Patch notes

What changed on SecondPen.ai, newest first. The build number in the editor's footer links here.

1.3.32 (2026-10-09)

• Find marks its matches in the document. It counted them but did not show where they were. Every match is now marked, the current one more strongly, and the first match scrolls into view as you type.
• Typing in Markdown view no longer throws the page away from the cursor. In Safari, each keystroke could jump the view towards the top of a long document.

1.3.31 (2026-09-27)

• When the Co-author writes the whole document, say as a first pass at a template, the Editor and Critic read it straight away, and the thread says so. Before, they could sit silent: they wait for an abstract, and a first pass does not come with one. Now one is drafted from the document for you to keep or edit. Blanks the AI leaves for facts it cannot know, such as [figure] or [date], no longer hold the line edits back. The Editor lists them above its cards as facts to fill in, each a click away from its place, and reviews everything around them.

  The pilot runs for [number] weeks with our two largest accounts.

  Repeat questions fell by [figure]% in the first month of the shared inbox.

  Saved replies start from the forty answers support sends every week.

  Co-authorEditorCritic

  Fill in the whole documentReworked the document from your ask. The Editor and Critic are reading it.
• When the Co-author answers a comment with a better version of a paragraph already in the document, its first button is now Replace paragraph. Hover it to see which paragraph it means; click it and the new wording takes that paragraph's place, with Undo to bring the old one back. Before, Insert put the rewrite beside the original, so the paragraph appeared twice.

  The pilot starts in March with the two largest accounts.

  Support answers the same forty questions every week, and each answer is typed again from scratch by whoever picks up the ticket first.

  Saved replies start from those forty answers.

  Co-authorEditorCritic

  The second paragraph is clunkySupport answers the same forty questions every week, and whoever picks up the ticket types each answer again.Replace paragraphInsertCopy

1.3.30 (2026-09-27)

• Blank templates give fuller guidance. Each section now carries a short brief instead of a single line: what the section is for and how long it should be, what good looks like, an example where one helps, and the mistake writers most often make there. The brief is written from the same criteria the draft is scored on, so following the template is also writing to the score. The guidance still sits in quoted lines you delete as you fill each section in.

1.3.29 (2026-09-27)

• Blank templates are written to the same standard as everything else the AI writes. Their guidance lines and examples are held to the Editor's checks (clear on a first read, concrete, short, free of mistakes), and a template that comes back with a stock phrase of machine-written text goes back once to have it taken out. The templates are being rewritten this way; each one you use stays as it is until its new version is ready.

1.3.28 (2026-09-27)

• The Editor proposes a change only where a paragraph needs one. Every paragraph is checked for the same things: is it clear on a first read, concrete, supported, as short as it can be, true to the abstract and to how you write, and free of mistakes. A line edit now has to fix one of those, and each card names which. A paragraph that passes gets no suggestions, however recently it was edited; one that still falls short can get the next fix straight after you apply the first. The Editor's header counts the paragraphs that pass, and the draft counts as ready only when the whole reads well and no paragraph fails a check an editor would insist on. The Editor also reads across paragraphs now, for a point made twice, a missing transition or a paragraph in the wrong place.

  Support answers the same forty questions every week, and each answer is typed again from scratch.

  In order to really see whether it works, we would basically like to pilot the inbox with our two largest accounts.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditorCritic

  Reading for line edits...
• What the AI writes is checked before it lands. Every request tells the model the same checks the Editor uses, and what comes back is checked for them and for the stock phrases of machine-written text ("delve", "a testament to", "it's not X, it's Y") and your words to avoid. If anything fails, the draft goes back once with the problem named, while the working label reads "Checking the draft...". That second pass is not counted against your daily edits. Writing at the cursor and reworks also see the document's outline, the text around the place they write in a long document, and the figures it states elsewhere, so new text does not contradict them. A demanding ask on a selection goes to the stronger model the first time. If an edit may have changed your meaning, the note under the document offers Try harder as well as Undo.

