By Vincent Chan
July has been an insane month for Capy, and I've been kind of radio silent. During my week in SF, and throughout the month, I've had the privilege of talking to some really amazing people — folks who have literally defined key parts of how the internet works and built unicorns from scratch.
Through those conversations I've made a few radical realizations that are about to shape the next evolution of my company and product, while still solving the needs of current users and customers.
-
For the past 4.5 months I've made CLI DX the main focus for Capy. That came from the initial idea that in the AI age, people will spend more time in their terminals and would appreciate better tools for the work they do there. That part is true. Where in the terminal they spend it is the critical detail.
Terminal use has gone up, but the time is going into LLM CLIs rather than direct interaction with shells. I predict even that will shift as agent orchestration tools like Conductor, Claude/Codex Desktop, and Cursor mature. Nobody wants to learn another CLI tool, no matter how good it is, when their agents are doing all their work for them.
One of my favorite design partners, Daniel Zoghalchali, made this especially clear. Capy has the best secrets management DX through a market-leading CLI, and we need to go far beyond the CLI to solve the broader UX problem secrets management faces. I still maintain that bad security practices are ultimately a design problem.
-
The next wave of companies will be agentic-first API companies with completely headless products, accessed by agents through standards like ID-JAG/XAA and AAuth rather than API keys.
Scott Motte, who created
.env, gave me an extremely tangible example: a 3D printing company where you sign up, create a design, print an object, and have it shipped to you without ever leaving an agent session. That's the clearest picture of the near future I've encountered, and visionaries like Dick Hardt, who created OAuth, are ahead of how it works. -
Michael Grinich built the Switzerland of enterprise readiness with WorkOS, which is the least "enterprise" enterprise product in existence. He credits much of that success to a bi-modality.
On one end, solve individual problems EXTREMELY well — SSO, auth, MFA — with focused, well-refined, easily adoptable point solutions. On the other end, compose those point solutions into a unified turn-key platform that's even easier to use and better than the sum of its parts. AuthKit is the perfect example. The former is hard, the latter is harder, and doing both at once is the hardest. Fortunately WorkOS pulled it off so the rest of us can build our own bi-modalities on top of the mountains they've moved.
-
I've been grasping for dear life onto an outdated notion that code is defensible — which is at odds with what I wrote about building Capy during the SaaSpocalypse. I think I've finally learned to let it go.
What clicked is that in the new world, code is no longer the relevant output. The relevant output is something more like Art. Art isn't a physical painting or printed notes on a sheet of music — those are commodities. The true value of art is the meaning, emotion, or ideas evoked by the medium. Code is now one commodified level of abstraction below the true output, and like art, the true output is something akin to meaning or emotion. I don't claim to have solved this fully, but July has been a step along that path.
To that end: once Capy's code is hardened enough, I'm open sourcing all of it — infrastructure, CLI, services. The moat has always been the roadmap, the factory that builds it, and how our human judgment evolves over time. No current AI can steal that from us, unless perhaps it's whatever Yann LeCun is working on. Open source is a channel of distribution that enables a greater, purer appreciation of the true output that lies above the code.
-
Speaking of open source, I had an insightful chat with Glenn Gillen. He cautioned against overindexing on open source adoption alone as a signal of product traction. There's a wide range of outcomes for people involved in successful open source software, and the path from creating a mass-adoption project to enduring relevance is rarely a straight line.
His cogent example was Kelsey Hightower, who didn't create Kubernetes but is one of the industry's most important voices for it. That got me reminiscing about TJ Holowaychuk of Node.js fame, his massive impact on the modern web, and the disproportionately small reward he got for it. He created apex/up, essentially what Vercel would become, many years before Guillermo Rauch turned it into a multi-billion dollar company.
July has been a huge month — for clarity and direction more than feature output. Watch this space, and prepare for what's next. It's coming sooner than you think.