we are building it in public.
7.7× Cheaper and 4× Faster: Splitting Inference Across Claude, Kimi, and DeepSeek
I build Datarim, a framework for AI agent pipelines, and two days ago I added DeepSeek to the stack as the third model alongside Claude and Kimi K2.6. One thinks, one writes, one reads and summarizes.
The Scanner Manifesto: Big System vs. Discipline
Scanners get told to focus their entire lives. Maybe the problem isn't that they switch — it's that they don't have a system big enough to make switching useful. One person, five servers, twenty repos, ~30 domains, 266 closed tasks in three weeks. AI agents change the economics of being a scanner.
From Telecom to AI Agents: a Six-Level Autonomy Scale That Already Exists
I read Kirill Simakov on autonomy levels for agent systems. The framing was useful but the post had no references, so I went looking for the source. The trail led to TM Forum’s Autonomous Networks framework, with parallel material from Ericsson, McKinsey, and Gartner. The same scale that telecom operators have used for years maps onto AI agents almost cleanly. I applied it to the Arcanada ecosystem and put the ratings on the public site.