I asked for a document in one message. It shipped branded, sign-ready paperwork.
The agent pulled my company logo off my website, built a proper letterhead, asked the three questions that actually changed the wording, and delivered editable Google Docs plus sign-ready PDFs — then added a matching second document to the batch on the same letterhead.
Paperwork is a recurring tax.
Verifications, letters, forms — all needing the same logo, the same letterhead, the same signature block.
Low-skill, high-friction, and easy to get subtly wrong.
Running a company means constantly generating official-looking documents from the same brand kit. By hand that means finding a template, dropping in the logo, matching the letterhead, getting the wording right, exporting to PDF — every time.
A small wording mismatch on an official document is a real problem — the difference between “has been employed” and “was employed” matters on something headed for a visa or loan application. I wanted to turn a vague “I need this document” into a finished, branded, correctly-worded, sign-ready file without doing any of the assembly myself.
Specify → clarify → produce → deliver.
The clarify step is what separates a useful document agent from a mail-merge.
I gave the agent the intent and the facts in plain language. It pulled the brand assets from my live website and business listing to build the letterhead — rather than asking me to supply them. Then it did the important thing: it asked the disambiguating questions that change the output — signatory title, address inclusion, an optional identity field, and current-versus-departed employment, which flips the tense and the whole framing. Only after I answered did it generate.
It flagged that end-of-employment changes the wording — “has been employed” versus “was employed” — before it generated anything. That’s the judgment a naïve generator skips.
— Proof moment · clarify before generateThe document workflow.
A plain-language request triggers a brand-asset pull, then a round of clarifying questions, then a Google Doc build and PDF export — delivered as an editable Doc plus a sign-ready PDF, per document, for a two-document batch.
A plain-language request reaches the agent, which auto-sources brand assets from the live website and business listing, then routes through a clarify gate — the four questions that change the wording — before building. Only after answers does it generate the Google Doc letterhead, export a PDF, and deliver both formats for each of the two documents on a shared letterhead.
Agent Runtime
Brand Sourcing
Document Build
Output Formats
Delivery
Judgment Layer
One message. Four artifacts.
Mar 10, 2026 · single session · verified from the session record.
The four questions it asked first.
A mail-merge fills a template. This asked what actually changes the document — then generated once it knew.
What signatory title should sign?
Sets the authority line in the signature block.
Include a physical address?
Affects the letterhead and formality of the document.
Include an identity / reference field?
Optional field, only added if the document’s purpose needs it.
Current or departed — “has been employed” vs “was employed”?
Flips the tense and the entire framing. Getting this wrong breaks the document for its real-world use.
One chat message → four finished artifacts.
Correct, on-brand, sign-ready — not a PDF generator.
Honest scope: this was a single ad-hoc session, not a productized document service.
What it produced: two documents on a real company letterhead, each delivered as an editable Google Doc and a sign-ready PDF, generated from a plain-language request. The agent self-served the branding, asked exactly the questions that affect correctness, and returned polished, correctly-worded, on-brand paperwork ready for electronic signature.
Its portfolio value is the judgment layer — it didn’t just fill a template, it disambiguated the wording a naïve generator would get wrong. Present it as a demonstration of capability, not a standing service.
I was the spec and the approval gate.
I specified the documents in natural language and provided the facts. The agent did everything else — sourcing branding, building the letterhead, and knowing which questions to ask before generating.
I made the content decisions it surfaced and approved the output. The agent was the production engine; I owned the spec and the sign-off. The interesting part isn’t that it made a PDF — it’s that it refused to guess at the wording that mattered.