Agentic AI
Treating tool-using language models as processes rather than conversations.
16 articles by Paul Scarrone.
- Agentic AI Engineering: Building Determinism Around a Non-Deterministic Engine A working definition of agentic AI engineering: treating tool-using language models as processes rather than conversations, and putting the determinism in the pipeline instead of the prompt.
- Judgement-Capable Circuits: What You Write Loops Out Of If the job is writing loops rather than prompts, the question is what you write them out of. A circuit is deterministic wiring with judgement localised to specific nodes, assembled from tmux, an event transport, and generated prompts, running on credentials you already hold.
- Orchestrating Agents on Amazon Bedrock: When You Cannot Push, Make Completion the Wake Multi-agent designs assume push delivery, which Bedrock-hosted sessions cannot receive. Making process completion the wake signal, and scoping agent credentials with session policies so blast radius is a property of the credential rather than the prompt.
- Five Techniques That Keep Paying Off When You Build for LLMs Six months of building tools for LLMs, distilled into five techniques: embedding and vector search, structural indexing, harness engineering, hooks, and why agents are better at writing code than using tools.
- Beamlet A minimum viable OTP over WebAssembly - supervised, capability-sandboxed processes on wazero, built as the substrate for a self-rewriting LLM harness
- Wavelet Google Wave's data model repurposed as a coordination substrate for multiple chat agents - forked conversations, cursor-based wake semantics, and deterministic targeted delivery that keeps an agent's context about what concerns it
- Cando Task-oriented AI agent - spawned for a task, does the work, exits
- Kwike Composable event-driven agent orchestration using Unix primitives
- kwike: LLM-First Agentic Workflow Composition An LLM-first tool for composing agentic workflows using Unix primitives - pipes, append-only logs, and event subscriptions instead of SDKs and harnesses.
- Catalyst: An Orchestrator That Stopped Asking and Started Deciding A devlog tracing catalyst-orchestrator through three architectural revisions — from LLM-driven gate decisions to deterministic routing to daemon-managed fix loops.
- Automatic Programming: Iteration 4 From looms to LLMs — Grace Hopper dreamed of programs writing themselves, and each generation of programmers has resisted the next layer of abstraction.
- Keep Your Eyes on the IDE, and Your Robots on the Tickets Building a Jetbrains plugin for Beads, an AI agent orchestration workflow using context graphs and deterministic prompts
- Agentic Patterns: Elements of Reusable Context-Oriented Determinism Exploiting tool-using LLMs as processes instead of conversations - patterns for keeping context focused and outputs deterministic
- A Deterministic Box for Non-Deterministic Engines Taming LLM non-determinism through structured workflows, task tracking with Beads, and memory management strategies for more predictable AI-assisted development
- Claude or Clod A year of vibe coding, agent orchestration, and learning to trust what predicts but doesn't think
- The AI Diaries Anecdotes and observations from working with AI coding tools - the hype, the reality, and the debt