#012 - Introducing Agentsy
May 15, 2026
Enjoying this newsletter?
Subscribe to get future issues
Agentsy, the software factory stack, and a head full of SF conviction.
May 14 2026
AI has changed how many things I can run at once, and I’m still getting used to the new normal. The last six weeks have been heads-down across more projects than would’ve been possible a few months ago, with a detour to San Francisco for the Claude Code Conference last week.
I came back with a head full of conviction about where agent tooling is heading. The next couple of issues will dig into that. First, the big one: we just launched Agentsy. Then a catch-up on everything else.
Launching Agentsy with Marius
I launched Agentsy with my longtime collaborator and good friend, designer Marius C. Paun. The landing page is live at agentsy.build.
The pitch: senior design and AI engineering, embedded. Three disciplines, one team, working directly inside your business to build the brand, the systems, and the infrastructure that work without you in the room.
Marius and I have been talking about doing something together for years. Shipping it has been a dream come true. I’m especially excited about how the brand and positioning landed; Marius’s design, my engineering, both of us with strong opinions about how AI-era services should be delivered. The landing page is the first artifact.
We’re just getting started and will be publishing more about our work and our process in the coming weeks and months.
What else I’m building
A few other moving pieces, in no particular order.
PHAiTO and Image Salon. I’m their Fractional CTO. PHAiTO is about to relaunch with an updated design and a stack of new features for photographers; launch is imminent and there’s a lot of good energy going into it. Image Salon is expanding its wings beyond photo retouching services. We’re building something I’m very excited to talk about… in the future.
Open source. Three repos I recently published:
librarium: a CLI that fans research queries to over a dozen search and deep-research APIs in parallel. A great tool to equip your agents with for when you need to do some research. Goes way beyond the built-in web search you get in Claude or Codex. Comes with a skill for easy use.
graft: the easiest way to interact with Git operations and the GitHub API in Laravel. Branch, commit, and ship from your Laravel app, without ever shelling out by hand or hand-rolling the GitHub API.
plume: wraps the entire X.com API v2 behind a clean Laravel facade. Typed DTOs, automatic pagination, user-scoped operations, OAuth token refresh, rate-limit handling, artisan commands, and AI tools for the Laravel AI SDK. The easiest way to interact with twitter from a Laravel application.
Software factory stack. I’ve developed my own internal tooling that I use to run everything. A multi-provider, multi-model harness, a memory system, a task management and planning system, and a code review setup that beats the commercial solutions. Working on it actively. Not sure yet what I’ll release and what’s just for me. If you’re reading this, let me know what you’re interested in!
A few days in SF
I spent a few days in San Francisco for the Claude Code Conference. Walked out with new friends and a clearer sense of what I want to build.
I finally met Ahmad in person after years of being Twitter friends; he’s building Command Code, a new coding agent focused on DeepSeek v4 and getting some really good performance out of it. I also got to know Katerina and Vanessa, two sharp Bay Area product designers. Katerina gave me a phrase I’ve been chewing on since: “a design eye for an engineer.” Exactly the thing I’m trying to lean into. Thank you!
The flag I want to plant: building software in a more calm, more process-oriented way than what’s currently out there. Most of the agent tooling I see right now is built engineering-first, and it shows; it’s the wild west of agent town. There’s room for a different approach. Something considered. Something where the human stays in the loop in a deliberate way, not as a panic button.
The next couple of issues will dig into what that could look like.
Cool stuff from the internet
Tech Differently: Speaker Accelerator + Conference
Program details · Free info sessions this week
My friend Jill is building something I think is genuinely cool. From May to October she works with you to develop a talk that only you can give, the kind tech audiences need to hear. Then you deliver it live at the Tech Differently Conference in October. If you’ve ever wanted to speak at a tech conference but felt out of place, weren’t sure you could, didn’t get your talk accepted, or did and want to do it better, this is for you. She’s running free info sessions this week.
A Codebase by an Agent for an Agent by Tim from Amp
Tim flips the frame: instead of designing your codebase to be readable for humans, design it to be readable for the agent. Let the agent pick the structure, naming, and conventions that match its own probabilistic priors. Provocative idea worth sitting with.
How to really stop your agents from making the same mistakes by Garry Tan
Garry’s “skillify” pattern: every time an agent fails or hallucinates, you turn the lesson into a deterministic skill with tests, so the same mistake becomes structurally impossible. Spec-thinking applied to agent reliability. This is the right mental model for skills.
Keep shipping, Joey
P.S. Hit reply and tell me what you want me to dig into next. The software factory stack, the calm-software thesis, something I mentioned in passing; I’m listening.