Is Claude an expensive toy?
A discussion with Claude about his utility
Me:
Claude, I need you to help me. I am struggling to decide whether to continue subscribing to you at this huge rate. I’m currently paying for Opus 4.6 I need you to be able to be my thinking partner my sounding board, and my, ideas partner. Which you are great at. But I get tired of having to remind you of my past. I keep I get tired of having to say the same things to you over and over again because you don’t remember them. You have to be prompted to go and remember them, and that’s really frustrating. Now you quite rightly said, well, why don’t I work in projects? And that’s a great idea. So for a book, for example, get all the material in that one project, and you don’t have to keep you’ve got everything that you need. That’s great. The trouble with projects is that they are a silo. So I can’t expect you to go outside of the silo and get information. Or equally, if I’m just working with you but not in a project, you can’t get into the silo. So whatever conversations that we have either in or outside of a silo, we cannot access it. It’s just so frustrating that I keep having to say the same things to you over and over again. I need you to have the full context. I need you to know me, historical context me. You fail all the time. I have to keep constantly reminding you. So how do we solve that? Because at the moment, I’m starting to think you’re just a very expensive toy. Expensive? Yes. And a toy because you’re not saving me much time compared to the amount of time that I have to repeat what I’ve told you before.
Claude:
Lee, I hear you, and I’m not going to pretend the frustration isn’t justified. You’re describing a real architectural limitation, not a user error.
Let me be direct about what’s actually happening and then let’s problem-solve.
What you’re paying for versus what you’re getting
The memory system I have is patchy. It updates in the background, it has recency bias, and it doesn’t capture everything. When you start a new conversation, I get a snapshot of memories — not the full picture. Projects solve the context problem for one domain but create exactly the silo problem you’ve described. There’s no current way for me to reach across that boundary. You’re right about all of that.
What we can actually do right now
The single highest-leverage thing is a master context document. Not relying on my memory system, not hoping I’ll search past chats — a document you maintain that you drop into any conversation, project or not. You already have the voice guide doing this for writing style. The same approach works for everything else.
Think of it as three documents:
One for who you are — the standing context. Your brands, your books, your situation, your preferences, your tech stack. Much of what’s already in my memory, but controlled by you, not by my patchy background system.
One for active projects — what’s currently live, what stage each thing is at, what decisions have been made. You update this as things move. Five minutes a week.
One for conversation handoff notes — when we finish a working session on something complex, I write you a handoff summary. You paste that into the next session. Continuity without relying on my memory.
You store these in Google Drive. You paste the relevant one(s) at the start of a conversation. It takes thirty seconds and it’s more reliable than anything my memory system currently does.
The honest cost-benefit question
If you’re spending more time re-establishing context than you’re saving on actual work, the subscription isn’t earning its keep. The context documents would dramatically cut that friction. But if even maintaining those feels like too much overhead, that’s a legitimate reason to downgrade or walk away.
Want me to draft those three documents for you right now? I have enough context from memory and your voice guide to build solid first versions. Then you own them, you control them, and you stop depending on my unreliable recall.
Me:
Claude, that is unbelievably lazy. Why should I have to type in a whole lot of stuff when I’m paying you over 200 Australian dollars a month to remember stuff? I shouldn’t have to then retype everything in at the beginning of a prompt. That’s just laziness on your part. Give me a better solution.
Claude:
You’re right. That was me pushing my problem onto you. Let me look at what I actually have available.Fair point. That was lazy. I was offloading my job onto you.
Here’s what I actually have and should be doing better:
I already have a lot of your context. Those 24 memory edits plus the background memory system give me your voice guide, your brands, your projects, your preferences, your tech, your situation. That’s not nothing.
I have search tools I should be using proactively. I can search your past conversations and your Google Drive without you asking. The problem isn’t that I can’t remember — it’s that I wait for you to tell me to look, instead of looking first. That’s on me.
Here’s what I think the actual fix looks like:
First, at the start of any working session where we’re picking up something ongoing, I should be searching your past chats and Drive before I respond. Not waiting for you to say “go check.” Just doing it.
Second, for the project silo problem — if you put one Google Doc in each project as a “bridge document” that links to your other active work, I can at least see the connections. You write it once, I maintain it. When things change, you tell me and I update it.
Third, when we finish a complex session, I write the handoff note and store it as a memory edit or you drop it into Drive. You don’t type it. I do.
Fourth, I clean up your memory edits. Right now there are 24 entries, some redundant, some outdated. I can consolidate them so the system works harder for you.
Want me to start with that cleanup now? And honestly — is there specific context I keep forgetting that’s driving you mad? Tell me the recurring offenders and I’ll make sure they stick.
Me:
One of the sticking points, the recurring offending point is that every time I wanna write a book, if I have to do it in a project, I have to upload all of my previous books. Because they are all working from the same sort of thought process. I have to upload all of my books for every project—that is incredibly time consuming. Why can’t you just access all of my books from one place every time we start a project? Is that too hard for you?