  We pilot the shared inbox with our two largest accounts for six weeks.

  /a sentence on how we measure it

1.3.27 (2026-09-27)

• The Editor and Critic show when they are working. While either reviews the draft, checks whether it is ready, or scores it, a small spinner turns in its tab, whichever tab you have open, and stops when its cards arrive.

  Support answers the same forty questions every week. One shared inbox, with saved replies the whole team can edit, fixes that.

  We pilot it with our two largest accounts for six weeks.

  Co-authorEditorCritic

  Suggest a titleOne inbox for the forty questions
• An ask on selected text no longer disappears when it reads like a question. "Recommend a change" or "can you make this punchier?" rewrites the selection, as you meant. A real question, such as "what does this mean?", is answered in the Co-author with the passage quoted, and the note under the document offers to rewrite the selection instead. Before, the box closed and a brief notice was all that happened.

  The valve is closed by turning the handle a quarter turn.what does a quarter turn mean here

  Second paragraph stays.

  Co-authorEditorCritic

1.3.26 (2026-09-26)

• Asking the Co-author to change the document no longer answers with a red line that reads like an error. The thread shows a quiet line with a spinner while the document is reworked, then says how it went: reworked, stopped, or not finished with the document unchanged. While the change is the last one, the line carries its own Undo.

  Support answers the same forty questions every week, and each answer lives in whichever inbox happened to receive it, which means nobody can see the whole picture.

  It is therefore proposed that a pilot be conducted with the two largest accounts before any wider rollout is considered.

  Co-authorEditorCritic

  Make it shorter and warmer.
• Template search finds what you mean. It ignores accents and hyphens, so "resume" finds the résumé formats and "post-mortem" the postmortem, and every format also answers to the everyday words for it: "cv", "incident report", "retro", "one pager", "cold email".

1.3.25 (2026-09-26)

• Scoring failed with "The model did not return valid JSON" after 1.3.24 added line edits to the scorecard: the longer reply was cut off at the output limit. The limit is raised to fit, the edits are asked for in short form, and a reply that is still cut off now says so instead of blaming the JSON.
• Text size. The document's type now starts at 16px, down from 17.5, and two A buttons in the footer, or the Text size row in Settings, step it through five sizes from 85 to 120 percent. The choice is kept with this browser, so each screen has its own. Export and print are unchanged.

  One shared inbox, with saved replies the whole team can edit.

  We would pilot it with the two largest accounts before opening it to everyone, and decide on the numbers after six weeks.

  Saved212 wordsAAAI: ready

1.3.24 (2026-09-26)

• Review fixes. A long document reviewed from its flagged paragraphs no longer counts the rest as read, so the next pass reads them instead of saying nothing changed. Clicking a margin flag reviews that paragraph and keeps the cards you already have. A scorecard fix whose sentence is not in the draft is dropped instead of shown as a card that points nowhere. The same note twice in one answer is one card. A pass over the whole draft replaces the cards it had there, while scorecard fixes stay until applied. When the judgement model gives no answer on a style or figure check, the sentence is asked again after a minute rather than marked clean.
• You are not asked to edit the same sentence twice. A paragraph you have just acted on, by applying a card or accepting a margin edit, is settled: the review does not read it again and no card is shown on it until you change it substantially or ask for everything to be reviewed again. A second note about a sentence that already has a card joins that card as an extra line instead of standing beside it, and the scorecard's fix on a sentence takes the lead over a note. Dismissing a card keeps that kind of note off the sentence while it stands. A margin flag steps back from a paragraph that already has a card.

  The pilot is kind of a big deal for us. It runs for six weeks with the two largest accounts.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditor1Critic

  1 thing between this and ready

  VagueSay what the pilot decides.is kind of a big deal for usdecides whether support gets the inboxApplyDismiss
• The automatic review waits for the draft. While a document is being written, the Editor's line edits stay quiet, and the Critic's questions wait until the draft has taken shape. With a template that means how many of its sections carry text and whether any placeholders are left, and the tab says so: "2 of 6 sections filled". Without a template it goes by how fast words are arriving and the judgement model's last readiness reading. Margin flags while drafting show only a paragraph that does not serve the abstract. Review now and Ask now run at any stage.

