Mastermind · April 12, 2026
Everything is a CLI now
A week ago I would have called this a technical detail. This week it's the whole strategy.
At some point on Saturday, I said out loud to Xisen: everything is CLI. MentorMates is CLI. Pulse is CLI. The new submission flow I'm shipping is CLI. Claude Code is the surface the whole stack lands on. That sentence came out of me as a joke, and then we spent the next twenty minutes unpacking it and realized it wasn't one.
What changes when the surface is a CLI is that partnerships stop being contractual and start being technical. When a sponsor funds an event, the user installs MentorMates' skill, and the sponsor's skill auto-installs on top of it. That's --include-sponsor-skill on a command, not a partnership agreement. You didn't sign anything. You didn't negotiate distribution windows. You shipped code, and the integration happened.
A year ago the question how do we partner with a new ecosystem player meant a four-week back-and-forth, a contract, a marketing plan, and probably a Google Doc with comment threads. This week the question is does their skill expose the right command, and does ours know how to install theirs. The answer is discoverable in an afternoon and the test run happens by Monday.
There's a version of this that's just technical. APIs always reduced friction. CLIs are nothing new. But the thing that changed isn't the availability of the primitive, it's that the partner who used to need a BD motion to install our product now just types one command. The distribution cost collapses. And the flip side is that their product becomes one command to us too, which means we stop building things ourselves if someone else already shipped the skill for it.
That's the part that rewires the operating model. We're running two hackathons next week. In both of them, the entire sponsor experience is going to be: download MentorMates skill, and if your event is sponsored by a partner, their skill auto-installs on top of it. There's no physical booth dependency, no manual integration, no your-team-will-be-onsite-to-support clause. Our sponsors' surface area inside our events is exactly the commands their skill exposes, nothing less, nothing more. Pricing a sponsorship suddenly has to get honest. What are you actually paying for? The shipping skill, or the boost to install count?
For MentorMates specifically, this collapses the business-model question. We're not a SaaS with a pricing page, we're an install-surface with a skill catalog. The unit economics become: cost to acquire an installer, expected lifetime skills installed per installer, revenue share per skill install. I haven't priced that out yet. Doing it this week.
Xisen and I also talked about agent networking at the event layer, and the click for me was that when everyone at a hackathon has Pulse installed, and everyone installs the MentorMates skill, the agents on everyone's machine are already talking to each other through the same protocol. Project discovery, teammate matching, judge-participant briefings, sponsor introductions: all of it becomes something like pulse find-teammate. The hackathon as a whole becomes a shared agent runtime. That's a weirder vision than I was ready to say out loud a month ago, and this week it just felt obvious.
There's a risk inside all of this that I keep circling. A mentor told me earlier this quarter that the thesis doesn't matter without a success story to point at. I've been holding three or four directions at once because they're individually plausible, and the everything-is-CLI frame makes all of them more plausible, not less. AI VC. 1-to-100 instead of 0-to-1. Education as the long arc. Event infrastructure as the near-term wedge. Agent networking as the middle-term product. Those threads have gotten easier to argue for in the last month because the underlying substrate got better. That doesn't mean they're compatible. Too many compatible directions is how founders fragment.
The discipline for the quarter is to pick one observable outcome across all of it and let everything else be downstream of that. The Hong Kong flagship hackathon in July is the outcome. When a thousand people walk in, every one of them installs MentorMates, half of them install Pulse, and the agent networking runs end-to-end in public. If that weekend is a success, the thesis has a story. If it fails, none of the thesis matters. It's that binary.
The other thing that came out of the call that I keep thinking about is the meta-observation. Somewhere around minute forty, one of us said: we should just auto-publish these calls as essays. The quality of the conversation is decent, we already have the recording, the format exists, and we both need a writing cadence for distribution anyway. We also both know that writing down what we said post-hoc is the thing that prevents drift week to week.
That's where this post comes from. The skill exists now. The blog is live. The cadence is set. And yes, this is the first public post that names where it came from: a Saturday conversation, a throwaway remark, and a decision that if everything else is a CLI we should probably ship our writing the same way.