This is not a drill.
For the past 20 days, I’ve been obsessed with an open source platform called OpenClaw.
It’s a simple piece of software that you can run on a Mac, PC, or in the cloud, and it’s everything we were told Siri and Alexa would be.
It’s dangerous, disruptive, and inspiring, because once you authenticate it with your online services, it actually “does things” as opposed to “tells you things.”
ChatGPT and LLMs are fantastic researchers and archivists.
They’re making middle managers 100% more effective, instantly, by summarizing endless meetings, creating action items, and preparing thorough plans on which humans can execute.
But they don’t execute, because the companies that make them are scared to death of the mistakes they would make.
When you install OpenClaw, you assume that responsibility.
OpenClaw will not just research potential guests for your podcasts. It will find their contact information, email them an invite, build you a CRM to manage these contacts, and schedule the appearance on your calendars.
It will fill your Uber Eats cart and order lunch for you. If it fucks up, that’s on you – because you gave it your password.
At my 20-person venture and podcasting firm, we’ve offloaded about 20% of our tasks to OpenClaw in 20 days, and it’s getting better every day because of “skills” and memory.
Skills are built by either you or third parties, and function like apps. My favorite so far is /last30days by Matt Van Horn. You can instruct your agent (aka “replicant”) to research the best practices for any skill on the web and then incorporate those learnings into its skill stack. (Shout-out to Scott Adams, RIP.)
We did this with thumbnails, and now our Producer Replicant conducts a deep research project every Saturday at noon, finding people discussing how to make better thumbnails in forums like Reddit and elsewhere around the open web.
Unlike humans, this replicant doesn’t take breaks and doesn’t forget. It just relentlessly makes itself better.
It’s recursive.
At this point, I don’t think we will hire another human for at least a year or two at our firm, but our efficiency will increase by at least 10% a month – the equivalent of two new full-timers.
Why?
Because it’s easier, faster, and more cost-efficient to set up replicants.
The replicants burn $1,000 in tokens a month (but never ask for a raise).
The replicants don’t make mistakes.
The replicants will get 10% better at their jobs each week, which typically takes humans about a year.
The replicants don’t complain, they train.
Adding a human to our team will be a luxury spend for now. Something we do because we can afford to. Something we do because we enjoy having another interesting person around.
And everyone on the team now buys in, because chores are not fun. Every time we tell a replicant to take over repetitive chores, we get to spend more time with other humans – like founders, podcast guests, and one another.
This isn’t dystopian. It’s nirvana!
It’s Star Trek’s vision of the world, in which humans solve problems alongside “Computer.”
It’s here.
This is not a drill.
Best, JCal
PS - We’ve been covering OpenClaw on This Week in Startups – a lot! Here’s a YouTube playlist.
PPS - We are launching a new podcast called This Week in AI, you can subscribe on YouTube, Spotfiy and Apple. You can also sign up for our newsletter here.
PPPS - We are looking to invest in OpenClaw-related startups. If you have one, email us what you’re building, why and a bit about yourself: openclaw@launch.co