Alex Finn (Creator Buddy): delegating tasks to a Clawdbot AI agent by chat

Alex Finn, CEO of the SaaS product CreatorBuddy, assigns work to Clawdbot (now renamed OpenClaw) through chat messages. He has publicly shown the agent booking a restaurant by phone after an online reservation failed, and says it coded on its own for 48 hours straight. None of the results have been independently verified.

Summary

Item	Detail
Company	Creator Buddy (Alex Finn)
Industry	Consulting
Use cases	Business intelligence, process optimization
Functions	R&D, marketing
Reported result	No cost or headcount-replacement data disclosed; token spend is hard to predict (one developer reported $300 in two days); default setup carries gateway-exposure risk
Moments rating	A-tier recommended
AI use-case score	3.65

Why it matters

The example shows an AI agent switching tools when the first one fails, which is the behavior that makes agents useful for real errands. The same sources also document the costs and security exposure that come with giving an agent terminal, file and key access.

Case details

Background

Clawdbot is an open-source AI assistant built by Peter Steinberger, founder of PSPDFKit (now Nutrient). It can browse the web, run terminal commands, manage files and make phone calls, and users instruct it through chat apps such as WhatsApp, Telegram, Discord, Slack, Signal and iMessage. It uses MCP to connect models like Claude and GPT to real actions, keeps persistent memory and can run locally (Decrypt, January 26, 2026). The original clawd.bot showcase now redirects to openclaw.ai, as the project has been renamed OpenClaw. Alex Finn is founder and CEO of CreatorBuddy.

What they did

Finn named his Clawdbot “Henry” and hands it work by message. His public example: he asked Henry to book a restaurant for the following Saturday. The OpenTable booking did not go through, so Henry used an ElevenLabs skill to phone the restaurant and complete the reservation. According to InfoQ, Finn also said Henry spent 48 hours non-stop “vibe coding”, the first time since Claude Code launched that he went two days without using it himself. He said he planned to replace his Mac mini with a $10,000 Mac Studio, using Opus as the “brain” and local models as the “staff”.

Results and data

Every source read is Finn’s own account. There is no third-party verification and no data on time saved, cost or output. Earlier claims that the agent built a project management board, replaced PM, developer and operations roles, or cost $15-25 a month do not appear in these sources. The “$300 in two days” figure came from an anonymous developer on Hacker News, not from Finn. On security, researcher Luis Catacora used Shodan to find Clawdbot gateways exposed to the public internet without authentication, which could hand over shell access, browser automation and API keys.

What to take from it

The agent recovered from a failed tool by taking another route (online booking failed, so it called). Giving an agent several skills is more reliable than wiring it to a single API. Small teams can start with low-risk errands whose outcome is easy to check, such as reservations and scheduling.

Limits

A solo founder’s demo is not proof of enterprise readiness. These agents hold terminal, file and credential access. Before any company deployment, at minimum bind the gateway to localhost only and add authentication, key rotation, rate limits and logging with alerts. Token usage is hard to forecast, so set a budget cap.

Source

Original source: Clawdbot showcase

Additional source: Decrypt · open-source AI assistant testing the limits of automation and safety

Additional source: InfoQ report (via 36Kr)

This is an industry research entry: a publicly documented third-party AI case, not a Moments client case. For projects Moments delivered itself, see client cases.