Author: @danbeksha
Compiled by: Peggy, BlockBeats
Editor's Note: AI is entering enterprises, but the real question is not "whether to use agents," but whether these agents can understand the company itself.
This article uses the author's first 100 days at Ramp as a starting point to discuss a more fundamental issue: A fast-paced company cannot rely solely on newcomers slowly reading documentation, asking colleagues, and filling in the gaps in context, nor can it allow each AI tool to operate independently. The truly crucial element is building a continuously updated "company brain" that consolidates meeting minutes, documents, Slack discussions, customer feedback, and product decisions, ensuring that both newcomers and agents can start from the same shared context.
When context is systematized, onboarding is no longer just a long adaptation process, and AI is no longer just isolated tools. The value of enterprise AI may ultimately lie not in how many agents are deployed, but in whether the company can first establish a trustworthy, readable, and reusable knowledge foundation.
The following is the original text:
In the 4x100m relay, the outcome is often not determined by the entire race, but rather compressed into a 20-meter exchange zone. Runners must complete the baton exchange at high speed: if the receiver starts too early, the baton will fall to the ground; if the receiver starts too late, the runner will have to slow down, and the entire team will instantly lose its advantage. If the exchange itself is not precise enough—any error in hand position, angle, or timing—the result can also be a dropped baton.
A team can have the fastest player on the field, yet still lose in those crucial 20 meters. Speed is important, but so is the passing of the ball. What truly determines victory is whether both can be achieved simultaneously.
Every job handover I've witnessed is essentially a relay race, except one runner is still standing at the starting line. The new employee joins on Monday, starting from scratch; but the organization doesn't slow down, continuing to move forward at its original pace. So, the new employee can only rely on reading documentation, lurking in Slack, repeatedly asking the same few questions, and spending three months figuring out how the organization operates until they finally become "useful."
We often view this gap as a matter of time, as if given enough time, newcomers will naturally catch up. But that's not the case. This gap will either be resolved by the system or it will persist.
Context is the true handover system of an organization.
I've been with Ramp for about 100 days. Before that, I worked at Plaid for five years, familiar with every product, every customer story, and the context behind every decision. I could recount those stories without hesitation. But when I came to Ramp, I knew almost nothing about any of that.
The core of product marketing is storytelling. If you don't know the characters, plot, and backstory, you can't truly tell the story well.
From day one, my goal was to build an AI-native product marketing organization. But to do this without context, I first had to expand my own knowledge base—the "context layer" that underpins all the work.
Ramp is a company known for its speed. There's no room for "catch up next quarter." The company releases, iterates, and moves forward every week. You either keep up with the pace, or you become an additional cost to the organization.
Meanwhile, I'm also going through another layer of onboarding. The pace is already fast, but AI is changing even faster, and I have to learn about a new company and a new work environment at the same time. I'm not an engineer; the last time I opened a terminal was in a university computer science class. In other words, I have to catch up on the organizational context while also adapting to the new way AI works, and these two things overlap, further amplifying the difficulty.
What ultimately freed me from this pressure wasn't completing a specific article, launching a product, or working through a particular workflow, but rather treating the "context" itself as a deliverable. Once the context layer is built correctly, all subsequent work becomes less costly.
So I started building something truly scalable: a system that could help me catch up on missed material, much like a good wiki helps researchers. By the third week, it was already drafting content based on my notes; by the eighth week, it was summarizing conferences I had missed. The learning and catching up didn't disappear, but as the system grew, their cost decreased day by day.
Personal versions of this idea have actually been around for a while. Karpathy, former head of AI at Tesla and one of the founding members of OpenAI, wrote an article in April describing what he calls a "personal LLM knowledge base": a folder storing raw input, including papers, articles, transcripts, and personal notes; an LLM that generates a wiki on top of these materials; and an editor like Obsidian as the front end. When the data accumulates to about 100 articles, the LLM can answer complex questions based on the personal corpus without requiring complex search techniques.
His assessment was that there was an opportunity to create a truly outstanding new product, rather than a bunch of hastily cobbled-together scripts.