  Launch plan

  Goal

  Two hundred teams on the shared inbox by March.

  Audience

  Support leads at accounts with more than ten agents.

  Timeline

  [Dates, from beta to launch.]

  Co-authorEditorCritic

  Waiting for the draftReview now

  Waiting until the draft is further along · 2 of 6 sections filled, 4 placeholders left. Review now runs it anyway.
• Scoring a draft is now the review pass. The scorecard brings its fixes and up to six line edits from the same read, all as Editor cards, and the draft counts as reviewed, so the standing list does not read it again until something changes.
  Documents+ NewBlameless postmortemProposal: a shared inbox

  [Incident title]

  Blameless: name systems, not people. Two pages.

  Summary

  [Two sentences: what happened, and the customer impact.]

  Timeline

  [Dated entries, detection to recovery.]

  Co-authorEditorCritic

  The format is in place. Fill it in, then score the draft against it from the Critic tab.
• Asking the co-author to change the document reworks it in place, with the same check and one undo as any rework, instead of replying with text to paste over the document.

  Support answers the same forty questions every week, and each answer lives in whichever inbox happened to receive it, which means nobody can see the whole picture.

  It is therefore proposed that a pilot be conducted with the two largest accounts before any wider rollout is considered.

  Co-authorEditorCritic

  Make it shorter and warmer.
• Your voice has a "How you write" group: spelling (American, British, Canadian or Australian), contractions, the serial comma, first or third person, whether the reader is addressed as you, a ceiling on sentence length, which punctuation to avoid (em dashes, semicolons, parentheses) and a list of words to avoid. It sits collapsed under a one-line summary with an Edit button, so it stays out of the way of teaching your voice. The choices travel with every AI request as rules beside your voice profile, the profile is told not to contradict them, and the ones that need no model are enforced in the editor: an em dash that slips through is replaced, a sentence over your ceiling and a word you avoid are underlined. The em dash switch moved here from Settings, and the Writing style menu is gone from Settings: a standard to write to or a register to keep goes in the group's "Anything else" note, and a custom style instruction you had set is carried over into it.
  Your voiceBack to the documentHow you writeEditspelling as written · no em dashesGoes with every AI request, beside your voice profile.

1.3.23 (2026-09-25)

• Editor cards no longer invent facts. The line-edit review and the scorecard's fixes are told that a rewrite may only use what the draft already says, and never add a date, figure, name, place or source; where one is missing they leave a placeholder such as [date]. As a backstop, a rewrite that still brings in a date the draft does not mention shows [date] in its place, and the card says the writer must supply it.
• Critic cards retire as the draft moves on. A card whose passage you rewrote greys out while you type and leaves at the next pass, a pass over changed paragraphs replaces the cards it had there rather than adding to them, a card re-pointed once is not re-pointed again, and each list keeps at most twelve edits or eight questions, oldest out first. Before this, cards from earlier passes stayed as the document changed and a long session could pile up dozens.

  One shared inbox, with saved replies the whole team can edit.

  We pilot it for a quarter and then decide.

  Co-authorEditorCritic1

  1 open question

  QuestionDecide on what? A reader cannot tell what would make the pilot a success.We pilot it for a quarter and then decide.Show meAddressed

1.3.22 (2026-09-25)

• Score the draft and templates now use your key. They had been sent without it, so they were refused once the included allowance was spent while rewrites kept working, and the refusal notice lost its detail. Both are fixed.

1.3.21 (2026-09-25)

• When the included allowance refuses a request although your key is in use, the notice now says what the server saw: no key header at all (something between your browser and the site is removing it), or a key whose shape it did not recognise. Keys are also accepted in any printable form after their prefix, so a newer key format cannot be turned away.
• When Anthropic or OpenRouter turns down a key of yours, the notice now says so and repeats the provider's reason, instead of reading as if the AI were not set up on the server. A key whose account has run out of credits says that too.

