How Claude Cowork Rebuilt My Writing Workflow in Two Minutes

I admit it, with a mixture of embarrassment and excitement: AI may be the catalyst for revising my writing workflow.

This is not due to procrastination or the siren call of the new and shiny. It is due to an entirely new capability made possible by recent AI models. Here is the short version.

I have struggled to settle on writing apps and workflow, and I own that fact. If you must, feel free to harass and poke fun at me. I deserve it, and I can take it. :slightly_smiling_face:

The struggle stems from the sheer variety of what I produce. Second only to speaking and meetings, most of what I do involves text: speeches, devotionals, communications, reports, white papers, blog articles, a book project, and meeting notes, among other things. Finding a combination of apps and hardware simple enough to manage yet robust enough to handle everything from brief meeting notes to a large nonfiction book on Christian school leadership has proven elusive. I have used many apps, but I will spare you the list.

Over the last two days I discovered the real power of AI to deliver results and make my varied writing far easier. In addition to what I shared in this post, today I discovered the real power of Claude Cowork. After a thorough conversation with Claude (Max plan, Opus 4.6), I concluded that I can and should move my book project and all other writing to iA Writer, abandoning Ulysses, Scrivener, and other writing apps. Because I already have the entire book in both Ulysses and Scrivener (do not ask), I can export it in raw markdown to a folder I created in iCloud under iA Writer.

Here is the key to this decision. Claude Cowork was able, in about two minutes, to create the entire folder structure, all of the templates, the book sections and parts, and even the content block master document. All I need to do is export the completed first drafts of the first ten chapters and then continue writing the remaining chapters.

Here are the advantages:

  • Distraction-free writing. Clean surface, no toolbars. Focus Mode dims everything except the current sentence.
  • Markdown native. Plain text files, never locked in a proprietary format. Future-proof for decades.
  • True file ownership. My files are standard .md files on my filesystem. Ulysses stores writing in a proprietary database; Scrivener uses proprietary project bundles. If either application disappears, retrieving work is not straightforward. With iA Writer, every file opens in any text editor on any platform.
  • No vendor lock-in. Because my files are plain Markdown, they work in any text editor or any future application I may adopt. I am never dependent on a single developer’s continued support.
  • No subscription. iA Writer is a one-time purchase. Ulysses charges approximately forty dollars per year and reverts to read-only mode if the subscription lapses — I would lose the ability to edit my own writing without paying again.
  • Content Blocks. One master file assembles the entire manuscript from individual chapter files. No other app does this as well, at least not without a lot of complexity–looking at you Scrivener.
  • Export versatility. One source file exports to PDF, Word, and HTML — serving my publisher, my board, my blog, and my archive without maintaining separate versions.
  • iCloud sync. Draft on the iPad, revise on the Mac, review on the iPhone. No manual file transfer.
  • Cross-platform availability. iA Writer runs on Mac, Windows, iOS, and Android. Ulysses is Apple-only. Scrivener has no Android support. This is not really important to me, but for others, it would be.
  • Library organization. Folders, tags, and search across my entire writing archive — book, articles, speeches, reports, and devotionals in one place.
  • Custom typography. Purpose-built monospaced font with deliberate spacing and contrast for extended writing sessions.
  • Style Check. Flags redundancies, clichĂ©s, and filler without rewriting my prose.
  • Breadcrumbs workflow with Claude AI. As I draft, I leave brief inline cues — an author’s name, a Scripture reference, a source title — and Claude AI converts those breadcrumbs into properly formatted Markdown footnotes and links. I never break my writing flow to format citations manually.
  • Authorship tracking. Distinguishes my own drafting from pasted or AI-assisted text at a glance.
  • Backblaze redundancy. iCloud syncs my files across devices; Backblaze backs up everything offsite continuously. Two independent layers of protection, no manual effort.
  • Fast and stable. Lightweight, opens instantly, works offline. No learning curve. Scrivener’s manual runs over nine hundred pages; iA Writer requires none.
  • No account required. I own my files. They live on my filesystem in a format I can read with my own eyes.