The personal version already exists. But the company version doesn't. That's the problem.
Broadly speaking, this is the system I built in the first 100 days of my employment. They weren't very sophisticated yet, but together they formed the "connective tissue" within the organization.
At its core is an Obsidian vault, which Claude reads and writes. Conference transcripts, documents, public opinions, and personal notes that I come across all go into this knowledge base. When I ask, "What exactly did Geoff and I decide three weeks ago about the homepage?" it looks for the answer in this vault, rather than relying on the model's generalized memory.
To continuously feed content into this vault, Granola logs every meeting by default and archives transcripts at night. So, a meeting I missed on Monday was already available by Wednesday. To ensure others in the company could keep up, I chose to make my work public—most of what I was building would first appear in #team-pmm or related release channels before entering the Notion documentation. The build process itself is a synchronization mechanism.
On top of this vault, there's a small library of naming skills that agents can invoke on demand. One skill can generate agendas based on my four most recent meetings with someone; another can scan Slack for a week's worth of product activity and convert it into article topics. Each skill is approximately 200 lines of markdown, designed to replace a type of task that previously required manual work.
In addition, I built a dynamic product roadmap based on Ramp's internal application platform. It reads the same context layer, so it doesn't expire because it wasn't a static document from the beginning. There's also a morning summary sent to me via Slack private message every morning at 8 AM: what went live yesterday, where things got stuck, and what needs my response. This content is already prepared while I'm sleeping.
Individually, none of these things are particularly impressive. But put together, they provide a working answer: what would a company look like if it had the kind of wiki Karpathy described?
You can call it a wiki, a graph, a context layer, or the company brain. The name doesn't matter; functionality does. It must be able to absorb all the signals generated by the company: meetings, Slack discussions, documents, code, transcripts, customer calls, and key decisions, and keep it continuously updated without relying on manual maintenance. It must also be the first thing every new employee and every new agent reads before starting their work.
If a new employee joins the company tomorrow, what should they read on their first day? If the real answer is a Notion document from 2024, plus a broken Confluence link, then it's essentially taking over from a stagnant state.
From standalone tools to the brain of a company, the real gap in AI.
Today, the primary way AI enters enterprises still relies on forward-deployed engineers. Whether it's OpenAI, Anthropic, or large consulting firms, they all choose to build specific workflows on top of models.
These efforts are real and valuable. However, they remain in the "chatbot era" of enterprise AI: narrow tools encapsulated around specific tasks, useful on their own, but not integrated into a system that can continuously generate compound interest.
A true "company brain" has yet to emerge. Customer service agents and HR onboarding agents may have been created separately by different teams in different months. They are unaware of what decisions were made at the last all-hands meeting, how the company understands its market, or what judgments the sales manager made at the last management offsite meeting. Each agent is merely a chatbot with specific responsibilities, but they do not share a common brain.
This is the biggest gap right now. And outside the lab, there are hardly any people building products around this problem.
If you're building a team or starting a company in 2026, the workflow will be different from 2022. Write the context documentation first, then install the tools. Record every meeting. Build the wiki first, then the dashboard. Deliver skills, not slides. Get new employees reading the wiki on day one and contributing content on day two. Hire and promote those who keep the "company brain" running, and also reuse agents who truly understand and utilize the company's brain.
Context is not a side project. It is the infrastructure that makes all AI investments truly pay off.
I'm currently building a part of it on Ramp: a wiki, a skillset, applications that read information from the same context, and an organization mechanism to continuously feed it content. It's still small and early. If you're also trying to build a company-level version elsewhere, I'd love to exchange experiences. More useful than one trustworthy brain is two brains in the same room.
Returning to the relay race, the true condition for victory is not the cleanest handover, nor the fastest leg, but rather that both occur simultaneously within the same 20-meter stretch.
New employees access the company's brain and then begin their sprint. New agents access the company's brain and then begin working. New customers access the company's brain and are operational from day one.
We know we've done the right thing when the word "ramp-up" no longer makes sense.
