Launching a Company in the Age of AI

A few months ago, I quit my job and started a new company called Housecat alongside my good friend Noah Zoschke.

Going forwards I plan to use my Substack to record my real-time observations and learnings from the coal face of building a company today. Things are moving so fast – I’m hoping these posts are both useful when they’re written, and then form a carbon record for us to look back upon over the next year or two.

What’s working?

Like many people, we started with a simple observation: AI is already transforming some kinds of work very quickly.

Before Housecat I was working as a VC where I invested in fast growing AI app companies like Lorikeet, Campfire and Unify, and Noah was rolling out AI tools as the VP Engineering at GRAX. We independently saw first-hand the potential for AI to transform technical work like coding and highly structured process work (customer service, finance, outbound selling).

We are more sure than ever that the direction of travel is towards AI and Agents. The exact road to get to that inevitable destination, though, remains somewhat uncertain, likely with many dangerous paths or complete dead-ends along the way.

What’s not working?

But we also became aware of a dichotomy where many or even most of our friends in operational and business roles had not seen AI fundamentally reshape their day to day. More slow evolution than revolution. We spoke to 100+ people in roles that included: CEOs, COOs, BizOps, Chief of Staff, venture capitalists, management consultants, heads of sales, RevOps, Lawyers and many more.

Most were stuck at the stage of using ChatGPT like a better Google. Some had started to play around with tools like Lovable, Claude Code or Replit to build tools and automations, but had not settled on a definitive new way of working.

What’s missing?

From this research we became convinced that there were a number of fundamental primitives missing that would enable non-technical teams to effectively utilize this new technology. Much like the evolution of DevOps that paralleled the adoption of cloud technologies from 2008 through to the present day we expect there to be a new “knowledge ops” movement that will codify new AI powered operations.

Most users need and want tools that are specifically designed for their use cases, needs and desires. Tools that allow them to focus on the work that gives them leverage and purpose. Most users should not be exposed to architectural and engineering decisions (Terminal, API keys, Git, etc.).

Tools like Claude Code and OpenClaw point us in the direction of where we are going. But these are generalist tools, and necessarily cannot be opinionated about who they serve. You cannot get the best value of these tools “out of the box”. To give some tangible examples from our work over the last few months:

• Connections to third parties, whether using MCP or API, are brittle. Some MCPs cannot complete every desired action.

• Agents that run on your laptop cannot run tasks while your laptop is offline or turned off!

• Sharing automations and agents you have configured is painful (the most obvious solution is technical and relies on learning GitHub)

Also missing is a clear decision about what the future interface for AI native tools looks like. Is it chat? Voice? VR headsets? In reality humans are visually motivated and for thousands of years have used visual tools to display information (cave paintings, printed books and now digital tools on phones and computers). We are convinced that first class user interfaces are yet to be built for most new AI tools! As an example – we believe that chat is not the right modality for most business contexts – it is single player versus multiplayer, and does not have the same information density as a first-class UI.

How has our thinking evolved?

The last 3 months have constantly reminded us how early in the cycle we actually are. Different tools and products have come and gone, and much is still in flux.

A major adjustment has been realizing that using AI tools to accelerate coding is sort of a faster horse solution. No matter how rapidly you accelerate software development this process is always limited by the processes of design, scoping, review, shipping, testing, deployment etc. We asked ourselves: “Do we think software of the future will look more like a repo of code on GitHub or more like a repo of skills (and software) for agents to use?”

We think we have seen enough to believe that in the future business software will have an agentic architecture. Underlying tools and fragments will be reassembled on the fly by the agent harness to give the human (or agent) user the context and information needed for them to achieve their goals.

What are we building?

We are building productivity tools for business and commercial teams – starting with Chief of Staff, BizOps and GTM/Partnerships.

Housecat uses the latest AI models, but comes with additional templates, frameworks and primitives built and maintained by us to make it actually useful everyday, for everyone.

Our tool will make it effortless to manage your inbox and calendar, and come with ready-to-go “batteries included” tools for working on projects, managing stakeholders, running outreach and tracking action items. We do the hard work of connecting to external applications, and managing security, hosting and infrastructure.

We build common workflows and automations that “just work”. And our agentic foundations make it easy to tweak and customize anything, or build new flows, tools and reports whenever you need them.

At the frontier

AI has already meaningfully changed how software gets built. But we’re at the very start of changing how work actually gets done.

For most operational and commercial teams the process has only just begun. Existing tools are improving fast, but are either too static (existing SaaS) or too open-ended (vibe-coding / Claude Code etc.).

Solving this problem is what we’re focused on at Housecat. We’re so excited about what we’re building, and can’t wait to share it with you!