Chris Thoburn

Two years ago, Sohil was building an agent framework and many of the primitives commonplace today (LLM router, tools, etc.) before they even existed. Biased but I would recommend this talk.

19

2 Comments

Founder observation after dozens of conversations in SF: Teams love to say “we’ll build it in-house.” What they underestimate is the experimentation cost. Invite flows, collaboration timing, and role sequencing need iteration, not one-time implementation. Velocity beats ownership more often than people admit.

13

Many of you ask: What exactly do you do? We match companies with fractional CTOs/CPOs and vetted dev agencies across five engagement models to build software with a human-in-the-loop: — Project work Going to a restaurant. You order. They cook. You eat. — Team augmentation You are cooking at home. Tonight is ambitious. You bring in an extra pair of hands. — Build-Operate-Transfer You love the restaurant so much, you open your own. We build it, run it, then hand you the keys. — Product discovery Before you cook anything, you taste-test. Figure out if the dish is worth making at all. — Fractional CTO/CPO You do not need a full-time head chef. But once a week, you need someone to fix the menu. We match you in hours as we have all the data to triage to your requirements.

20

20 Comments

Historically, engineers have stared at terminals all day. Today, perhaps some stare at Claude/Codex all day whether it's the CLI or app. An old coworker of mine once said, "if you are building a product people use daily, they should feel delighted using it." Design is a top priority for us at Galaxy - both DX and UX. Now that AI can spin up a working tool in minutes, design is one of the few differentiators. One fun thing we did on the Filament landing page is a train station departure board, every row a source and a destination flipping by. Filament is our open-source data movement engine and our whole pitch is that it's free and fast, so the page had to represent that. Watch the counter in the corner climb. The "rows moved" counter is moving at 2.6M rows per second, just like Filament does for Postgres to Postgres replication. 🥲 Would love to hear what you think! https://lnkd.in/eXBFeuGg

17

2 Comments

The best sign your team is moving fast? When other engineers start pointing it out publicly (unprompted). This type of speed doesn’t come from top down mandates (you can't fake it). It comes from a culture where everyone feels real ownership and obsesses over the customer experience. Speed is the only real moat startups have. And once you lose it, it’s almost impossible to get it back. That’s why we’re constantly watching for early signs of drag and doing everything we can to protect velocity. For us, that starts with who we hire. We’ve taken our time building the team, and set a very high bar for who we bring on. It's also a reason why 50% of our engineering team are former founders, who naturally bring that mindset to everything they do. That compounds week over week. And before you know it, people outside the company start noticing too.

57

4 Comments

We're launching Gambit, an open source agent harness designed for reliability. If you’re not familiar, agent harnesses are sort of like an operating system for an agent... they handle tool calling, planning, context window management, and don’t require as much developer orchestration. Normally you might see an agent orchestration framework pipeline like: compute -> compute -> compute -> LLM -> compute -> compute -> LLM we invert this so with an agent harness, it’s more like: LLM -> LLM -> LLM -> compute -> LLM -> LLM -> compute -> LLM Essentially you describe each agent in either a self contained markdown file, or as a typescript program. Your root agent can bring in other agents as needed, and we create a typesafe way for you to define the interfaces between those agents. We call these decks. Agents can call agents, and each agent can be designed with whatever model params make sense for your task. Additionally, each step of the chain gets automatic evals, we call graders. A grader is another deck type... but it’s designed to evaluate and score conversations (or individual conversation turns). We also have test agents you can define on a deck-by-deck basis, that are designed to mimic scenarios your agent would face and generate synthetic data for either humans or graders to grade. Prior to Gambit, we had built an LLM based video editor, and we weren’t happy with the results, which is what brought us down this path of improving inference time LLM quality. We know it’s missing some obvious parts, but we wanted to get this out there to see how it could help people or start conversations. We’re really happy with how it’s working with some of our early design partners, and we think it’s a way to implement a lot of interesting applications: - Truly open source agents and assistants, where logic, code, and prompts can be easily shared with the community. - Rubric based grading to guarantee you (for instance) don’t leak PII accidentally - Spin up a usable bot in minutes and have Codex or Claude Code use our command line runner / graders to build a first version that is pretty good w/ very little human intervention. We’ll be around if ya’ll have any questions or thoughts. Thanks for checking us out! https://lnkd.in/evPNf2bE Walkthrough video: https://lnkd.in/etw__HbA

13