Skip to content

Array AI Labs

We are builders before we are investors. The lab is where we try new models and agents in the open, on our own work first, and write down what we learn.

2 experiments · 4 notes · since May 2025

Research areas

If you are researching or building in one of these, email us at deals@array.vc.

  1. Memory and context

    How agents remember people, priorities and past work, and share it with other agents.

  2. Harnesses and sandboxes

    Where always-on agents run: their sandbox, their schedule and the channels we reach them on.

  3. Loops, evals and AI spend

    Designing agent work as loops with gates and budgets, and measuring what each result costs.

  4. Relationship management

    Agent-native CRM: tools that let agents track people, log calls and brief us before meetings.

  5. Voice agents

    Agents you can talk to, from AI office hours with Shruti’s clone to generated podcasts.

  6. Environments and protocols

    How agents act on websites, real phones and each other, through MCP, WebMCP and A2A.

Experiments

  1. Essay ·

    Loop to Graph Engineering for Measuring AI Spend

  2. Field notes ·

    How Our 24/7 AI Coworker Replaced Half Our Internal Tools

  3. Explainer ·

    Web World Models, a Simple Explainer

  4. Release ·

    Office Hours: Grateful For You

    Office Hours with Array VC

    Tell the agent who you are, then talk it through.

    Voice by default, uses your microphone. You can type instead.

  5. Experiment ·

    AI Generated Looks and Videos

    Read the write-up +

    So our model friends turned me into a model! Using Nano Banana Gemini 2.5, Kling 2.1, and Seedance 1.0, we took regular photos and transformed them into Paris Fashion Week runway looks and even Taylor Swift-worthy engagement outfits. No photoshoot required.

    The wildest part? These AI models don't just swap clothes - they understand fabric physics, lighting, and how different outfits would actually move and drape on a person. We went from "wouldn't it be cool if..." to "wait, this actually looks real" in about 30 minutes.

    What started as us playing around with AI tools turned into something bigger: a glimpse at how creative industries might work tomorrow. Imagine trying on clothes without buying them, or seeing yourself in any outfit from any era. The fashion industry's $3 trillion market is about to get a whole lot more interesting.

    Built with Nano Banana (Gemini 2.5), Kling 2.1, Seedance 1.0

  6. Experiment ·

    AI Generated Podcasts

    Read the write-up +

    At Array Ventures, we're constantly experimenting with the latest AI tools. It helps us stay on top of new developments in AI and just as importantly, it sparks ideas on how we can better support our founders with go-to-market strategies and product innovation.

    Recently, we've been building some internal AI tools, and one experiment came from a pain point I had while running the [Array] podcast a few years ago. We had some incredible guests who shared unique insights but I often found myself thinking, "Haven't they said this elsewhere?" With so much content already out there, I started wondering if we really needed to take up their time to get the same information again.

    So when Notebook LM launched, a lightbulb went off: What if I could solve two problems at once?

    • Relaunch the podcast without a heavy lift on production
    • Capture expert insights without requiring time from the guest

    The result? AI-generated podcasts featuring insights from top entrepreneurs and VCs, created entirely from their existing interviews and content available on the web.

    Built with NotebookLM, Veo 3, Hedra, ElevenLabs, Runway, Kling, HeyGen, Synthesia