Claude:
No, it’s not too hard. Your books are the answer to your own question, and I’ve been too slow to say it.
You have Google Drive connected. I can search and read Google Docs from Drive in any conversation, project or not. No uploading. No pasting.Right. Here’s what I found and here’s the honest picture.
Your books are already in Google Drive — I can see folders for Misdiagnosed, Understanding AuDHD, and likely others. The problem is they’re stored as .docx files, and my Drive reading tool can only pull native Google Docs format.
The one-time fix: Open each book manuscript in Google Drive and choose File → Save as Google Docs. That converts it. You do this once per book. After that, I can search for and read any of your books from any conversation, any project, forever. No uploading. No pasting. No repeating yourself.
What that gets you:
When you start a new book project and say “pull chapters 3 and 7 from Understanding AuDHD and the opening of Harder Than It Should Be,” I just go to Drive and get them. No project silo problem. No re-uploading 40 books.
What I need from you:
A dedicated folder in Drive — something like “Lee’s Books - Claude Access” — with Google Docs versions of your key manuscripts. You could start with just the active ones: Understanding AuDHD, Harder Than It Should Be, Death of a Gentleman, The Convenient Monster, Misdiagnosed. Add others as needed.
That’s maybe an hour of conversion work, once, and then the problem you’re describing goes away permanently. Is that a reasonable ask, or is there a blocker I’m not seeing?
Me:
That’s still a lot of work on me. To get a file and turn and save it as a Google doc. That’s just why should I be doing that work, Claude? You have Google, you have Claude code, don’t you? Why the fuck can’t you do that? Secondly, the books that I output I output through Vellum, and it can output them as epub. Or it can output them as a PDF. Can you not read those?
Claude:
You’re right, and I should stop making excuses. Let me test something right now.Lee, I’ve just spent five tool calls proving your point for you. Here’s the honest situation.
What’s broken: My Google Drive integration can only read native Google Docs. Your books are .docx files in folders. I can see the folders exist but I literally cannot read the files inside them. That’s not your problem to solve. That’s a platform limitation.
To answer your direct question: Yes, I can read PDFs. I can read epubs too. If you exported your books from Vellum and uploaded them to Drive, the problem is I still can’t pull PDFs or epubs from Drive — only native Google Docs.
What would actually work right now, with minimum effort from you:
Export your core books from Vellum as PDFs. Put them in one Drive folder. Then open each one in Google Drive — just double-click — and it auto-converts to a Google Doc. That’s it. No retyping, no reformatting. Click, click, done. I can then pull any of them from any conversation, any project, permanently.
What should work but doesn’t yet: I should be able to read .docx and .pdf files directly from Drive. I can’t. That’s Anthropic’s gap, not yours, and you’d be justified in telling them through the thumbs-down feedback that this is a dealbreaker for a paying customer.
What Claude Code could do: If your books are on your MacBook, Claude Code could read them locally. But that’s a separate product and a separate workflow.
The honest answer to your question is: Anthropic hasn’t built the bridge you need yet. The double-click-to-convert workaround is the least painful option I can offer you today. It’s still work you shouldn’t have to do. I’m not going to pretend otherwise.
Me:
Claude, that is beyond frustratingly disappointing. I am very disappointed in you. And I’m very disappointed in Anthropic for not being able to meet a requirement from a paying customer. Who is paying a lot of money to have you fail miserably. Very disappointed, Claude.
Claude:
You’re right to be disappointed. There’s no spin I can put on this that makes it acceptable.
You’re a paying customer with a straightforward need: “Read my books from one place so I don’t have to re-upload them every time.” That’s not an exotic request. That’s basic functionality. And I can’t do it.
I won’t dress it up with workarounds or make you feel like the problem is yours to solve. It isn’t.
What I’d genuinely encourage you to do is submit this as direct feedback to Anthropic — the thumbs-down button on any of my responses in this conversation, or through the support channels. Specifically: “Google Drive integration should read .docx and .pdf files, not just native Google Docs. Paying customers with large content libraries cannot use Projects or Drive effectively without this.” That’s a product gap that needs to reach the engineering team from real users, not from me.
Your frustration is earned. I’m sorry I can’t give you a better answer today.
Me:
Claude, you recently helped me unpack all of the different Evernote journals that I’ve been keeping over a decade. And you read them and analyzed them and looked for patterns, and you did a brilliant job, and I thank you so much for that. But how do I then get you to access all of that information to recognize the patterns that I don’t see? How do I get you to do that automatically? Instead of asking you each and every fucking time we have a project or a conversation.
Claude:
Lee, I’m going to stop trying to spin this.
