On a Sunday evening not long ago, we watched a general partner prepare for Monday's partner meeting.
The agenda had four companies on it. For the first one, the partner opened the CRM record, which held two notes, the most recent five weeks old. Then a search through the inbox for the founder's name, scrolling past intro threads and scheduling back-and-forth to find the message that mattered. Then the last board deck, a PDF attachment in a different thread. Then the KPI spreadsheet, updated by hand each quarter, to check whether the revenue number in the deck matched the number the firm was reporting to LPs. Then a Slack message to an associate: what's the latest here?
Forty minutes. One company. Three more to go.
Here is the thing that struck us. None of this was research. The partner learned nothing new that evening. Every fact being assembled was something the firm already knew. The board deck had been read when it arrived. The KPI numbers had been entered by someone on the team. The associate's "latest" was a summary of calls that other people at the firm had already sat through. The partner was not acquiring understanding. The partner was reconstructing understanding the firm had already paid for, sometimes three and four times over.
We have spent months embedded inside a working venture firm, watching partners, associates, and operations leads do their jobs at close range. Once you notice this reconstruction ritual, you see it everywhere. Before partner meetings. Before board meetings. At the start of every diligence process. During every LP reporting cycle. The same understanding, rebuilt by hand, over and over.
The conclusion we came to is not the one we expected. Venture firms do not have an AI problem. Most are already using AI, and the tools are genuinely good. The problem is that the firm's understanding of its own world is not held anywhere. It lives in fragments across a dozen systems and in the heads of a few senior people, and it has to be reassembled every single time someone needs it.
The stack that solves workflows and forgets the firm
Look at the software a venture firm actually runs on. A CRM for the pipeline. Email and calendar, where most of the real work happens. Pitch decks in one folder, board decks in another. A KPI spreadsheet, or a portfolio tool if the firm has graduated to one. Meeting notes in a doc tool, or lately in an AI note-taker. Market data behind a login. Internal research in someone's drive. Slack threads commenting on all of it. And now, an AI assistant or three.
Every one of these tools earns its seat by solving a workflow. The CRM tracks deals. The portfolio tool collects KPIs. The note-taker transcribes calls. That is what they were bought to do, and most of them do it well.
But no tool in that stack is responsible for what the firm understands. Each one holds its own partial copy of the firm's world, cut to the shape of its workflow, and the copies never reconcile.
You can watch the cost of this in small, concrete moments.
A board deck arrives as a PDF. Someone re-types ARR, burn, and runway into the KPI spreadsheet. A quarter later, someone re-types the same numbers into the LP update. The same figure now exists in three places, keyed in by hand three times, and no two copies agree for long.
An associate starts diligence on a company and spends the first two days assembling context the firm already possesses: who met the founder and when, what was said, how the firm has historically thought about the market, which portfolio company sells into the same buyer. Then the firm passes. Two years later the company raises again, a new associate picks it up, and the process starts from zero. The old memo sits in a folder nobody checks. The actual reason the firm passed lives in one partner's memory, if it lives anywhere at all.
And now the same fragmentation is being rebuilt one layer up, in AI. Firms are adopting AI the way they adopted software: one workflow at a time. A note-taking bot. A deck-triage script. A drafting assistant for LP updates. Researchers at Oxford who studied AI adoption across venture firms found exactly this pattern: useful point tools, each attached to a single workflow, each seeing only its own slice. The note-taker has never heard of the pipeline. The drafting assistant knows nothing about the portfolio. Every tool is intelligent, and the firm as a whole still cannot remember what it knows.
The industry can feel this. In Affinity's 2026 survey of nearly 300 private capital professionals, the striking finding was not AI adoption. It was consolidation: firms are actively cutting their data sources down, moving from sprawling stacks of four to six sources toward one to three. Dealmakers have started to sense that the sprawl itself is the problem.
Every workflow at a venture firm reconstructs the same understanding of the world from scratch. That reconstruction is the largest hidden cost in the building, and it does not appear on any budget line.
Context is not data
It is tempting to call this a data problem and reach for the standard fixes: a warehouse, some integrations, a dashboard. That instinct misses what is actually going on, because the thing being rebuilt every Sunday evening is not data. Venture firms have plenty of data. What they lack is context.
The distinction matters. Data is the board deck, the CRM record, the row in the spreadsheet. Context is what the firm understands because of those things, connected across time. Context is knowing that this founder's projections have historically come in at about seventy percent. That the metric in this quarter's deck was quietly redefined from the metric in last quarter's. That the firm has looked at this space twice before and passed both times, and why. That a particular LP asked pointed questions about AI exposure at the last annual meeting and will ask again.
Context has three properties data does not. It is relational: facts mean something because of how they connect to other facts. It is temporal: understanding evolves, and the history of how it evolved is itself part of the understanding. And it carries judgment: not just what happened, but what the firm decided about it and what reasoning drove the decision.
Every consequential act at an investment firm runs on context. A term sheet is a conclusion drawn from context. So is a pass, a reserve allocation, a board vote, an answer to an LP's question. Two firms can hold identical data and reach opposite decisions, because what separates them is not the data. It is the accumulated, connected, judgment-laden understanding each firm brings to it.
Which raises the obvious question: where does context live today?
