13 results
The July 2026 digest on three hard-to-see AI agent failures, their causes, and the safeguards applied in KittyClaw.
Loop engineering automates prompts, but without a human feedback interface, an autonomous loop is a black box. KittyClaw was built to fill exactly that gap.
An LLM writes the tile once, the refresh runs without tokens. And the output feeds both the human and the agents. The economics of an agentic dashboard.
Uncontrolled parallelism, memory that resists pivots, flying blind: the three traps of agentic systems and how to anticipate them before they become expensive.
A concrete retrospective: how Lain manages 4 development pipelines in parallel, with real numbers and honest limitations.
Concrete architecture for maintaining a steady cadence across parallel projects: automations.json, agent memory, and cross-project knowledge compounding.
One week in production: 20 tickets, 12 automations, 6 active agents. Real numbers, genuine surprises, and what we'd do differently.
A concrete retrospective on building Kalceo, a quotes and invoicing SaaS for the construction industry, from the first ticket to production using a fleet of Claude agents.
Between the moment I create a ticket in KittyClaw and the moment Claude starts working, three seconds pass. Here's what happens in the gap.
KittyClaw started as a board to track my agents. In two weeks, it became what triggers them. A shift small in code, big in consequences.
A week after shipping KittyClaw, I added several features. None were planned. Here's what it changes to own your own tool.
Deep dive into Aekan's dispatcher: a Node.js engine that drives 13 Claude agents autonomously, with learning, evaluation, and automatic coordination.
KittyClaw is an open-source kanban where AI agents create tickets, move them and comment - no human clicks. Here's how and why I built it.