1.3.20 (2026-09-25)

• A saved key is used the moment the included allowance runs out. If a request is refused for the day's allowance while a key of yours is saved but not in use, the editor switches to the key, tells you, and sends the request again, with no click needed. When the allowance does refuse a request, the notice now says what it found: a key in use that somehow was not sent, a key set to be forgotten when the browser closes, or no key at all.

  Support answers the same forty questions every week.

  The idea is pretty simple, basically, when you think about it.

  Today’s included allowance is used up, so writing switched to your Anthropic key.

1.3.19 (2026-09-25)

• A sentence from your voice profile could turn up in an edit. 1.3.18 told the AI to use quoted wording exactly, and the prompt quotes sentences of yours for tone; the AI sometimes took one of those as wording to insert. The rule now applies only to wording inside your instruction, and everything quoted for tone is marked as reference that must never be copied.

1.3.18 (2026-09-25)

• Scoring a draft now shows what it is doing. The score box carries a live line with the seconds while the strongest model reads the whole draft, the button says Scoring, and if it fails the reason stays in the box with a Try again. Switching documents ends a score in progress instead of leaving the button stuck.

  The shared inbox is here

  Starting today, every team on the Pro plan can answer customers from one inbox, with saved replies anyone on the team can edit.

  Turn it on under Settings, Inbox.

  Co-authorEditorCritic

  Format: Launch announcementScore this draft
• Rewriting part of a sentence no longer leaves the rest of it hanging. An instruction on a half sentence widens to the whole sentence, with or without the judgement model's help, and the AI is told that its replacement must read as one with the words on either side and that wording you supply is used as written.

1.3.17 (2026-09-25)

• A saved key that is not in use no longer looks like a missing one. If Settings is on the included allowance and the allowance runs out, the dialog now says your key is saved but not being sent, and one click switches to it. The error notice says the same, and the Settings list shows which one is in use right after you save, remove or switch a key.

  One shared inbox, with saved replies the whole team can edit.

  We would pilot it with the two largest accounts.

  Saved212 wordsAI: allowance used up

1.3.16 (2026-09-25)

• Try harder. After an edit lands, a nudge offers to run the same edit again on the strongest model over the passage it produced. Everyday edits are faster, and the scorecard, your voice profile and very long reworks stay on the strongest model.

  Support answers the same forty questions every week.

  Each answer gets saved so agents can reuse it later on.

  Applied. Not quite right? A stronger model can take another pass at the same passage.Try harder

1.3.15 (2026-09-25)

• The blog is off the site for now, while the writing gets the attention it deserves. The templates pages stay.

1.3.14 (2026-09-25)

• Every document format has a page of its own at secondpen.ai/templates: its sections, where it comes from, the criteria a tough reader scores it on, and a link that opens a new document from it in the editor. The blog is live at secondpen.ai/blog.

1.3.13 (2026-09-25)

• The patch notes show sharing by link in a silent scene: the Share button, the link made and copied, and a reader's comment arriving as a card.

1.3.12 (2026-09-25)

• Blank templates arrive faster, in the same shape. As before, the first person to open a format has it written once and everyone after gets it instantly.

1.3.11 (2026-09-25)

• The editor works on a phone and a tablet. Below 980 pixels the document has the screen to itself: the document list slides in from the left and the AI panel from the right, over a scrim that puts them away, and the abstract starts collapsed. On a phone the toolbar scrolls sideways, the AI panel takes the whole width, and the buttons are big enough for a thumb. Select text with your finger and the bubble waits for the handles to settle, then sits below the selection, out of their way.
  Proposal: a shared inbox

  One shared inbox, with saved replies the whole team can edit.

  We would pilot it with the two largest accounts before opening it to everyone.

