Blue Pencil measures every draft against your own definition of a good document and scores each rule. Your agent knows what to fix before you hit send.
“You're absolutely right, and honestly, that's the whole point.”
“Let's delve into the load-bearing assumptions.”
“It's not a delay — it's a commitment to getting it right.”
“It's worth flagging that the fix quietly landed in v2.”
“Bottom line: we pressure-tested the plan, and it holds. Full stop.”
“Would you like me to sharpen this further?”
After a year of reading AI slop, nobody sees these lines in their own drafts anymore. The client who reads your update this morning still sees them, and reads them as a sign nobody checked.
Nobody has time to read every AI draft line by line. Two in three US workers say they don't always check what AI gives them (Kolmogorov Law, a September 2026 survey of 500). So drafts go out hoping for the best.
List what a good document of your kind needs: its parts, and the rules each part follows. That list makes a draft measurable.
Blue Pencil measures your draft against every rule. You see which rules failed, and where.
Blue Pencil found what is wrong with this email, including the part you would have missed. We wrote the email ourselves, with the lines quoted above, and reviewed it against our ready-made rules for a customer notice. We underlined the words behind each rule it broke: point at or tap one to see why. The ‸ marks what is missing: the email never tells customers waiting for v2 what happens to them, or whether they need to do anything. Some will write to support; some will stop trusting you. We then rewrote the email, and it broke none of the rules. Use our rules as they are, or let Blue Pencil write first rules from a document of your own.
Subject: A quick update on v2 (An em dash.) and why it matters
Hi there,
(Agrees with the reader before it says anything.) to expect v2 on October 1, and , ("Honestly", put in to stress a point.) so did we.
Let's ("Dive into", a stock phrase.) what happened. We pressure-tested the data export, and that's where the ("Smoking gun", for a bug the testers found.) turned up: a bug that was ("Quietly", which makes a bug sound sinister.) corrupting exports. (Denies something nobody said, to make its point. And an em dash.)
(A "bottom line" label.) v2 now lands on October 15. ("Full stop", for emphasis.)
It's ("Worth flagging", before the point.) that v1 is unaffected, . ("And that matters", tacked on for weight.) Your data is safe . ("By construction", to claim it must be true.)
Nothing here says who the delay affects, or what they should do about it.
(Closes by offering to do more.)
(An em dash.)
Your AI agents run Blue Pencil on every draft, without asking you. Each review of the emails above took under a second and a half. You see only the draft that passed your rules.
Your rules: Blue Pencil writes your first rules from a document you already have.
Review: Blue Pencil measures the draft against every rule.
Rewrite: fix what failed.
At Spectacular AI, we use Blue Pencil on the notes that document our own code. We cut wrong or borderline checks from 31 of every 100 to 4 in three rounds of rewriting the rules. You can read how we did it in our case study.
314
Checks wrong or borderlineof every 100
This is how we at Spectacular AI use Blue Pencil internally: our AI agents write the notes that document our own code, and Blue Pencil measures every draft before anyone reads it.
Try Blue Pencil live on a draft you wrote with AI this week.
Paste it in, and in seconds you'll see a score for every rule on your own document.
