
Six professionals who do not code organized their agent workspaces the same way.
Six professionals in five industries, none of whom write code, organized their agent workspaces the same way. This note describes the two-layer pattern they reached, why agents push people toward it, and what six cases can and cannot show.
Three kinds of sea snail, each working alone, built the same spiral.
Tanush Jagdish · June 2026
Once people move from chat to agents, many of them organize their work the same way. In early 2026 we watched six professionals in five industries, none of whom write code, arrive at the same two-layer workspace: one layer that tells the agent who they are and what they are working on, and one layer of reusable skills that cuts across every project.
Treat this as a field observation, not a study. Six people is a small sample, and not a random one: they came to us through our coaching program, our interview research, and one-to-one conversations, and one of them later joined Olito as a co-founder. What makes the pattern worth recording is how often it recurs, and that the major agent tools have formalized the same two layers.
#1Six people, five industries, one architecture
Six people make up the observation: a consulting partner, who later joined Olito as a co-founder; a strategy-consulting associate in Germany; a venture capital partner in Miami; a supply-chain consultant in Bangalore; the chief executive of a 60-person government-software company; and a corporate strategist who used to run online courses. Each started in a different place with a different problem. Once an agent was operating their computer rather than only answering questions, each of them built the same structure.
Partner
Strategy consultingLater joined Olito as a co-founder
Associate
Strategy consulting · GermanyPartner
Venture capital · MiamiConsultant
Supply chain · BangaloreChief executive
Government software · 60 peopleStrategist
Corporate strategy · former online-course operatoridentity and projects, down · skills, across
A field observation from coaching, interviews, and conversations in early 2026, not a study.
A week after the consulting partner began using the one agent tool his firm allowed, he sent me the most compact statement of the pattern. Because he later joined Olito, read it as an illustration rather than as independent evidence:
“I can clearly see how I really hate when BAs do this deck a certain way. I just need to create a skill for the agent to not do this anymore. And if I can just give that skill to my associates, that’ll save us so much time.”
a consulting partner one week into using agents, later an Olito co-founder
In one message he described the skill layer and its payoff: encode a correction as a reusable skill, then hand the skill to the team.
#2Two layers: identity going down, skills going across
One layer is vertical: who you are and what you are working on. At the top of the workspace sits an instruction file, usually named CLAUDE.md or AGENTS.md, that states the person’s role and standards. A venture analyst’s version might begin: “I’m a VC analyst. I review 20 deals a week. My deliverables are memos and partner updates.” Beneath it sit project folders (deal flow, portfolio, fundraising), each with its own brief and files. Before the person types a word, the agent knows whom it works for and on what.
A second layer runs horizontally: skills, meaning reusable instructions for tasks such as deck formatting, research synthesis, or meeting notes. Skills belong to no single project and apply to all of them; each is written once, used everywhere, and shared with the team. In short, the workspace instructions run down and the skills run across.
Deal flow
Portfolio
Fundraising
- Vertical layer
- Who you are and what you are working on: an instruction file, then a folder per project.
- Horizontal layer
- What you can do anywhere: skills written once, used in every project, shared with the team.
Tool makers have formalized both layers. OpenAI released AGENTS.md, an open format for giving coding agents project-level instructions, in August 2025 and later contributed it to the Linux Foundation’s Agentic AI Foundation. Anthropic introduced Agent Skills, folders of instructions and resources that an agent loads when relevant, in October 2025 and published the format as an open standard that December. Part of what people “discover,” then, is what their tools make available. What stood out in these six cases is that people with no engineering background put both primitives to work in the same way and for the same reasons.
#3Why agents force structure
Agents impose a discipline that chat never did. In chat, sloppiness is cheap: paste in whatever you have, take a decent answer, and move on. With an agent, file structure, naming, and the organization of your knowledge decide whether the agent can find and use that knowledge at all.
So people clean up. They build reusable components and start treating their work the way a well-run engineering team treats code: modular, documented, and shareable. An agent works as a forcing function for legibility. I describe it with an analogy from physics: nobody invents gravity; people run into it. None of the six write code. They described what they wanted in plain English, and the structure came out of the work itself.
#4Why non-engineers land in the same place
This architecture looks like software engineering, but the skill behind it is organizational, and these people already had it. A consultant already organizes work into workstreams, writes project charters, and keeps a standing team brief. With an agent, folder architecture turns out to be workstream organization, a README turns out to be a project charter, and the CLAUDE.md file turns out to be the team brief. Such people need translation, not new mental models.
Context architecture, in other words, is an organizational skill applied to AI. Competent operators already carry the structure in their heads; an agent makes them write it down.
#5A recurring pattern, with caveats
Coaching sessions between April and June 2026 produced three more instances, each small on its own.
On April 21, two clients at a similar stage raised the same problem in separate calls: many projects and data sources, and no way to bring them together. One described working on everything at once and feeling overwhelmed. My recommendation in the second call was the architecture in this note: a central brain with control of everything, and a knowledge base for each area beneath it. Both clients converged on the problem; the architecture was my recommendation.
In May, a client two sessions into coaching told me he wanted his own version of our agent, even if that meant running a small server, because without something like it, “it’s just going to be impossible to, frankly, operate in today’s environment.” He had reached the end-state I describe to most clients: an agent with its own computer, a brain you control, and a way to message it from anywhere. Most people get there; they differ in how fast.
In June, a client showed our community call a diagram of her personal-brain design, produced with a different AI tool. Its five columns matched the layers we were building toward in our own system: data sources, an ingestion layer, a storage layer that agents read from, reasoning agents, and outputs such as briefs and dashboards. I expect many personal-agent products to settle on some version of those five layers. That is a prediction, not a finding.
#6What this means in practice
If you have moved from chat to agents and feel an urge to organize (naming things properly, putting a brief at the top of a folder, turning a correction into something reusable), that urge is the pattern at work, not fussiness. Follow it.
If you have not made the move yet, this structure is what waits on the other side: not a better chatbot, but a way of working that organizes itself.
Our curriculum covers each layer in depth. Part 3, Skills, covers the horizontal layer; Part 4, on the personal brain, covers the vertical one.
More from our research.
Working on the same question?
We look for people to interview, data to compare against, and teams willing to test what we find. Leave a note and we will write back.