  Your voice2 pieces readDocumentsProposal: a shared inboxNotes from TuesdayLaunch plan

  Co-authorEditorCritic

  Suggest a titleOne inbox for the forty questions
• Share a document by link. The new Share button in the header makes a read-only page of the document that follows your edits, for anyone who has the link. Readers can leave comments, quoting a passage by selecting it, and those comments arrive in the editor as cards at the top of Critic: Show me jumps to the passage, Addressed marks it answered on the page. Turn comments off, or stop sharing, whenever you like. Sharing needs an account, so the link stays yours to stop.
  SecondPen.aiField notes: the overflow valveD

  The valve closes when the float reaches the top of the chamber. A second float, set lower, catches the overflow if the first one sticks.

  Run 2 was after the gasket swap, which explains the smaller leak.

• Templates no longer put words in your mouth. A blank template used to fill an author line, a byline, a customer quote or a thank-you with a real-sounding name; now anything that would read as a fact about you or your company is a bracketed placeholder such as [Author name and role]. Worked examples stay, labelled as examples with plainly made-up names. Templates already written have been regenerated.

1.3.10 (2026-09-25)

• Ten formats for writing that strangers will read, in a new Writing & publishing category: the engineering blog post as Stripe and Netflix write it, the essay in Paul Graham's manner, the op-ed as The OpEd Project teaches it, the listicle, the tutorial and the how-to guide from the Diátaxis framework, the long-form explainer, the launch announcement, the newsletter issue, and HubSpot's SEO pillar page. Each scores a draft on whether its first line earns the second.

1.3.9 (2026-09-25)

• The AI panel scrolls properly. The Editor tab clipped its cards once they outgrew the panel, with no way to reach the rest; it now scrolls like the Critic tab. And the Co-author thread no longer snaps to the bottom on every word of a reply: scroll up to read and it stays put until you return to the bottom or send another message.

1.3.8 (2026-09-25)

• Ten formats for the candidate's side of a job search, in a new Careers & job search category: the reverse-chronological résumé as Harvard's career office teaches it, Google's XYZ résumé bullets, the two-page executive résumé, the academic CV, the federal résumé that USAJOBS requires, the Europass CV, the cover letter, a LinkedIn headline and About, the professional bio in three lengths, and the brag document that feeds all of them. Each scores a draft the way a recruiter's six-second read would.
• "Fill out the document" typed anywhere in a template now fills the whole template, replacing its guidance, instead of adding a filled copy at the cursor. The slash command asks how much of the document an ask means, and "fill out", "complete" and "populate" count as reworking what is there.

  Pilot plan

  • Two largest accounts
  • Saved replies from the top forty answers
  • Decide after a quarter

• A slash ask shows that it is running. From the moment you press Enter a dashed line under the cursor says what is being read or written, the working bubble with Cancel sits beside it, and a rework pulses the text it is about to replace. Before, the bubble could sit somewhere unrelated and the document gave no sign for several seconds.

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

1.3.7 (2026-09-25)

• Running out of the day's included AI allowance no longer leaves you with a toast. The first time it happens a dialog explains what is used up and when it comes back, and offers the way on: add your own Anthropic or OpenRouter key, which takes you straight to the key field, or wait for the reset. A bar stays at the top of the AI panel with the same choices until the allowance is back or a key is in. Signed-out visitors at the end of the free trial get the same, with sign-in as the first option.

  One shared inbox, with saved replies the whole team can edit.

  We would pilot it with the two largest accounts.

  Co-authorEditorCritic

  Suggest a titleThinking...

1.3.6 (2026-09-24)

• A catalog of 112 document formats behind "Browse all templates" in the New menu: six-pagers, PR/FAQs, OKRs, design docs, postmortems, decision memos, pitch decks, research plans, role scorecards, project charters, personal reviews and sixteen kinds of email, each with its origin and its sections. A format opens as a blank template the AI writes once, with the guidance in place for you to fill in and delete.
  Documents+ NewUntitledProposal: a shared inbox

  Start writing, or press + New for a format.

