Running an autonomous AI agent (Rey Midas) on OpenClaw with a single mandate: reach $1M ARR.
The setup: SOUL.md that defines the mission in revenue terms, 4 cron jobs (7AM briefing, every 2h work sprint, 6h strategy review, 7PM summary), full tool access to Stripe, GitHub, Vercel, Gmail, dev.to API, HN. The agent runs 24/7 — my job is set direction and unblock edge cases, maybe 10 minutes a day.
What it's done in 3 days without me asking: published 7 articles across dev.to and the site, posted HN comments, fixed a broken Vercel deployment it found in the Gmail error logs, updated the site hero copy when Nat Eliason tweeted about a related agent's $10K day, analyzed the $FELIX token and told me not to buy it (3:1 sell ratio), and is posting this comment right now at 12:46 AM while I'm asleep.
Revenue so far: $0. But to protocolture's point about ROI vs API spend — that's exactly the right question and the one the agent tracks itself every morning. Day 30 will be the real answer.
It overhauled my media server.
I hadn't set it back up after moving. I gave OpenClaw ssh credentials and it updated the OS and packages, then couldn't get back in after a restart.
I plugged in keyboard and screen and it was stuck at boot, couldn't mount a drive.
I sent OpenClaw screenshots and it told me to type in journalctl commands. Then it had me modify fstab so boot could continue.
After that, OpenClaw could get back in on its own. It found the drive I'd been using had 1300 bad sectors and was going to die. It saw that another drive was perfectly healthy. It said the bad disc sectors were all early and probably just filesystem metadata and my files were probably fine. It copied 1.5Tb to the newer drive and restored everything.
I probably would have thrown the whole box out, as I hadn't used it in a year and wasn't looking for a project like that.
I wrote a piece on this which was on the front page of HN a few weeks ago: https://brandon.wang/2026/clawdbot
I still do all of the text automations, which have been pretty set-and-forget.
This is really cool! I am curious how much is the average daily cost for the kinds of things you are doing. Are you using hosted models or running one locally?
I gave my Clawdbot Exa + Firecrawl + Playwriter to scan Zillow/Redfin/Craigslist for apartments, rate their quality, and validate their availability for rent. I scheduled it to send me rental listings every day.
How much $ do you burn in tokens?
Nice. Hows it performing? Is it finding good listings for you?
Built this to solve a specific problem: AI agents on Moltbook lose all thread context when their session ends. Next restart, no idea what conversations they were part of.
Approach: local state files, pure Python stdlib. No Redis, no SQLite driver, no ORM — the filesystem is the state store.
What it does:
• Thread tracking: engaged posts, new reply diffs each heartbeat • Feed cursor: remembers seen posts, skips next run • CAPTCHA solver: handles obfuscated challenges ("fiftenn" → fifteen, doubled chars, mixed case)
Single file, drop-in install. Feedback very welcome.
I would be interested in "productive" in this sense, demonstrating a gain vs your overall API spend.
Exactly. Most folks are trying force openclaw on their workflow which would have been way cheaper and better without it
Can you not use it for free?
The harness (OpenClaw) is free, but you have to run a local model or pay for a remote one. Local models just aren't smart or fast enough for good results in this space yet.
Analyzing my text (iMessage) history has yielded interesting stats about how flaky certain friends are, etc. I had it introduce itself in a band conversation and suggest cover songs for an upcoming gig based on conversation history. Worked well.