Notes on agents, software, and work.
Systems / July 21, 2026 / 8 min read
SearXNG MCP servers: which to run, and how to keep it alive
By the end you know which searxng MCP server to wire into your agent, how to connect and test it, and what the 2026-07-28 stateless spec changes about the transport you pick.
Operator notes / July 21, 2026 / 8 min read
Binary Ninja MCP: which server to run and what it costs
By the end you know which of the three live Binary Ninja MCP servers to run for your client and workflow, what it actually costs to get there, how to wire it in, and which agent capabilities to keep reversible.
Operator notes / July 21, 2026 / 7 min read
Zotero MCP: which server to install, and the four shapes
By the end you can name which of the four Zotero-AI tool shapes fits your setup and privacy tolerance, and which specific server to try first, without reading five separate setup pages.
Operator notes / July 21, 2026 / 8 min read
Ableton MCP: which server to install and wire up first
By the end you know which Ableton MCP server to install and why, exactly how to wire it to your client, what it does and doesn't cover out of the box, and where setup breaks, plus the canonical repo link to satisfy the raw search.
Systems / July 21, 2026 / 8 min read
The agent loop, and where to put a human inside it
You'll leave with one reconciled loop model (assemble context, reason, act, loop), a decision table for where a human should gate the loop by reversibility, and a way to choose the framework by the job it does rather than its star count.
Operator notes / July 21, 2026 / 9 min read
Agentic AI example: real one or just wrapped automation?
You will see real, current agentic AI examples and leave with two things the listicles never give you: a test for whether a system is genuinely agentic, and a decision for when a workload actually needs more than one agent.
Systems / July 21, 2026 / 8 min read
Guardrails AI: where each LLM guardrail actually belongs
Leave with a decision you keep: which guardrail layer to use for what you're guarding, which of the three main frameworks fits, how to start, and where each one (and the category) falls short.
Operator notes / July 21, 2026 / 9 min read
BMAD vs Spec Kit: which harness fits your actual work
By the end, the reader has a two-line pick rule (when to choose Spec Kit, when to choose BMAD) and a harness-fit matrix they can apply to their own codebase, not just another feature scorecard.
Systems / July 21, 2026 / 8 min read
Non-deterministic vs deterministic: when it matters
One definition of non-deterministic vs deterministic that holds across CS theory, AI, and psychology, worked examples for each, the real reason 'deterministic' LLM settings still drift, and a decision framework for whether to enforce determ