Collaborative Workspace.
Wikis, notes, and project surfaces where shared thinking accrues into a queryable corpus.
Overview.
Every firm has two kinds of information. There are the facts — who the company is, what round it raised, the valuation, the status of the relationship. And there is the thinking — the half-formed thesis, the debate, the draft memo, the pros and cons that get rewritten ten times before anyone decides anything. Most firms throw both into the same place and lose the ability to tell them apart. This module is about giving the thinking a home of its own.
Most VCs don't actually need a heavily structured productivity or project-management tool. Startup evaluation isn't a pure sales cycle — there aren't always repeatable steps, and a single deal can take months or even years of relationship-building that no pipeline stage cleanly captures. Force that work into a rigid process tool and it starts feeling like a wall of half-open tickets and missed dependencies, when really you're just nurturing something on a longer clock.
The collaborative workspace is where shared thinking accrues. It is the layer where the deal team works a company over time: writing opinions, tracking diligence, arguing in comments, capturing the nuance that never fits in a database field. Done well, that accumulated thinking becomes a corpus your AI can query — a structured record of how the firm actually reasons, kept deliberately separate from the canonical facts in your CRM.
Our recommendation is Notion. It is flexible enough to model exactly how your firm works, interoperable enough that your LLM can read it, and intuitive enough — once the team gets the hang of it — that people actually use it. You can substitute another tool, and we say where that holds and where it does not. But the architecture matters more than the vendor, and the architecture starts with one principle.
The separation principle: system of record vs. system of thought.
This is the most important idea in the module. Get it wrong and everything downstream degrades. Keep hard data and opinion in different systems.
Your CRM
Structured, canonical, as close to ground truth as you can keep it: the company, the sector, the round, the numbers, the status. Changes slowly and deliberately. Meant to persist and to be trusted.
Your workspace
Opinions, working theses, draft writing, temporary scores, the make-or-break questions you are still chewing on. Meant to be provisional, rewritten, and sometimes discarded. Messiness here is a feature.
These are different kinds of data, and they belong in different places. The failure mode is convergence — writing opinions into CRM fields, or treating workspace notes as settled fact. When the two blur together, you lose the line between what is true and what someone guessed on a Tuesday.
A model reasoning over your data needs to know what is ground truth and what is provisional opinion. If facts and opinions sit in one undifferentiated store, the model cannot tell a verified figure from a hunch — and neither can you. Point at the CRM for facts and at the workspace for reasoning, and treat each on its own terms. If they have already merged, that choice is gone.
Keeping the two separate preserves data integrity and keeps your options open. The CRM stays clean enough to trust. The workspace stays free enough to think in. Your AI stack can reason over both without confusing one for the other.
Why Notion — and the alternatives.
We use Notion because it does three things at once that this layer requires: it is flexible, it is interoperable, and it is intuitive enough — once learned — to drive real adoption. Flexibility lets you model your firm's actual workflow instead of inheriting someone else's template. Interoperability lets your LLM read the corpus. Intuitiveness is what gets the team to live in it, and a workspace nobody uses captures nothing.
Flexibility matters especially for this job. Because venture work isn't a clean, repeatable sales process, you need a surface that lets you create just enough structure to keep track of what's open without making every loose thread feel like a hole or an overdue task. Notion lets you spin up a page, a database, or a checklist on demand and walk away from it without the tool nagging you — you can do real work and keep a record of it without the constant sense of dependencies and process debt that more opinionated PM tools impose.
You can sub in Asana, Trello, or another tool if you prefer it. Whatever you choose, one criterion is non-negotiable: it must connect to your LLM. A workspace your AI cannot read is a silo with extra steps.
| Tool | Strength | Weakness | Fit for this layer |
|---|---|---|---|
Notion Recommended | Flexible data model (databases + free-form pages in one place); strong API and MCP connector; captures nuance well. | Learning curve; can slow on very large/complex workspaces; discipline required to keep it organized. | Recommended. Best balance of flexibility, interoperability, and nuance capture. |
Asana Viable | Clean, structured project and task management; low learning curve. | Rigid relative to Notion; weaker as a free-form writing and wiki surface. | Viable if your need is task tracking more than shared thinking. |
Trello Viable | Dead-simple Kanban; instant to adopt. | Shallow; not built to hold long-form reasoning or rich databases. | Viable for lightweight pipelines; outgrown quickly. |
Coda / ClickUp / others Viable | Comparable doc-plus-database models. | Smaller ecosystems; verify LLM/connector support before committing. | Viable if they clear the interoperability bar. |
Decision criteria.
Evaluate any candidate against these, in roughly this order:
- 01Interoperability with your LLM.Can your AI read — and ideally write to — the workspace through an API or MCP connector? If not, stop here.
- 02Flexibility.Can you model your firm's real workflow, or are you bending your process to fit the tool's assumptions?
