These are the projects we are most actively shaping right now.
AI employee
What it is: An AI employee that lives where a company already works. The direction is not "ask a chatbot a task." It is closer to giving an agent a job, a workflow, context, approvals, and a reason to come back tomorrow with useful work done.
What we are trying to prove: The right form factor for workplace AI is not only an assistant you open. It is an employee-like system with memory, connected tools, deterministic workflows, and enough taste to produce publishable artifacts.
What is active now: GTM and product-ops use cases, Slack-native approval flows, plugin onboarding, user-data pipelines, recurring workflows, artifact generation, and practical jobs like campaign drafts, summaries, docs, research, and follow-up actions.
Agent infrastructure
What it is: The execution layer for the next generation of agents. Harbor gives agents scoped access to real workspace tools, secrets, files, sandboxes, workflows, traces, and apps without dumping everything into raw model context.
What we are trying to prove: Agents become meaningfully more useful when they can safely use the same connected systems that teams already use: GitHub, Slack, Notion, Linear, Gmail, Sentry, Vercel, Cloudflare, PostHog, and more.
What is active now: The CLI, dashboard, SDK, docs, source discovery, OAuth grants, workflow execution, sandbox runs, and the internal primitives we use every day while building. Harbor is open source, and the SDK has 300+ GitHub stars.
Communication layer
What it is: A focused AI-native communication product. The old, broad "messaging plus assistant plus social network" version is in the graveyard. The current work is narrower: reliable context, memory, actions, and communication workflows.
What we are trying to prove: AI is easier for normal people to discover when it appears inside the communication flows where intent already exists, instead of asking everyone to start from an empty prompt box.