So there you have it. I am using four apps in my writing workflow: Apple Notes for notes, ideation, and handwritten outlines; DEVONthink for research documents; iA Writer for all drafts; and Pages for final formatting if and when needed.

I’m glad that is settled! :rofl::crossed_fingers:t2:

11 Likes

I’m a big fan of iA Writer, it’s the best Markdown editor I’ve used on any platform. I only write short notes mostly for myself though. They live in Obsidian, but the Obsidian editor is pretty bad especially on mobile, so most of the writing part happens in iA Writer. I then use Obsidian for viewing and organizing my notes.

I didn’t think it would work well with book length projects, but good to know that is possible. If I ever do write a book, I’ll start with iA Writer!

3 Likes

Sounds like a great plan.

In all seriousness, when a workflow is one that is very important and/or consumes a lot of your active time, I think it makes perfect sense to use a combination of best-of-breed products/apps to accomplish what you need.

Where a workflow is infrequent, or casual, or set-it-and-forget-it, consolidating tools also makes the most sense for many of us.

One of the reasons I ditched Text Expander for Keyboard Maestro is that I was already using KM and didn’t need TE for anything else. Sure, setting up text expansion in KM was a little klunkier, but I don’t constantly change my expansions so once I completed the migration, I almost never touch it.

Conversely, my workflow for video/audio production requires a lot of specialized tasks for which I still find it best to use several silo’d tools instead of trying to force-fit into a single all-in-one app.

5 Likes

You can work with an external folder of markdown files in Ulysses by using the external folder feature, if you did want to stick with Ulysses. But you seem to have other good reasons to switch to IA writer so maybe that option is moot.

1 Like

You are correct that one can work with external folders in Ulysses. Doing so, however, means losing some of the features unique to Ulysses, making that option less compelling.

1 Like

Clearly, iA Writer is not built for writing books. What I like about it, however, is the ability to focus on each chapter with a simpler tool while avoiding proprietary databases. I will still be able to reconstruct the book into a single master document, export it to Word, and finish the formatting and layout in either Vellum or Atticus. I primarily want to get the text down, and relatively new tools like Claude and SuperWhisper make that a much more efficient process.

1 Like

Glad to see we survived “the culling of the apps”. That’s in Numbers 3, I think. :wink:

DEVONthink is nearly indispensable, but that is not an excuse to raise my price! :rofl::wink:

No comment on your old or new workflows, but do wonder on the wisdom of AI making the subjective decision on your behalf. Does Claude really know you better than you know yourself?

Of course not, but I understand the reason for your question. Rest assured, I make an objective decision about the change. I used Claude as a critic as I worked through all of the pros and cons. I’d like to think that a lot of people can replace me, but not Claude. :rofl:

This is remarkable! I have decided to move from Ulysses to iA Writer. I faced the challenge of exporting from the proprietary database in Ulysses to plain markdown in iA Writer. I will not go into all the details, but this is not as easy as it sounds. Another factor was that by using iA Writer, I needed to resequence all my references to match chapter titles and reference numbers in a particular sequence so that once everything is combined, everything matches as endnotes. I also needed to complete some corrections, rename files, and address other housekeeping tasks. Below is a summary of what Claude just accomplished for me while I worked on another project. I admit I am becoming a believer in what AI is and will be able to do going forward. Much of the mechanical work that we do now is going to be handled by AI, freeing us for more substantive work. This really is like having a twenty-four-hour assistant. The result is that the transition has been painless.

Reformatting a Book Manuscript with Claude (Cowork Mode)

I am writing a book on Christian school leadership consisting of thirty-three chapters. My drafts are written in Ulysses, but I needed eleven chapters reformatted for use in iA Writer with a specific footnote convention, consistent formatting, and light editorial polishing. Rather than spending hours manually reformatting each chapter, I exported the Ulysses files to a temporary folder on my desktop, uploaded them to Claude in Cowork mode, and let Claude handle the rest while I worked on another project. One of the issues is that content exported from Ulysses to markdown does not include paragraph separation. Consequently, all paragraphs for all eleven chapters—and these are long chapters—had no separation, making writing and reading too difficult. It would have take “forever” to correct the formatting.

