Resources
Whitepapers and technical briefs from Prompt One.
Every enterprise has line-of-business teams that run the same motions over and over: lead follow-up, contract renewals, warranty analysis, account health. Those teams almost never have engineering resources to automate their work. Prompt One is the platform built for them. Describe the work in plain language, and the Composer turns it into a Motion (a named, versioned definition of that work) run end to end by a workflow agent.
At the core of every agent is its workspace: a live, two-way copy of everything the Motion touches, unified into one canonical vocabulary, serving as the workflow's state, the context for every inference, and the agent's memory across runs. The team manages and works the whole task from one live view. Questions go to Ask Chat and come back answered from live records, as text, charts, tables, and even forms the team can act on in place. Consequential changes wait at an approval gate for a person to decide, and once the Motion is published, teammates subscribe in one click.
Under the hood these are compiled workflow agents, where the model's intelligence is applied at two distinct stages. First, at authoring time, AI generates and validates the code implementing each step, once; then, at runtime, the workflow executes as ordinary, deterministic code, invoking the model only where judgment is genuinely required. Important benefits accrue from that split.
Security: an agent can only speak the sanctioned nouns and verbs of the Service CLI, and every action crosses the Service Gateway under the acting user's SSO identity, so existing enterprise permissions apply. Every run is replayable, and because the workflow is fixed code, untrusted data cannot become a prompt injection.
Economics: intelligence is purchased once, not rented on every run, so a unit of work costs zero or only the intended LLM calls. In a measured comparison of the same agent built both ways, the compiled version cost roughly 11 times less per run, and on a day with no new work, the compiled agent spends zero tokens.
Operations: beyond automation that finally works where past initiatives stalled, the team that owns the work builds and runs it, completely self-serve, with no engineering resources required.
Prompt One Motions, executing as compiled workflow agents, improve the security, economics, and operations of enterprise AI. This paper makes the case. Evaluate Prompt One on one of your own motions and see for yourself.
- 1. The job to be done
- 2. Compiled workflow agents
- 3. Motion lifecycle
- 4. The agent workspace - fresh, live context
- 5. Agent capabilities
- 6. Security by construction
- 7. Benefits - security, economics, operations
- 8. Conclusion
Every business has teams that run the same motions over and over — lead follow-up, contract renewals, warranty analysis, account health — and those teams almost never get engineering time. Prompt One is the platform built for them. You describe the work in plain language; the Composer turns it into a Motion, a named, versioned definition of that work; and a compiled workflow agent runs it, end to end. The agent reaches into every connected system and builds its own agent workspace — a continuously synced, live copy of everything the Motion touches. The workspace is where everything lives, and it is what makes a long-running agent possible: it holds the workflow's state, supplies fresh context for every inference, and serves as the agent's long-term memory across runs. It drives the live view the team works in and grounds every answer in Ask Chat — and once a Motion is published, it spreads across the team by subscription.
A compiled workflow agent uses AI to generate the code implementing each step of a workflow once, at authoring time; the workflow then runs as deterministic code, invoking the model only where judgment is genuinely required, as an explicit, bounded step. The companion paper, Choosing an Enterprise Agent Architecture, makes the general case: a decision framework for matching agent architectures to enterprise tasks, showing why recurring, structured, cross-system work profiles toward this architecture — knowable steps, concentrated judgment, unattended cadence, untrusted input, audit obligations, and a cost curve that punishes per-run inference. In that paper's measured comparison of the same agent built both ways, the compiled version cost roughly 11 times less per run.
This paper describes the implementation in the order a team meets it: the Motion you build (§1), the Composer that codifies it (§2), the agent workspace (§3), the Service Gateway's canonical schema (§4), the deployed surfaces — live view, Ask Chat, approval gates (§5), compiled security (§6), and subscription (§7). A summary and an appendix on the execution graph close the paper; limitations are stated where they arise.
- 1. Motions: what you build on Prompt One
- 2. The Composer: the LLM as compiler, not the runtime
- 3. The Service Gateway: one canonical schema over every system
- 4. The Prompt One CLI: universal service abstraction
- 5. The Admin Console
- 6. The Agent Runtime: deployment and execution
- 7. Summary