The State of AI at COTU.

We are at an inflection point for how venture capital firms operate. AI is no longer a tool you bolt on to existing workflows — it is becoming the infrastructure around which the best investment teams are reorganizing themselves.

At COTU, we've been building in this direction for over a year. Some of what we've done is deliberate and structured. Some of it emerged organically from individual team members experimenting on their own. The result is a patchwork of genuinely impressive automation alongside significant gaps that cost us time and signal every week.

This document is an honest map of where we are. It covers each of our four core processes — deal sourcing, due diligence, portfolio monitoring, and LP relations — assessing what's working, what's missing, and what our next step should be. The goal is to give the whole team a shared view, and to lay the groundwork for building the AI-native version of COTU.

Before mapping where AI fits, it helps to be explicit about what we're optimizing for. Our investment process is designed around seven principles that define how we want to operate as a partnership.

These principles translate into a clear decision pipeline with target timing:

We bias toward tools that integrate well over tools that do everything. Each tool in our stack does one thing exceptionally well — the goal is a setup where data flows automatically between them, not one where we manually move it between silos.

This is our most automated process today. The inbound pipeline is built around a centralized email address — dealflow@cotu.vc — that triggers a Zapier workflow the moment a pitch arrives.

The pipeline handles edge cases intelligently: if the email contains a pass, the AI extracts the rejection reason and updates the Attio status automatically. If a Docsend link is present, it extracts the URL and passcode, then downloads the deck directly into the company's Google Drive folder. Confidence scoring flags low-certainty extractions for manual review. Any deal arriving at dealflow@cotu.vc is fully logged, filed, and visible to the team within minutes — with zero manual work.

For outbound, we use Harmonic for founder discovery, but this remains largely ad hoc — no structured workflow or tracking exists yet.

Our DD process is designed for speed and rigor simultaneously — first call to term sheet in under two weeks, with all three partners engaged before any decision. Meeting notes are automatically captured via Granola, organized into folders by type, and founders are assessed on four dimensions.

The ambition is to auto-draft roughly 70% of the deal memo from notes, deck, and DDQ responses — with partners adding the remaining analytical layer and the system flagging missing questions. In practice, this vision is partially realized. Claude is used to fill the deal memo template, but each partner does this independently, without a shared, standardized workflow.

We have built a portfolio monitoring platform — backed by Airtable and a Lovable front-end — that extracts KPIs from founder updates automatically and surfaces them in a structured dashboard. The core workflow is validated and functioning as a POC.

The vision for this platform has sharpened significantly. A web dashboard is a solid foundation — on top of that, the priority direction is a Slack-native interface where any partner can ask a question and get an instant, data-backed answer without opening any tool.

Focus on the present, not the past. The priority is knowing how a company is performing now, and what we can do to help. Historical data remains useful as reference context — for example when onboarding a new team member — but the primary use case is current performance and actionability.

Where we capture data:

Our LP reporting process is one of the most distinctive applications of AI at COTU. Quarterly, we produce Fund I and Fund II reports. Ismail builds the PowerPoint from our template, Amir reviews and adds per-slide commentary, then ElevenLabs generates Amir's voice over each slide. The result is a narrated video presentation sent to LPs alongside the PDF on Docsend — distributed via Attio's bulk send. This personal touch is a meaningful differentiator that very few funds are doing today.

The next phase of COTU's AI journey is moving from automated workflows to autonomous agents — systems that don't just react to triggers, but actively monitor, reason, and act on behalf of the team.

The pattern that works: give each agent a name, a clear role, and access to the same tools a human teammate would use. Onboard them like employees. Let the team interact with them naturally.

The highest-leverage starting point is Maya — the memory agent. Organizational memory benefits every other process simultaneously: better DD alignment, better meeting follow-through, better outbound context. Build one thing well, and the next problem reveals itself.