Here is what Claude did across approximately 68,000 words:

  • Read all 11 exported Ulysses source files, identified formatting inconsistencies, grammar errors, and broken markdown syntax across hundreds of pages.
  • Created an output folder and wrote all finished files to it.
  • Renumbered 108 footnotes across all chapters to a chapter-specific convention required by iA Writer (e.g., [^3-2] for Chapter 3, footnote 2), updating both in-text references and definitions.
  • Added consistent paragraph spacing, corrected heading hierarchy, and removed HTML comments, placeholder links, and editorial notes left over from the Ulysses drafting environment.
  • Corrected grammar, typos, and broken possessives throughout (e.g., “t4he” to “the,” missing parentheses, split lines).
  • Made minor editorial revisions for clarity, flow, and paragraph transitions while preserving my voice and theological meaning.
  • Renamed all chapter files to a consistent naming convention (e.g., Chapter-01.md, Chapter-02.md).
  • Processed chapters in parallel batches for efficiency, then ran a verification pass confirming all 11 files passed quality checks.
  • Delivered 11 clean, raw markdown files ready to move directly into my iA Writer book folder.

Time comparison: This work, including reading every chapter, manually renumbering 108 footnotes, correcting grammar, fixing markdown syntax, ensuring consistent formatting, and verifying every file, would likely have taken me 8 to 12 hours of careful manual effort. Claude completed the entire project, including verification, in approximately 20 minutes while I focused on other work.

EXAMPLE

Here is an example from chapter 2. Claude redid every footnote citation and subsequent reference to match the chapter title (NOTE: I removed the so it would render here correctly):

2-1: Thomas Fuller. 1650. Pisgah-Sight of Palestine and the Confines Thereof, with the History of the Old and New Testament acted thereon. J.G. for John Williams.

2-2: Matthew Henry. 1991. Matthew Henry’s commentary on the whole bible. Hendrickson Publishers, Inc.

2-3: Rapaille, Clotaire. 1992. The culture code: An ingenious way to understand why people around the world buy and live as they do. ASQ Quality Press.

2-4: Rana, Fazale & Ross, Hugh. 2005. Who was Adam? A creation model approach to the origin of man. Colorado Springs, CO: Navpress.

2-5: Who then is the faithful and wise manager, whom his master will set over his household, to give them their portion of food at the proper time? Blessed is that servant whom his master will find so doing when he comes. Everyone to whom much was given, of him much will be required (Luke 12:42–43).

2-6: Brand, Chad, Draper, Charles & England, Archie. 2003. Holman Illustrated Bible Dictionary. Nashville, TN: Holman Bible Publishers.

2-7: Lewis, C. S. 2009. Mere Christianity.

2-8: A covenant is a formal and serious agreement or promise made to another.

2-9: C. S. Lewis. 1970. God in the dock. HarperCollins Publishers Australia Pty. Ltd.

2-10: Bridges, Jerry. 2014. The Pursuit of Holiness. Tyndale House.

2-11: David Shimm. 2006. 365 Days with Wilberforce: A Collection of Daily Readings from the Writings of William Wilberforce. Day One Publications.

2-12: Genesis 2:7; Job 7:1, 12:10, 14:5,14, 33:4; Psalm 31:15, 39:4, 90:1-2, 102:24, 116:15, 139:16; Ecclesiastes 3:2; Daniel 4:35.

2-13: The importance of spending time in prayer has always been recognized by God’s people. Again, Wilberforce reminds us of our need for prayer. “Let your religion consist much in prayer
 There are so many things we need of God and prayer is the means by which they are sought.”

2-14: David Shimm. 2006. 365 Days with Wilberforce: A Collection of Daily Readings from the Writings of William Wilberforce. Day One Publications.