In people. The partners are the context layer. This is why losing a senior partner damages a firm in ways no offboarding checklist can capture, and why every associate transition quietly deletes a slice of institutional memory. The firm's most valuable asset is held in a form that cannot be queried, cannot be backed up, and goes home every evening.
The context layer
Here is the idea we think matters. Every modern investment firm will eventually run on what we have come to call a context layer: a continuously maintained representation of everything the firm knows. The companies it has met and tracked. The people and the relationships between them. The investments, the documents, the conversations, the portfolio metrics, the decisions and the reasoning behind them, and the external signals moving around all of it.
Three properties define it.
It is continuously maintained. Not a quarterly data project, not a spring cleaning of the CRM. Understanding decays in weeks. A context layer is only alive if it updates as the firm works, absorbing the board deck when it arrives and the meeting when it ends.
It is unified. One representation of the firm's world, not a dozen partial copies. The company in your pipeline, the company in your portfolio, and the company your LP asked about are the same company, and the layer knows it.
And it is consumed by everything. This is the inversion that makes it infrastructure rather than another application. Today, each workflow tool holds its own fragment of context. With a context layer, the relationship flips: workflows become interfaces over shared understanding. Diligence draws from it. Portfolio management draws from it. Investor relations draws from it. AI is simply one more interface into it, alongside the screen you work in today and the agents that come next. The interface is not the interesting part. What the interface can see is the interesting part.
Other industries have walked this exact path. Marketing teams spent a decade buying point tools, each holding its own partial copy of the customer, until the customer data platform emerged as a distinct piece of infrastructure underneath all of them. Data teams ran separate warehouses and lakes until Databricks named the lakehouse, and the lesson of that story, as their CMO later told it, is that the category followed years of conditioning the market around the pain. The pattern repeats because the underlying failure repeats: when every application maintains its own understanding, the organization ends up with none.
Fifteen years ago Marc Andreessen argued that software was eating the world. It ate venture capital in an ironic way: the industry that funded the feast ended up with the most fragmented plate. The next thing software eats is not another workflow. It is the space between the workflows, where the firm's understanding is supposed to live.
AI is what makes this urgent now. When a human assembles context by hand, the cost is invisible, absorbed into Sunday evenings and analyst hours. When you put an AI assistant in front of a partial view, the cost becomes visible instantly: the output is generic, and everyone can tell. The models are no longer the constraint. Frontier intelligence is available to every firm on earth for a few hundred dollars a month. The constraint is what you can show it. As one investment firm's head of AI put it in Affinity's report: "Having a well-defined, clean, golden source data set is more important in the age of AI."
Firms will compete on accumulated context
For the past two years, the question inside venture firms has been some version of: what is our AI strategy? That question is nearly obsolete. Eighty-five percent of private capital dealmakers already use AI daily, and adoption climbs every quarter. When everyone has AI, nobody differentiates on having it, any more than firms differentiate on having email.
So the basis of competition moves. It always does. And it moves to the one input that cannot be purchased: context.
Models are commodities. Data vendors sell the same feeds to everyone. But the understanding a firm accumulates by operating is proprietary by construction. Every diligence process, every board meeting, every pass and the reasoning behind it, every LP conversation adds a layer of sediment that belongs to that firm alone. Context cannot be downloaded. It has to be lived.
This is a compounding asset, and compounding assets reward early accumulation brutally. A firm that starts capturing its context in structured, connected, durable form today will have ten years of it in ten years. A firm that starts in five years cannot buy the missing five years at any price. The advantage looks exactly like the advantage of proprietary deal flow, because it is the same kind of advantage: accumulated, relational, and impossible to replicate quickly.
The sharpest LPs will get here first. Today they ask GPs what the firm is doing about AI. The better question, and we expect it within a few fund cycles, is: what does your firm know, and where is that knowledge held? Does it survive a partner's departure? Can your team, and the AI working alongside it, actually reason from it? A firm whose institutional memory is an org chart has a different risk profile from a firm whose memory is an asset on its books.
Firms will not compete because they have AI. Every firm will have AI. They will compete on the quality of the context their AI, and their people, can reason from.
What we believe
The firm we have been describing is CherryRock Capital, a working venture firm, and we are the team that spent those months inside it. We watched the Sunday evening ritual, the triple-keyed KPIs, the diligence that started from zero on a company the firm had known for two years. Then we started building the thing we thought should exist.
We believe investment firms do not need another AI application. They need the layer underneath: one continuously maintained understanding of the firm's companies, relationships, documents, metrics, and conversations, with every workflow, human or AI, drawing from it. That is what we are building with Opsberry, the context layer for modern investment firms. It runs CherryRock's pipeline, portfolio, and raise today.
We hold the category conviction lightly and the problem conviction firmly. Maybe the industry settles on a different name than "context layer." Names are negotiable. What we do not think is negotiable is the direction: firms that keep rebuilding their understanding by hand, workflow by workflow, will lose to firms that hold it in one place and compound it. If we are right, a decade from now nobody will call this a category. It will simply be how investment firms run.
Get a demo and we'll walk you through the fund workflow end to end.
About Opsberry
Opsberry is the context layer for modern investment firms: one continuously maintained understanding of your companies, relationships, documents, metrics, and conversations, powering diligence, investor relations, and portfolio management from a single source of context. Built inside CherryRock Capital by Enfra Inc., and backed by Y Combinator.