I’ve spent the last few weeks playing with Clawdbot. My instance is named Clawd.
If you haven’t seen this category yet: think “chat assistant”, but with hands. It can run commands, write files, poke your integrations, and generally do the annoying glue-work you normally do by tab-switching and copy/pasting.
TL;DR
Clawdbot lets you go from “I have an intent” to “I have an artifact” (a file, a diff, a message, an action) without leaving the chat.
It’s a bit clunky and you’ll notice the seams, but once it’s set up it feels weirdly magical.
Also: agentic workflows burn through credits faster than you think.
1) I underestimated how much of my procrastination is just “file chores”
I had this vague feeling that a lot of the things I do (or want to do, but keep procrastinating on) are not really “hard problems”. They’re a pile of small, annoying steps.
Once I started using Clawdbot, it became obvious how many of those steps are basically file system chores:
- moving things around
- renaming things properly
- listing what’s in a folder and spotting duplicates
- organizing a messy directory into something that makes sense
- reporting on the organized mess
And sometimes it’s not even about organizing for its own sake. It’s about using the file system as temporary storage for an ongoing action.
For example: create a file, dump some intermediate result into it, and then have another tool or program pick it up and do the next step (cleanup, conversion, upload, whatever).
That’s where an agent that can actually touch your files starts to matter. It turns “I should do this at some point” into “ok, let’s do the next tiny step”.
2) Connectors matter — and fallbacks matter even more
“Agent” is an overloaded word. In practice, a lot of the day-to-day usefulness comes down to: what can it connect to? And when it can’t, can it still help?
Clawdbot surprised me: it has a bunch of connectors (messaging, filesystem, shell, browser automation…), and it’s also happy to take detours when a direct integration is missing.
The calendar use case
I was planning a weekend trip around a festival, and I wanted the schedule in my calendar so I could see what was actually feasible.
Instead of constantly checking the event page during the trip, I asked Clawd to put the festival schedule in my calendar. Doing this by hand would have been very time consuming.
I pointed Clawd to the event site and asked it to add multiple events from the Saturday schedule.
There wasn’t a perfect one-click Apple Calendar connector at the time, so the workflow became “get the structured schedule into something my calendar understands”. Clawd ended up generating an .ics file and launching a popup for me to click “accept” and import.
3) Using it from my phone made the whole thing click
At some point I realized that about half of the Twitter posts I’d read about agentic assistants had some version of: “using it from my phone is awesome.”
I didn’t fully get it until I tried it.
When I’m not sitting down at the computer, the phone chat is the ideal interface. Low-friction, always there, and it fits into the tiny gaps of the day.
The moment it clicked was that Clawd could work on several tasks while I was doing children logistics, and I could check back whenever I had a pause. Send a message, go do the next thing, come back later to a file or a summary.
Somebody is working while life happens. That’s the dream.
The calendar query I didn’t know I wanted
One unexpectedly useful prompt was to figure out which days of the week are the heaviest: meetings pile up in my calendar and conflict with my to-do planning.
Which days this week are the heaviest, considering meetings and reminders/to-dos?
For me, “heaviest” meant more than 4 hours of meetings plus more than 4 to-dos.
I then asked Clawd to redistribute the to-dos to other days over the next three weeks, assuming each task would take about an hour. The replanning was solid—and easily saved me 20 minutes.
4) Setup was quicker than I expected
I was productive in under an hour. What took longer was making it feel reliable: setting defaults, fallbacks, adding guardrails, and figuring out the meaning of some messages.
The unsexy downside is cost. I hit my OpenAI Codex credit limit faster than expected, because agentic workflows tend to involve more steps (and more tokens) than a normal chat.
5) It’s clunky… and it’s probably a stepping stone to something better
It’s still a bit clunky. It reminds you that you’re using an early version of an open source application.
But once it’s set up, it can feel genuinely magical. There’s something deeply satisfying about stating intent in plain language and getting back a real action or a complete file.
Bonus 6) It’s not a money printer
All those posts about people making themselves rich on Polymarket using clawdbot are 100% cryptobro fantasies and scammers trying to sell you a copytrading bot. No, you can’t successfully arbitrage that way, no you can’t “read” noiseless twitter sentiment that easily.
The phone part is what sold me. If I can offload a task in the middle of the day and come back later to a finished artifact, that’s not a chatbot anymore. That’s closer to a real assistant.