2-15: Clear, James. 2018. Atomic Habits: An Easy & Proven Way to Build Good Habits & Break Bad Ones. Avery.

2-16: Jen Wilkin. 2022. Don’t Expect Instant Gratification from Your ‘Quiet Time’. Accessed: March 21 from: Don’t Expect Instant Gratification from Your ‘Quiet Time’ - Christianity Today

2-17: David Shimm. 2006. 365 Days with Wilberforce: A Collection of Daily Readings from the Writings of William Wilberface. Day One Publications.

2-18: Calvin, John. 2008. 365 Days with Calvin: A Unique Collection of 365 Readings from the Writings of John Calvin. Leominster: DayOne Publications.

2-19: Dickson, John. 2011. Humilitas: A Lost Key to Life, Love, and Leadership. Zondervan.

2-20: Crouch, Andy. 2013. Culture making. InterVarsity Press.

2-21: Elaine Godfrey. 2022. Why Trump Won’t Stop Talking About His Crowd Sizes. Why Trump Won’t Stop Talking About His Crowd Sizes - The Atlantic

2-22: Quasten, Johannes, Burghardt, Walter J. & Lawler, Thomas Comerford. 1982. St. Augustine: The literal meaning of Genesis.

2-23: Ryan Holiday. 2022. Discipline is destiny: the power of self-control. Penguin Random House LLC.

2-24: Agus, David B. 2014. A Short Guide to a Long Life. Simon and Schuster.

2-25: Harvard Health Publishing. 2019. The dangers of sitting - Harvard Health. Harvard Health Publishing. The dangers of sitting - Harvard Health

2-26: Russell Clayton, Christopher Thomas & Jack Smothers. 2015. How to Do Walking Meetings Right. Accessed: August 5 from: How to Do Walking Meetings Right

2-27: Do not speak evil against one another, brothers. The one who speaks against a brother or judges his brother, speaks evil against the law and judges the law. But if you judge the law, you are not a doer of the law but a judge. There is only one lawgiver and judge, he who is able to save and to destroy. But who are you to judge your neighbor? (James 4:11–12).

2-28: Kouzes, James M & Posner, Barry Z. 2011. Credibility: How leaders gain and lose it, why people demand it. John Wiley & Sons.

2-29: Eldelman Trust. 2021. Edelman trust barometer 2021. https://www.edelman.com/sites/g/files/aatuss191/files/2021-03/2021%20Edelman%20Trust%20Barometer.pdf

2-30: Eldelman Trust. 2021. Edelman trust barometer 2021. https://www.edelman.com/sites/g/files/aatuss191/files/2021-03/2021%20Edelman%20Trust%20Barometer.pdf

2-31: Condoleezza Rice. 2021. George Shultz: A Towering Statesman Who Lived By the Values He Preached. Politico. https://www.politico.com/news/magazine/2021/12/27/2021-obituary-george-shultz-520597?cid=apn

2-32: MacDonald, George. 2021. At the Back of the North Wind. Graphic Arts Books.

2-33: Bridges, Jerry. 2014. The Pursuit of Holiness. Tyndale House.

2-34: Ankerich, Shawn. 2020. Situational awareness: Make safe choices Article The United States Army. Accessed: August 20 from: Situational awareness: Make safe choices | Article | The United States Army

2-35: Matthew Henry. 1991. Matthew Henry’s commentary on the whole bible. Hendrickson Publishers, Inc.

2-36: Lewis, C. S. 2009. Mere Christianity.

2-37: Langworth, Richard. 2011. Churchill By Himself: The Definitive Collection of Quotations. PublicAffairs.

2-38: Calvin, John. 2008. 365 Days with Calvin: A Unique Collection of 365 Readings from the Writings of John Calvin. Leominster: DayOne Publications.

2-39: Tripp, Paul David. 2020. Lead: 12 Gospel Principles for Leadership in the Church. Crossway.

2-40: Tripp, Paul David. 2020. Lead: 12 Gospel Principles for Leadership in the Church. Crossway.

