AI-Assisted Document Workflows via MCP — Deterministic Setup, Human-Approved
The slowest part of document automation isn’t running it — it’s building it
For most professionals who automate documents, the work that eats the day isn’t generating the paperwork. It’s the setup: taking a stack of fill-in-the-blank templates, figuring out every variable inside them, writing the intake questions a client will actually answer, mapping each answer to the right field in each document, and defining the rules that decide which documents belong in which situation.
Done carefully, that’s hours of precise, repetitive configuration for every new workflow. It’s also exactly the kind of work that has to be right, because everything downstream depends on it.
We’re building something to compress that setup step — without changing anything about how the finished workflow runs.
Can AI build a document workflow for you?
Not the way most “AI document” tools mean it. We have a firm line on this, and it isn’t moving.
The direction we’re building toward is an assistant that helps you set up a workflow — reading your uploaded templates, detecting the variables, drafting intake questions, proposing how fields map to each template, and suggesting the conditional rules for document selection. It proposes. You review. You adjust it in plain conversation until it’s right. Then you approve it.
What it will never do is decide anything at run time. The model’s job ends at the draft.
How do you keep output deterministic if an LLM is involved?
This is the part that matters most, so it’s worth being exact about it.
The principle is simple: AI proposes, a human approves, DocupletionForms executes.
The assistant only ever operates during setup, and only ever produces a draft — a proposed form, a proposed set of mappings, a proposed set of selection rules. Nothing that draft contains goes live until a person has reviewed and approved it. And once it’s approved, the workflow runs the way every DocupletionForms workflow runs today: rule-based, repeatable, and identical every time for identical inputs. The same submission always produces the same documents. There is no model in the execution path — there never was, and this doesn’t change that.
That separation is deliberate. It’s what makes the output auditable, predictable, and safe to rely on for professional work. You get help with the tedious part of building the workflow, and you keep every guarantee about how the workflow behaves once it’s built.
Who this is being built for
The setup burden falls hardest on the people who convert other professionals’ documents into reusable systems:
Legal Document Assistants who receive attorney templates and turn them into client intake packets, over and over, often with only small variations between matters. The repetitive scaffolding of each new packet is exactly what an assisted setup step is designed to shorten.
Salesforce and implementation consultants whose engagements routinely start with the same groundwork — reviewing a client’s forms, drafting intake questions, mapping merge fields, wiring up the conditional logic. Assisted setup is meant to get that first draft on the table quickly, so the consultant’s time goes to review and customization rather than initial assembly.
In both cases the value is the same: less time spent on the mechanical first draft of a workflow, with a human firmly in control of what actually ships.
Tell us about your setup workflow
What stays exactly the same
Everything you already rely on. Conditional multi-document selection, deterministic data merge, PDF completion, and the existing integration paths continue to work as they do now. This is an assisted way to build a workflow faster — a layer on top of the platform’s existing strengths, not a replacement for any of them, and not a change to how a finished workflow runs.
When will it be available?
It’s in active development. We’re being deliberate about it precisely because the deterministic guarantee is the whole point — we’d rather get the human-approval and execution boundaries exactly right than rush an assistant that blurs them.
If you build client workflows and want to hear when assisted setup is ready to try, let us know — we’re talking with a small group of document professionals as it takes shape.