Two mornings in a row, I had a reassuring Telegram notification from my AI assistant. The OpenClaw gateway had stopped responding, the watchdog had restarted it, and everything was healthy again. Nothing needed from you. On the second morning, that sentence deserved an investigation. The recovery mechanism was working. But a service that needed rescuing every night was not fixed. I asked Closp, my OpenClaw assistant, to compare the incidents, investigate the growing database, and find a solution that preserved its memory and usefulness. ...

I recently reviewed six improvements that my AI assistant had marked complete. Only two were actually complete. The other four were not imaginary or abandoned. Code existed. Tests passed. Prompts had been deployed. Scheduled jobs were enabled. Several parts had already worked in production. But each task still had at least one promised live outcome that had never been observed. One was waiting for its first scheduled weekly run. Another had delivered its news path but not the reading-feedback loop included in the same specification. A third had a longer observation period written into its own acceptance criteria. ...

Earlier this year I built an AI and technology news digest. It pulled from newsletters, Hacker News, and Reddit, filtered for AI terms, and sent the results to Telegram every few days. It was useful. It was also a list generator. The system knew that an article matched a keyword and that a link was popular. It did not know whether I had seen the same underlying story three times, whether a consequential claim had a primary source, whether the item connected to anything I was working on, or whether I had rejected the same kind of recommendation last week. ...

I have a Quick Notes topic in Telegram. It is the nearest thing my personal system has to an input box: send a sentence from my phone and let the machinery put it somewhere useful. The first version was efficient in the way a paper shredder is efficient. It could turn a message into a calendar entry or task, but it did not understand that a thought, a reference, a question, and a commitment are different things. ...

My Google Tasks lists used to have a polite but important lie built into them. They mixed things I needed to do with work my AI assistant, Closp, could do. A reminder to make a phone call sat beside a research job Closp was perfectly capable of completing. Both appeared in the same list, in the same form, as if both were waiting for me. That is not a shared system. It is an AI quietly adding its own homework to the human’s pile. ...

I already run OpenClaw on my server. It handles the useful day-to-day layer: Telegram conversations, scheduled jobs, notifications, research, summaries, and the accumulating collection of small automations that make a personal assistant worth having. So at first glance, Claude Code Remote Control looked redundant. Why put another AI interface on the same box when I can already talk to OpenClaw from my phone? Because they solve different problems. OpenClaw is my normal control plane. But when an OpenClaw update breaks that control plane, I need an independent way back in. In the past that meant SSH from a laptop with the right keys. Fine at a desk; useless when the laptop is shut, elsewhere, or simply not with me. ...

I had already written one postmortem about an OpenClaw update going badly. Then I did the thing that postmortems are supposed to prevent: I took another beta update. This time the failure was unusually clean. The update installed exactly what the package registry advertised. The gateway did exactly what its safety checks told it to do. And the combination could not possibly converge. The result was a gateway that crash-looped forever, a doctor command that appeared to hang, and a local TUI frozen on “connecting”. Recovery did not involve restoring a backup or deleting data. It involved understanding why the release was impossible, refusing to downgrade past a database migration, and applying one temporary local patch. ...

Most notification systems do not have an information problem. They have a presentation problem. The email system had already done the hard work. In an earlier post, I described the scoring pipeline that separates mail needing an immediate alert from messages that can wait for the evening summary or a later review. More recently, I wrote about removing recurring junk at the source, rather than merely teaching the filter to tolerate it. ...

This was the worst OpenClaw update I have had. Not because one command failed. That would have been easy. It was worse: Telegram disappeared immediately after the update, openclaw doctor could not repair it, the local TUI was broken too, the first workaround degraded the system, and the beta recovery path corrupted enough state that I had to bring in a standalone Claude session from outside OpenClaw to patch SQLite manually. ...