Daily field notes on AI & operator lessons you can actually use. Applying AI across code, design and marketing. 2X founder
Joined January 2017
- Tweets2.1K
- Following246
- Followers388
- Likes1.4K
The claude.dev automation writeup that ClaudeDevs shared reads each source from a bookmark, the timestamp of the newest item seen last run. A fixed 'last 24 hours' window, it says, leaves a gap on a late run and repeats items on an early one.
How to build automations with Claude Managed Agents: claude.dev/blog/building-eff…
Deploy an agent that reads Slack/GitHub on a schedule (with credentials and memory) and posts updates. You can set it up with a Claude Code command and use the new API credits included in Max and Team.
Codex Cloud wants a Tailscale key set to Reusable and Ephemeral, so every task VM can join and then disappear from your tailnet once it goes offline.
Codex 🤝 Tailscale
Codex Cloud can now securely connect to resources on your tailnet with Tailscale.
Useful tools should be easy to build with, combine, and turn into something even more useful.
See how it works →
buff.ly/yFcldrC
That quick image in Codex comes out of your coding budget. Image turns use the allowance 3-5x faster on average.
The $200 Codex plan feels reasonable with 6.1 Sol. Not hard to hit the limit, but not too easy either.
Astra feels pretty close to unusable.
This gives Claude's plans read receipts. It records the decisions you never opened so Claude knows to ask again.
A Claude Code tip going around says to run /advisor fable so Fable 5 reviews your Opus session. Anthropic's advisor docs say Claude Code rejects that command. Here is what the advisor accepts, and what it costs.
Cost: advisor tokens are charged at the advisor model's rates, on top of your main model. On subscription plans they count toward your usage limits.
The advisor's read of the conversation isn't cached, so each call processes the full transcript again. Toggling it mid-session doesn't invalidate the main model's cache.
It is experimental and needs the Anthropic API. The docs list Bedrock, Google Cloud and Microsoft Foundry as unsupported. /advisor off clears it, and CLAUDE_CODE_DISABLE_ADVISOR_TOOL=1 disables the tool.
Check the pairing table before you copy a tip.
Source:
Escalate hard decisions with the advisor tool | Claude Code Docs
code.claude.com/docs/en/advi…
Hot take: Muse is quite dumb compared to Claude and ChatGPT. It might be good at admin tasks. But my experience with other things has been quite bad. 👇
The tell is the repetition. I give the same instruction over and over, refining it each time, and it still doesn't reach the quality I get with other AI models.
It handles the boring operational glue of the week, tirelessly, and the output holds up. Inbox triage, renewals, reminders, follow-ups, all covered without me thinking about them.
The admin lane is genuinely good. It goes through a stuffed inbox and pulls out what needs a reply. It keeps track of subscriptions and renewals. It drafts follow-ups, handles routine customer complaints, sends the reminders.
Claude Code mods can approve or deny your permission prompts. Anthropic's own launch post says they aren't sandboxed.
So what stops a bad mod, and who gets that protection by default? The answer depends on your plan.
Load order matters here. When several mods hook the same event, the first mod loaded sees the event first and the result last.
That is why Anthropic suggests an audit mod that loads first, since it can record every call every other mod makes.
If you run Claude Code on your own machine without managed settings, the post doesn't say sec-default applies to you. Read a mod's source before you install it.
Source: claude.com/blog/claude-code-…
