On-ramps: How Requests Find Their Process in an Agentic Organization
1. Summary
Most of an organization's daily work arrives as requests, and many of them have a right way to be handled, written down somewhere. The hard part is finding it at the moment it is needed, by a new person or by an agent. This note proposes on-ramps: every process lists, in its own record, the requests that start it; one index of processes is built from those lists rather than written by hand; and every agent checks the index before acting, says which process it is following, and follows it in order. Requests that match no process are kept, and when one recurs it becomes a process 12. The approach is Early: it is running in the author's own organization since October 30, 2026.
2. The research question
Can an organization of people and agents route each request to the process that handles it, and learn new processes from the requests that have none?
- Does one rule for every agent, check the processes first, keep agents doing things the organization's way as processes are added, without a new instruction per process?
- Is an index built from each process's own on-ramps more reliable than a list kept by hand?
- Do the requests that match no process show, early enough to act on, where the organization works without one?
3. Background and related work
Process modeling has long named the trigger that starts a process its start event 3. Organizations also write standard operating procedures and playbooks, but these are found by people who already know they exist. Agents change both sides: they can check every request against every process, and they can notice when a request matches none.
Ikujiro Nonaka and Hirotaka Takeuchi described how organizations learn by turning what people know how to do into something written that others can use, which they called externalization 4. On-ramps are a way to make that step routine: the gap is noticed at the moment of the request, by the agent handling it.
4. Setting and method
The approach. Each process in the organization's knowledge lists its on-ramps in its header, for example on_ramps: ["add an organizer", "a new organizer joins"]. After every change to the knowledge, a program rebuilds one page, the process index, from those headers, at each access level, so the index cannot drift from the processes. Every agent has one rule: before acting on a request, check the index; if a request matches, say which process it is following and follow it in order; if nothing matches and the same kind of request has come before, propose a process to its person and add the request to a short list of gaps 12.
What we watch. How often requests match a process; how often a matched process was followed in order; how long a gap stays on the list before it becomes a process; and whether processes with on-ramps are kept more current than those without.
5. What we have observed so far
The approach is new. On its first day, seven of the organization's guides were given on-ramps, the index was built from them and rebuilt on its own when a guide changed, and five gaps were listed: requests that came up often with no process. Building the index also surfaced one guide that contradicted a recent decision, which was rewritten before it was given on-ramps. There are no findings yet on how agents follow it over time.
6. What would change our mind
What would count against the approach. If agents match requests to the wrong process often; if people stop reading processes because their agent follows them; or if the gap list grows without becoming processes.
Open questions. How specific should an on-ramp be? Who owns a process an agent proposes? Should an index ever be shared between organizations, so that agents moving between them find familiar processes?
7. Ethics and privacy
The index lists processes, never people's requests: what a person asked their agent stays private to them, and only the fact that a kind of request has no process is recorded. A process proposed by an agent becomes one only when a person approves it. This note names no one but the author and describes no organization's keys, accounts, or systems.
8. How it reaches students
Students who build agents in Building AI Agents write instructions for one agent; on-ramps are the next question for them: how an organization's many agents find the right instructions for a request, and how new ones are written from experience.
9. What this means for your organization
List the requests that come up every week where you work. For each, is there a written way to handle it, and how would a new person, or an agent, find it?
10. References
- Mike Lee, Building Agentic Organizations: How to Build an Organization of People and Agents (GSSF Press, 2026), chapter 8, "From Strategy to Execution". https://buildingagenticorganizations.org/book/ch08
- Mike Lee, Building Agentic Organizations: How to Build an Organization of People and Agents (GSSF Press, 2026), chapter 17, "How the Organization Improves Itself". https://buildingagenticorganizations.org/book/ch17
- Object Management Group, Business Process Model and Notation (BPMN), Version 2.0 (Object Management Group, 2011), section 10.5.2, "Start Event." https://www.omg.org/spec/BPMN/2.0/
- Ikujiro Nonaka and Hirotaka Takeuchi, The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation (Oxford University Press, 1995).
marks our own words: our view, our summaries, and our book. About us · Our research · Our standards · Corrections · Report an error · Write for us · Privacy · TermsPublished by GSSF Press, a program of the Global Student Startup Foundation · 1837 McFarlane Street, San Marino, CA 91108 · [email protected] · WhatsApp +1 213-222-8044