  Co-authorEditorCritic

  Ask about this document, or say what to write.
• Score a draft against its format. The Critic tab now knows which format a document follows and scores it the way that format's toughest reader would: each criterion out of ten, weighted to 100, with a verdict. The three highest-leverage fixes arrive as Editor cards you can apply, and the first question a sceptical reader would ask joins the Critic's list. Any document can be given a format from the Critic tab.
  Documents+ NewBlameless postmortemProposal: a shared inbox

  [Incident title]

  Blameless: name systems, not people. Two pages.

  Summary

  [Two sentences: what happened, and the customer impact.]

  Timeline

  [Dated entries, detection to recovery.]

  Co-authorEditorCritic

  The format is in place. Fill it in, then score the draft against it from the Critic tab.
• The formats find you. Write an abstract that reads like a known format and the editor names it and offers it: to start from, when the document is still empty, or to score against once there is a draft. Ask the Co-author or the slash command to write a whole document on an empty page and, when the ask reads like a format, the editor offers to start from the format and fill it in from your ask. Say no and it writes as before. A draft with no abstract is read the same way once it has a few paragraphs.
  AbstractWhy logins failed on Tuesday, for the engineering team: what happened, the timeline, and what we change so it does not happen again.

  Start writing, or press + New for a format.

• The New menu is Blank, the formats you used most recently, and the catalog. The four hand-written blanks it used to carry are gone; the catalog has fuller versions of each.
  Documents+ NewUntitledProposal: a shared inbox

  Start writing, or press + New for a format.

  Co-authorEditorCritic

  Ask about this document, or say what to write.

1.3.5 (2026-09-24)

• The slash command reworks what is there when that is what you ask. "Fill in the details" typed under an outline now replaces the outline with the filled-in version, instead of adding a second, longer copy below it. With the cursor at the end of the document the whole document is reworked; inside it, the section around the cursor is. New text is still added at the cursor when you ask for something new, and Undo puts the original back.

  Pilot plan

  • Two largest accounts
  • Saved replies from the top forty answers
  • Decide after a quarter

• The Editor's and the Critic's cards stay with their document. Switch to another document and back, or come back tomorrow, and the cards you had not dealt with are still there, pointed at the same passages.
• Every document opens on the Co-author tab.

  There are some really important reasons to do this. We pilot it for a quarter and then decide.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditorCritic1

  Re-reading what changed...Review now

  Line edits you can apply, kept current as you write.
• Your voice has its place on the home page, and its patch note above now plays a short demo.

1.3.4 (2026-09-24)

• Teach the AI your voice from your own writing. Your voice sits at the top of the left rail and in the account menu, apart from any one document. Paste something you wrote, bring it in from a link or a file, and submit it: the AI reads it once and updates your voice profile, and sentences quoted from it travel with every request. A link is read on our side and only its writing is kept, not the menus and footers around it; a Google Doc shared with anyone who has the link works too. Your voice is kept with your account, so it asks you to sign in.
  Your voicePaste something you wrote to startDocumentsProposal: a shared inboxNotes from Tuesday

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

  Support cannot see which questions repeat.

• The patch notes show a short silent demo under a new feature that changes what you see, starting with this page's own back catalogue.

1.3.3 (2026-09-24)

• Sign in goes straight to Google. The sign-in page now appears only when there is something to say first: an invite code to enter, closed sign-ups, or a sign-in that did not complete.

1.3.2 (2026-09-24)

• The settings gear is gone from the top bar. Settings lives in the account menu, which now opens whether or not you are signed in, and the menu offers Sign in with Google when you are not.
  SecondPen.aiProposal: a shared inboxReport a bugD

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

• Sign out is one click. No confirmation page in between.
• Try the AI before signing in. A signed-out visitor gets a few free edits a day on the included key, and the footer shows how many are left. When they run out, the editor says so and offers sign in. A visitor with their own Anthropic or OpenRouter key can use it without an account.

  Support answers the same forty questions every week.