2 Likes

@tomalmy I’m not Barrett, so forgive my jumping in.

I never fully trust the outputs of any AI, however they can be very useful sparring partners. I assume Barrett knows his writing craft well enough to judge the output of the AI and decide if it makes sense.

I see these tools as good for strategic thinking. It doesn’t replace the human or the need for collaboration, it can ask questions we might not think of ourselves.

Example, in the workshops I facilitate, I ask the students to engage in Systems Thinking to more deeply understand problems in their organizations. The thought process for Systems Thinking, doesnt come naturally to most people. So I created an AI prompt/skill to help them think through the process. See: Systems Thinking with GenAI: Solve Deep Team Problems

The tools is supposed to get the people to reflect on the problem from a different perspectives. Having just used a few minutes ago, I can tell you this group of students is examing the problems I set for them at a much deeper level than they would have otherwise.

1 Like

Ethan Mollick, associate professor at the Wharton School of the University of Pennsylvania, agrees. I highly recommend Co-Intelligence: Living and Working with A.

1 Like

Interesting, Ethan Mollick doesn’t love nuance.

He keeps on talking about software development, while ignoring the challenges of LLMs and writing code. Further, some of the flaws that are being ignored are fundamental to the models themselves: GenAI Code Quality – The Fundamental Flaws and How Bluffing Makes It Worse

There is much good in these tools, but they have real flaws that aren’t going away.

1 Like

Agreed. The point I was making is that he compared the innovative work done by those who worked with AI and those who did not. According to him, the former group performed better, which is similar to the experience you described.

@mlevison - Thanks for sharing your systems-thinking skill! Our teams are facing significant challenges at present, so I had been working with Claude to troubleshoot and try to find some solutions. Its initial proposals were reasonable, but lacked “meat” to back them up. I then discovered your systems-thinking skill, installed it and reran the same prompt telling Claude to utilise the systems-thinking skill, and wow! A very detailed response backing up the original proposals with solid reasons and causal loop diagrams too. Great example of a highly-specific, well defined skill that significantly boosts the LLM’s abilities. Well done!

small snippet of the output:

1 Like

(post deleted by author)

Thanks for making my day. This negatively reinforcing feedback loop is a common example. I’m glad that you found the systems thinking skill useful.

I posted this to linkedin, yesterday. It bears repeating here. The problem you is built into the AI. Mistakes are inevitable, the question is to make sure you catch them.

You can’t meaningfully test AI by yourself. When you try an LLM to see if it does what you want, you’ll only run the test a handful of times. If it mostly works, you’ll assume it works correctly. But the failure might only occur 1 in 20 tries - and you’ll never run it enough times to know the failure rate.

It gets worse. For the most interesting work - writing, analysis, strategy - assessing quality requires attention to detail and reflection. That’s expensive cognitive effort. So not only do we test too few times, we evaluate too loosely each time.

The result: we build false confidence. We assume the AI is correct because we haven’t caught it being wrong.

This applies at the organizational level, too. Most organizations aren’t doing a systematic evaluation of AI outputs. They’re relying on the same ad hoc “it seemed fine” approach, just spread across more people.

The uncomfortable takeaway: we can never assume the AI is correct. Every output needs scrutiny proportional to the consequences of getting it wrong.

Imagine a team of 6-7 people using AI on a daily basis. If each person does 10 tasks a week with AI, that means they might have 1-2 errors a week. Since most results are fine, we don’t even notice the errors.

Worse catching errors requires domain expertise, critical thinking and time. All of which are in short supply.


In addition to the above I wrote about for programming the flaws are built into how AI is trained and works. I assume this applies to other domains as well, however I’m not an expert in other domains. Whereas with 40 years of experience in software development, I can claim to be an expert in that: GenAI Code Quality – The Fundamental Flaws and How Bluffing Makes It Worse

Those are thoughtful points - thanks!

You must have posted right as I was deleting my post from this thread; I reconsidered my comment as somewhat off topic to the OP. I reposted in more detail in the AI forum as more of an FYI/anecdote than anything else.