Production Agents 2026
10 min read6 primary sources
MCP connects agents to tools. A2A coordinates independent agents, including through stateful tasks. Learn where each belongs and what neither protocol solves.
Articles & Journal
Writing
I think in public. Long-form articles work through the engineering ideas in depth; the journal is a dated log of what I'm building and learning along the way: retrieval, multi-agent systems, evaluation, and applied machine learning.
More topics (20)
15 results
Articles
Production Agents 2026
13 min read8 primary sources
Build agent memory as governed state: control writes, preserve provenance, supersede stale beliefs, enforce deletion, and verify actions against evidence.
Production Agents 2026
12 min read5 primary sources
Treat prompt injection as a source-to-sink security problem. Contain the agent, limit authority, and gate consequential actions outside the model.
Production Agents 2026
11 min read5 primary sources
Evaluate the whole agent system: verify final state, inspect traces, control the environment, repeat trials, and turn real failures into regression tests.
Production Agents 2026
11 min read4 primary sources
Task horizon, uninterrupted runtime, and parallel agent-hours are different clocks. Learn how to read agent autonomy research without overstating it.
7 min read
AI can build a component from a screenshot. It still can't tell you whether it matched. The fix is the same one that applies to every agentic loop: separate the measurement from the judgment.
5 min read
Anthropic's new Claude Tag turns @Claude into a persistent, shared teammate inside Slack. Here's what it actually does, how it's governed, and why the form factor matters more than the feature list.
8 min read
The unit of agentic work has moved from the prompt to the loop. A working definition, an anatomy of what a loop is made of, and the one principle that separates loops you can trust from agents that agree with themselves.
9 min read
/loop runs a prompt or a slash command on a cadence inside your open Claude Code session, so you stop re-asking 'did it finish yet?' and let the machine do the polling. This is what it is, the three ways to run it, where it earns its keep, and the session-scoped limits that tell you when to reach for durable automation instead.
12 min read
Running a coding agent unattended is tempting and mostly a trap. The thing that makes it safe isn't a better prompt. It's an external, deterministic gate the model cannot talk its way past. Here is the principle, a working pipeline that embodies it, and the failure modes that matter.
4 min read
Fetching the right documents is necessary but not sufficient. Grounding (answers that are actually entailed by the retrieved evidence) is a separate property you have to design for and measure. Here is how I think about the gap, and the evaluation that closes it.
4 min read
When an agent system misbehaves, the instinct is to add more instructions. Usually the real fix is structural: explicit states, hard guards, and small tools with narrow contracts. Reliability is an architecture decision, not a prompting one.
Journal
- exploring
Notes on chunking: smaller isn't always better
Spent the afternoon re-running a chunking ablation because a retrieval metric drifted and I wanted to know why. The folklore is "smaller chunks → better recall." That is true right…
- note
An eval harness before the feature
Rule I keep relearning: build the evaluation before the feature, not after. It feels slower. You sit down to add a capability and instead spend the first hour assembling twenty lab…
- reading
Re-reading BM25 in the age of embeddings
Went back to the BM25 literature this week, partly out of nostalgia from my IR research days, partly because a hybrid retriever I'm tuning keeps reminding me how good the old lexic…