- 03Nuance capture.Does it hold long-form writing, opinions, and context as well as structured fields? This layer is where the soft, high-signal thinking lives.
- 04Intuitiveness and adoption.Will the team actually live in it? Account for the learning curve, but weight sustained usage heavily — an unused workspace captures nothing.
- 05Database capability.Can it run relational databases (companies, calls, competitors) alongside free-form pages, so the structure and the thinking sit together?
- 06Separation discipline.Does it make it easy to keep workspace content distinct from the CRM, rather than tempting you to recreate canonical data here?
Recommended setup — treat each company as a project.
The core move: treat every company as a “project.” Recreate your deal pipeline as a Notion database that mirrors how your CRM is structured — but use it for the extended work the CRM is not built for, where you write opinions, track diligence, and keep temporary fields that will be rewritten. Same companies, same pipeline shape, different purpose. The CRM holds the record; the Notion project holds the thinking.
The pipeline database carries a light set of canonical-looking attributes for navigation — name, status, sector, a one-line description, and a link — while the depth lives inside each company's page. Status mirrors your pipeline taxonomy (e.g., Live, Portfolio Companies).
| Company | Status | Sector | Description | URL |
|---|---|---|---|---|
| NextSilicon | Portfolio Companies | Semiconductors | Advanced computing chips with effortless acceleration and innovative architecture | nextsilicon.com |
| PsiQuantum | Portfolio Companies | Quantum | Commercially viable quantum computing for solving significant global challenges | psiquantum.com |
| Stoke Space | Portfolio Companies | Space | Fully re-usable launch | stokespace.com |
| Albedo Space | Portfolio Companies | Space | High-res satellite imagery in VLEO | albedo.com |
| Whisper Aero | Live | Energy | Ultra-quiet electric propulsion systems for efficient aviation solutions | whisperaero.com |
| Positron | Live | Semiconductors | AI infrastructure for scalable, low-cost, power-efficient inference systems | positron.ai |
Every company page opens to the same structure, so the team always knows where to look and your AI always knows where to read.
[Company Name]
- Diligence Call Tracker
- Call Notes
- Key Competitors
- Investment Process Checklist
- Due Diligence Process Checklist
- How and when did we meet the company?
- What are the company's ambitions for its next fundraise?
- What is the expected timing and valuation?
- What are the key issues that will make or break this investment?
- What resources are, or should be, involved in evaluating the potential?
- What are the next steps to push toward a decision?
The discipline that makes this work: everything on this page is thinking, not record. Opinions, drafts, and temporary fields belong here precisely because they will change. The moment something becomes a canonical fact, it goes to the CRM — not into a Notion field that an AI might later mistake for ground truth.
Adjacent databases — feeds.
The same workspace runs the firm's standing databases, not just deal pages.
New deals that fit the firm's thesis are pumped in as they happen, giving the team a live, in-thesis view of market activity in one place.
News about the companies in your universe runs through Notion, so coverage accrues alongside the deals you're tracking instead of scattering across inboxes.
Each feed is its own dedicated Notion database, separate from the CRM's canonical deal and company records — they're context and signal living alongside the system of record, not inside it.
Interoperability and the queryable corpus.
The payoff of doing all this in an interoperable tool is that the accumulated thinking stops being dead text and becomes context an AI can use. Notion exposes a public API and an MCP connector, so an LLM — Claude, for example — can read the workspace and reason over it: pull the pros and cons on a company, summarize a diligence thread, surface the open key questions across the pipeline.
Two things keep this clean:
- It stays separate from the CRM. The AI reads the workspace for reasoning and context and the CRM for facts. Because the two never merged, the model — and you — always know which is which.
- It permissions by structure. Sensitive pages and databases stay scoped, and AI access inherits those boundaries, so you get the context without overexposing it.
This is what “shared thinking accrues into a queryable corpus” means in practice: every well-kept page is one more piece of context your stack can reason over, while the canonical record stays untouched.
Downstream implications and discipline.
What this setup buys you:
Your models reason over facts and opinions separately, so answers are grounded and you can tell which is which.
Because the canonical and speculative layers never merged, you can decide later what to trust, surface, or train on. Convergence forecloses that choice; separation preserves it.
An AI-native firm captures more of what it actually believes — the nuance, the debate, the dead ends — in a place the systems can read.
Where it breaks, so you can guard against it:
When to use your collaborative workspace vs when to use a system of record like a CRM or file drive needs to be extremely clear to all team members.
A flexible workspace that the team does not maintain captures nothing and is worse than a simpler tool they would use. Budget for the learning curve and for keeping it organized.
Flexibility invites mess. The fixed deal-page template and a consistent database schema are what keep the corpus queryable instead of chaotic.
The shorthand: pick Notion (or another tool if you have a different bias), treat every company as a project, keep opinion out of the CRM and facts out of the workspace, and wire the whole thing to your LLM. Do those four things and your firm's thinking becomes a corpus instead of an archive.