The Rise of Shared AI Workspaces

For 40 years, enterprise software giants have built systems of record—ERP, CRM, and HRIS—static databases that look backward. Their core workflow remains unchanged: an interaction occurs, a decision follows, and a human, often reluctantly and imprecisely, logs the data for future reference.

But what if the system wasn’t just a storage vault for structured data? What if the system was the conversation—with AI participating, augmenting, and automating work in real time?

A New Paradigm: The Shared AI Workspace

This is the shift I’m betting on: the rise of vertically-focused shared AI workspaces that replace static systems of record with intelligent collaboration layers. Instead of structured fields waiting for human inputs, these AI-native workspaces will work like persistent, shared ChatGPT windows— deeply aware of the participants, their work, and the industry in which they operate.

Rather than requiring users to input data, these systems continuously learn from interactions, manage work automation, and dynamically structure information in real time. In effect, they become AI intermediaries that not only record but also understand, anticipate, and act on business processes.

And unlike legacy horizontal systems of record of the past, my bet is that shared AI workspaces will form around industry-specific networks, where domain expertise and real-time collaboration unlocked by bringing network participants into a single system create compounding network effects. 

Industries Ready for AI-Native Workspaces

Shared AI workspaces will emerge first in highly fragmented industries with a high frequency of interaction—where coordination is critical, but existing tools are slow and manual. As these systems bring on more stakeholders and gain more context they can effectively “organize the industry” –becoming the connective tissue that aligns industry players, optimizes work, and drives compounding efficiency.

Some examples of industries we think are primed for AI-native workspaces to emerge: 

• Construction: Compliance, design iterations, and change orders require frequent interaction between owners, GC’s, subcontractors, and local permitting / building authorities
• Supply Chain: Inventory, delays, and logistics require real-time updates across suppliers, manufacturers, distributors, and retailers
• Agriculture: Forecasting, pricing, and distribution hinge on coordination and data sharing between wholesalers, retailers, processors, and logistics providers

In a shared AI workspace, information isn’t just stored—it’s translated in real time, ensuring everyone gets the context they need without manual rework. Legacy systems forced everyone to structure and re-structure information manually. LLMs—built for translation and context synthesis—eliminate that friction, dynamically shaping information for each participant in real time.

Ichi: The Shared AI workspace for Construction

Our first investment in a shared AI workspace is Ichi, a small but mighty team building the shared AI workspace for construction. 

Today, coordination delays and miscommunication between local jurisdictions, architects, and contractors over building codes are among the biggest bottlenecks in construction. Nowhere is this pain more severe than in the permitting process, where GCs routinely wait 6–18 months to complete plan reviews and secure permits, even for relatively simple projects. These delays significantly impact project costs, often adding tens or even hundreds of thousands of dollars for every month of delay.

Ichi’s initial opportunity is to dramatically accelerate the review process by bringing designers, builders, and local jurisdictions into a shared workspace that increases plan examiners’ productivity 10x. It achieves this through AI-powered automation and streamlined collaboration. The interface combines familiar chat-based tools—customized for the construction industry—with purpose-built workflows that help all parties achieve code compliance faster, and at higher quality.

Plan review and permitting is just the beginning of Ichi’s vision. As Ichi continues to bring on network participants and gain more context / industry knowledge, one can imagine the types of complex multi-stakeholder workflows that it will be primed to execute. For example:

1. A plan examiner writes a terse violation report.
2. Ichi expands that into detailed resolution paths for an architect and GC.
3. The architect executes on design changes in hours, not days or weeks. 

Instead of fragmented email threads and scattered PDFs locked in siloed systems, AI-native workspaces like Ichi automatically structure, analyze, and route information—turning frustrating workflows into seamless, actionable collaboration.

Moats in the age of commoditized software

The value to network participants of a shared AI workspace is obvious. But for founders and investors, there’s another critical point–in a world where the cost of writing software is rapidly trending to zero, the shared AI workspace creates network effects that form enduring moats. Unlike traditional SaaS, where switching costs are often low, shared AI workspaces become deeply embedded in workflows by continuously learning from industry-specific interactions. This accumulated intelligence creates a moat—competitors lacking access to these proprietary interaction datasets will struggle to match the same level of automation, accuracy, and domain expertise, while the AI’s vertical specialization enables it to orchestrate complex, multi-party workflows that generic LLMs cannot replicate.

Looking ahead

The next generation of enterprise software will be built around systems of collaboration—AI-powered communication layers that don’t just store information but actively shape, structure, and act on it.

These shared AI workspaces will emerge first in highly fragmented industries with a high frequency of interaction—where coordination is critical, but existing tools are slow and manual. They won’t look like an ERP, or feel like a CRM. They’ll feel like a living, AI workspace, where work happens fluidly without human operators needing to re-enter the same information across multiple places.

The fundamental job won’t change—but the way it gets done will. And everyone, I think, will be happier.