Thoughts
on building software, reasoning about systems, and the occasional philosophical detour. Written to think out loud. RSS Feed
Featured
-
New Domain: bjro.dev
The blog has a proper home now — bjro.dev.
-
Linear Types for Agent Safety - An Approach to Trustworthy AI Systems
Discussing a type system that rejects unsafe code before it ever runs.
-
MarkyMarkov - Markov Chain-Based Code Guidance for LLM Agents and Humans Alike
MarkyMarkov - a PoC leveraging Markov chains for coding - shows that small, interpretable probabilistic models add high value to code workflows. By learning patterns from your own codebase, Markymarkov provides fast, explainable guidance that complements existing tools rather than replacing them.
-
Building Smarter AI Agents With Ideas From Philosophy
Updated:Philosophically informed agent design means explicitly modeling what an AI believes, how it knows things, and which moral and epistemic norms guide its decisions, so that its behavior becomes more reliable, inspectable, and aligned with human expectations.
-
Context, context, context
Context—or how I learned to love the MINDMAP
-
Cloud LLMs in prod...
Modern infra teams are quietly feeding their most sensitive data—logs, configs, schemas—into cloud LLMs just to get unstuck. It feels like harmless troubleshooting, but in reality it extends your trust boundary into a black box you don’t control or fully understand.
-
A couple of notes on some software development topics.
Thoughts on some software development topics with regards to modern agentic tooling
-
Using casq in CI
Content addressable storage for improved CI
-
Introducing casq (a simple content-addressable file storage CLI and library)
casq is a small, no-frills content-addressed file store and Rust library that gives you git-style hashing and deduplication locally, without trying to be a full backup system or version control tool, and is available under the permissive Apache-2.0 license.
-
Terminal multiplexers
Notes on terminal multiplexers.
-
Locking down agents
Discussions about how one can limit the blast radius of agents.
-
Making use of the Caps Lock key
Remap Caps Lock to Left Control.
Recent Posts
-
Introducing cairn: append-only event storage with SQLite
cairn is an append-only event store backed by SQLite where immutability is enforced by triggers, not convention. SDKs for Go, TypeScript, and Rust share one spec and 21 test vectors.
-
The npm problem nobody wants to work on
Every production Node.js system depends on a single corporate-controlled registry with no meaningful public alternative -- and the ecosystem treats this structural risk as a solved problem because it hasn't broken yet.
-
Debian
Debian is a marvel.
-
Currying agents
How one can leverage partial application of agents - borrowing from the functional programming paradigm.
-
Markov chains and LLMs - hybrid architectures for smarter agents
How to combine Markov chains with large language models to build more structured, interpretable, and efficient agents. Markov chains provide structure and efficiency; LLMs provide semantic understanding and flexibility.
-
Notes on agentic applications in business processes
Updated:This post explores how to pair agentic workflows with durable workflow engines (specifically Temporal), add structure using finite state machines, plug in MCP servers for tools, and grow from a small pilot into stable, production-ready workflows.