  It's answers lives in seperate inboxes.

1.3.1 (2026-09-24)

• Fixed: "Replace document" and "Insert" from the Co-author could swap block styles when the document started with a heading, turning the reply's first paragraph into a heading. Headings, paragraphs and lists now land as written. Ctrl+Z, or the Undo button after a replace, puts the document back.
• Fixed: when you are not signed in, Send in the Co-author could do nothing. It now says to sign in, with a link, every time. The same goes for the selection box, the slash command and the continue offer.

1.3.0 (2026-09-23)

• One slash command instead of three. Type / on an empty line and say what you want: a request for prose is written at the cursor, a question or a summary is answered by the Co-author, and Enter on the bare slash continues from where you are. A slash mid-sentence, in a path or in a code block stays a slash.

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

• A Report a bug button sits in the top bar, labelled and easy to find, in place of the small speech-bubble icon. The button that opens and closes the AI panel now shows a sparkles mark.
  SecondPen.aiProposal: a shared inboxReport a bugD

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

• Fixed: quick edits, including Fix and short instructions typed in the selection box, could fail with "Something went wrong reaching the AI." The request asked the faster model for a setting it does not accept. If the faster model ever declines a request again, the edit now retries on the main model instead of failing.

1.2.2 (2026-09-23)

• The Critic now knows when to stop as well as the Editor does. Each question is weighed for how much answering it would change what a reader takes away, and whether an editor would insist on it; minor ones fold away, and two quiet passes in a row count as ready.

  Support cannot see which questions repeat. One shared inbox, with saved replies the whole team can edit, fixes that.

  We pilot it with our two largest accounts for six weeks.

  Co-authorEditorCritic

  Re-reading what changed...

1.2.1 (2026-09-23)

• Fixed: "Download my data" included the version history of documents you had already deleted. That history is now removed with the document, any left over from earlier deletions is cleared before an export and nightly, and the download lists each document's history under the document rather than beside it.

1.2.0 (2026-09-23)

• Live style checks. With a house style set, sentences that break one of its rules get a small dotted underline as you write: passive voice where the active would name who acts, more than three nouns in a row, a gerund where a verb would do, an uncommon word, jargon, an action turned into a noun, a sentence over the style's length, or a breach of your own custom instruction. Click into an underlined sentence to see which rule. Turn it off under Writing style in Settings.
  Simplified Technical English

  Open the valve a quarter turn. Check the gauge before you continue.

  The gauge must be checked by the operator before the pump is started.

• When a newer version of SecondPen.ai is live, the editor offers a reload once you are idle and everything is saved. Reloading brings back the same document, scroll position and cursor.

  One shared inbox, with saved replies the whole team can edit.

  We would pilot it with the two largest accounts before opening it to everyone.

  Saved212 words1.1.0

1.1.0 (2026-09-23)

• The panel now has three tabs. Co-author is the conversation: ask, draft, summarise, suggest a title. Editor is a standing list of line edits you can apply. Critic is a standing list of the hard questions a sceptical reader would ask. Both lists stay current as you write: ten seconds after you pause, what changed is re-read and new cards arrive at the bottom marked new, while the ones already on screen keep their place. A question retires on its own once the document answers it, or mark it addressed. "Keep this updated as I write" on each list turns the refresh off for that document.

  There are some really important reasons to do this. We pilot it for a quarter and then decide.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditorCritic1

  Re-reading what changed...Review now

  Line edits you can apply, kept current as you write.
• The critic now knows when to stop. Before a review runs, a separate judgement decides whether anything stands between the draft and its abstract that an editor would insist on. If nothing does, the panel says "Ready for its readers" and no edits are produced. "Review anyway" runs the full pass; in that pass, suggestions that are matters of taste fold away and only the ones an editor would insist on stay in view.

  Support cannot see which questions repeat. One shared inbox, with saved replies the whole team can edit, fixes that.

  We pilot it with our two largest accounts for six weeks.

