Skip to content

#ai-agents

10 posts on this topic, newest first. all tags →

.2026

10 posts, newest first.

TypeSafe's Jev in production: thresholds and two bugs

Jev is a decision model: it picks, rates and judges, and writes nothing. I wired it into ten places on this site and measured what it cost. The model was the easy part. The hard part was choosing a threshold for each decision, giving moderation a token budget nothing else can spend, and finding out the counter meant to cap the whole thing had never counted anything. These are the eleven thresholds, the two bugs, and the rule I kept.

llmai-agentsarchitecture+2 more
19 min read24 min listen

Letting an agent touch production

Over five months we gave an agent access to production on-call in stages: read only, then propose, then act on a short list, then act on a longer one. Each stage had to prove a specific thing before the next. The rules that came out of it fit on a page, and they aren't the ones I'd have guessed at the start.

sreai-agentson-call+1 more
8 min read10 min listen

The code nobody read

A client asked me to take over a product two people and an agent had built in four months. It worked. It was sixty thousand lines, no human had read most of them, and the first bug took a week to find because there was nobody to ask. This is what inheriting vibe coded software is like, and what I'd have done differently in those four months.

ai-agentsengineering-practicemaintenance+1 more
9 min read12 min listen

Hiring engineers when everyone has an agent

Our take-home test stopped working in the autumn. Every submission was clean, tested and about the same as the others, because every candidate had used the same tools and the test was measuring the tools. We rebuilt the process around what still tells candidates apart, what they do with an agent's output. This is the loop, the exercise and what it found.

hiringengineering-practiceai-agents+1 more
10 min read13 min listen

Context engineering is cache management

A million token context window didn't make the long-running agent problem go away, it just moved it. The model still forgets what you told it an hour ago, only now it forgets after paying for it forty times. The techniques that work, compaction, retrieval, memory files, are the same ones we use to manage a cache.

llmai-agentsarchitecture+1 more
8 min read10 min listen

How we review pull requests that an agent wrote

By spring an agent was opening more than half of the pull requests on our main service. Our review process was built for people who understood their own diff, and it broke in specific ways. 'Read more carefully' didn't fix it. Changing what a PR has to contain before a person looks at it did.

code-reviewai-agentsengineering-practice+1 more
9 min read11 min listen

Who is the agent acting as?

Every agent I audited this year had the same kind of problem. It authenticated as a service account with more access than any human on the team, on behalf of a user whose identity was lost the moment the request left the browser, so the logs couldn't say who a call was for. Fixing it is mostly an OAuth problem, and the pieces exist now.

securityai-agentsoauth+1 more
8 min read10 min listen

Tracing an agent like a distributed system

An agent that runs for forty turns, calls twelve tools and talks to three models is a distributed system with one process. We spent two years learning to debug those with traces. OpenTelemetry's GenAI semantic conventions let you do the same here, and the first trace I looked at explained a bug that logs had hidden for a month.

observabilityai-agentsopentelemetry+1 more
8 min read9 min listen

Skills are not tools

Every team that adopted MCP hit the same wall by spring: forty tools in the context, a model that picks the wrong one, and a bill that grew with every server added. Skills, the folder-of-markdown pattern that spread through coding agents this year, fix a different problem from the one MCP fixes. Knowing which is which decides whether your agent gets better or just bigger.

ai-agentsmcpllm+1 more
9 min read11 min listen

Prompt injection is a data plane problem

One GitHub issue was enough to make a coding agent leak a private repo through the official MCP server. In January the Git MCP server fell to path traversal from a prompt alone. I don't think either was a model failure. Both were the mistake we made with SQL in 2004, sending data and instructions down the same wire.

securityai-agentsmcp+1 more
9 min read11 min listen