simmisBetter AI agents are only the beginning. We write about organizations that people and agents can operate, program, and improve from within, turning useful work into lasting organizational capability without losing memory or control.
-
Measuring agent workflows on public benchmarks
Which model, which prompt, which steps: how Dvergr measures agent workflows on the same cases with intervals, cost and time, what it found on SpreadsheetBench and BIRD, and how a workflow can be improved by measuring its own variants.
read article → -
Superficie: Clojure you can read without learning Lisp
Why every Clojure example on this site has a second, block-and-infix view, how that view stays exactly the same program, and how libraries teach it to show numerical kernels and proofs.
read article → -
Stuttgart in a posterior: a city simulation you can question
A research demonstrator: 936,597 simulated people, shops chosen by size and distance, nine parameters inferred from the city's published retail turnover, and a policy question answered with its uncertainty attached.
read article → -
A KV cache you can fork
How pretrained-rstr gives a language model's KV cache an identity, a durable home, and a copy-on-write fork, so exact context can be reused across restarts, workers, and parallel attempts.
read article → -
A REPL you can fork
Fork a live SCI interpreter in the browser, compare independent runtime worlds, and see where branching stops at continuations, host objects, and external effects.
read article → -
Agents can act. Who keeps the company coherent?
Where agent harnesses, worktrees, and human-in-the-loop controls stop, and why multi-agent work needs shared organizational state.
read article → -
Shared context for AI agents: five architectures compared
How Wato, PromptQL, Dust, Buzz, and Simmis use memory, connected data, workspaces, event logs, or versioned organizational state to coordinate AI agents.
read article → -
What Bayesian inference does, in pictures
A research foundations guide to how probabilistic models turn evidence into distributions and how governed organizational history could support future learning.
read article → -
Copy-on-write branching: why isolated attempts can stay cheap
How structural sharing records only what changed and why that matters when people and agents work in parallel.
read article → -
Reading a causal graph: what your model knows
A research foundations guide to how causal graphs represent interventions and how governed organizational history could support future prediction.
read article →