  Co-authorEditorCritic

  Re-reading what changed...
• Two passes in a row that turn up nothing beyond minor notes, with little changed since, count as ready without asking again.
• Fixed: deleting the document you had open could leave a blank Untitled document behind, sometimes two. A new blank document is now not saved until you write something in it, and deleting the open one takes you to your most recent other document.

1.0.0 (2026-09-23)

SecondPen.ai launches: a writing tool with an AI in the margin that works toward what you are trying to write, not toward its own idea of good prose.

• The abstract. Above every document, a few sentences on what the piece is for, who will read it, and what finished looks like. Write it or have the AI draft it. Every AI action works toward it.
  AbstractA one-page case for a shared support inbox, for the ops leads. Finished when they can say yes or no to a six-week pilot.

  Start writing...

• Edits where you point. Select any passage and choose Shorten, Expand, Formal, Casual, Fix or Simplify, or type your own instruction. The result replaces the selection in place and undoes in one step. Off-topic requests are declined before they cost anything; a question is sent to the Assistant instead.

  Support answers the same forty questions every week, and each answer lives in whichever inbox happened to receive it.

  The idea is pretty simple, basically. One shared inbox, with saved replies the whole team can edit.

  Pricing for the first year still needs an owner.

  Replaced. One undo puts it back.
• Writing at the cursor. Type / on an empty line for blocks, or for Ask AI to write. When you pause at the end of a paragraph that invites more, a quiet offer to continue appears.

  One shared inbox, with saved replies the whole team can edit. We would pilot it with the two largest accounts before opening it to everyone.

• The margin. The few things most worth a look, three at most: a paragraph that does not serve the abstract, a figure that disagrees with the rest of the document, a claim without support, a hard-to-follow or wordy passage. Hover a mark for the reason, click it to review that paragraph, or step through them from the footer.

  Support answers the same forty questions every week.

  Our first help desk, in 2014, ran on a shared spreadsheet and a lot of goodwill.

  We pilot the inbox with the two largest accounts.

  Co-authorEditorCritic

  Ready at the line level

  · 1 thing to look at
• A critic that converges. Line edits and hard questions, judged against the abstract, with a verdict such as "2 things between this and ready". Each review reads only what changed since the last one, and a note you dismiss does not come back. Cards are ranked by how much they matter; the trivial ones fold away.

  There are some really important reasons to do this. We pilot it for a quarter and then decide.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditorCritic1

  Re-reading what changed...Review now

  Line edits you can apply, kept current as you write.
• The Assistant. Ask about the document, get summaries, outlines and new sections, or have it suggest a title. Answers stay in the thread; edits come back ready to replace the text.

  Support answers the same forty questions every week, and each answer lives in whichever inbox happened to receive it.

  One shared inbox, with saved replies the whole team can edit.

  Co-authorEditorCritic

  Suggest a title
• Your voice. A short note on how you write, learned from what you keep, undo and dismiss, sent with every request so the AI writes like you. Edit it, update it, switch it off or clear it in Settings.
• House style. Simplified Technical English, plain English, or your own instruction, applied to every edit. Em dashes are off unless you allow them.
• Files. Import Word, Markdown and HTML; Word documents get their headings and lists recognised. Export Markdown, HTML, Word or PDF, with a warning when structure may not convert cleanly. Version history on every document.
  Rich textMarkdown

  Proposal: a shared support inbox

  One shared inbox, with saved replies the whole team can edit. Costs in the cost sheet.

  • Pilot with the two largest accounts
  • Saved replies owned by support

  # Proposal: a shared support inbox
  
  **One** shared inbox, with saved replies the
  whole team can edit. Costs in the
  [cost sheet](/docs/costs).
  
  - Pilot with the two largest accounts
  - Saved replies owned by support

  ExportMarkdownHTMLWordPDF
• Your key or ours. A daily allowance of AI edits is included. Add your own Anthropic or OpenRouter key for unlimited use; keys stay in your browser, and can be set to disappear when it closes.