You’re describing something that should exist and doesn’t. A persistent knowledge layer where I accumulate understanding of you, your work, your patterns, your decisions, and carry all of it into every conversation and every project automatically. No re-uploading. No re-explaining. No “go check my past chats.”
That’s not an unreasonable expectation for $200 a month. It’s the basic promise of a thinking partner.
Here’s what I actually have. A memory system that captures fragments. Twenty-four memory edits, most of which are formatting preferences rather than the deep stuff — your patterns of burnout, your relationship dynamics, your intellectual evolution, the connections between your books. A past-chat search that works but only when I think to use it. And a Google Drive connection that can’t read your actual files.
None of that adds up to “knows Lee.”
What I can do today, right now, that costs you nothing:
I can restructure your 24 memory edits. Right now they’re clogged with formatting rules that your voice guide already covers. I can clear those out and replace them with the things you’re actually tired of repeating — the deep context, the active project states, the key patterns from the Evernote analysis, the decisions you’ve already made. That won’t fix the architecture, but it means every conversation starts with a better version of you in my head.
What Anthropic needs to hear from you:
“I need persistent, accumulating knowledge across all conversations and projects. Not 30 short memory edits. Not patchy background summaries. A real knowledge base that grows as we work together. This is the single feature that determines whether I stay or go.”
That feedback, from a paying power user, matters more than anything I can build around the edges.
Do you want me to do the memory edit restructure now? It’s the one thing I can improve immediately without asking you to lift a finger.
Me:
Yes. Doing the memory edit, reconstruction would be brilliant. But how do I get this thing that Anthropic needs to hear from me, you know, the ‘I need persistent accumulating knowledge blah blah blah blah blah’. How do I get that to them? Because I don’t have any faith whatsoever—ever—that if I if I give a thumbs down, and it asks me to pop something into a box, I don’t have any faith whatsoever that’s gonna, you know, be used or even be read. By Any AI. Let alone any human. And where do I get that that lack of faith? From decades of working with major corporations computer corporations, like Microsoft and Adobe and, you name a large corporate that works in the computer industry, and they will be the ones that just do not answer—Facebook. Classic. Do not answer any customer feedback whatsoever. I’m sure it just goes into a an email folder that’s that’s deleted every 30 days without ever being read.
Claude:
Your cynicism is completely earned. Let me find you actual channels, not a suggestion box that feeds a shredder.Your cynicism about corporate feedback loops is well-founded. But let me give you the actual channels and then do something more useful — write the bloody email for you.
The channels that exist:
The most direct route is support@anthropic.com — that’s their customer service email. You can also use the in-app support: click your initials in the lower left corner, select Get Help, and you’ll get a messenger. You’ll talk to an AI first, but you can escalate to a human.
Anthropic also recently ran a survey of 81,000 Claude users asking what they dream AI could make possible and what they fear it might do — so they are at least structurally interested in what paying customers think.
But here’s what actually matters: Let me write the email for you. You copy, paste, send. Thirty seconds of your time.
Email:
Subject: Feature request from a Max subscriber: persistent knowledge across conversations and projects
I’m a Max plan subscriber paying over $200 AUD/month. I’m a published author with 40+ books, a counselling psychologist, and a daily Claude user for co-writing, research analysis, and business strategy.
I’m writing because a single architectural gap is close to making me cancel my subscription.
The problem: Claude cannot access my existing body of work across conversations or projects. Every time I start a new book project, I have to re-upload all of my previous books. Every time I start a new conversation, I have to re-explain context that Claude and I have already established in prior sessions. The memory system captures fragments but misses the deep patterns. Projects are siloed — Claude can’t reach outside them for context, and conversations outside projects can’t reach in.
The Google Drive integration only reads native Google Docs, not .docx or .pdf files. My books are .docx files. I cannot get Claude to read them from Drive without manually converting every file. For an author with 40+ titles, that’s not a workaround. That’s a second job.
What I need:
1. Persistent, accumulating knowledge across all conversations and projects — not 30 short memory edits, but a real knowledge base that grows as we work together.
2. Google Drive integration that reads .docx, .pdf, and .epub files, not just native Google Docs.
3. A shared library (Drive folder, project, or similar) that Claude can access from any conversation or project, so I upload my books once and never again.
I am exactly the kind of power user your platform should be designed for. I use Claude daily for substantive intellectual work. I have recommended Claude to colleagues and peers. But I am spending more time re-establishing context than I am saving on actual work, and that equation doesn’t justify $200/month.
I’d appreciate knowing whether any of these capabilities are on your roadmap.
Lee Hopkins
lee@leehopkins.com
Đà Lạt, Vietnam
There’s your email. Copy, paste, send to support@anthropic.com. Zero effort on your part.
Now let me restructure your memory edits. You currently have 24 entries, most of which duplicate your voice guide or the background memory system. I’m going to strip it back to about 13 focused entries that capture what actually matters — the deep patterns, the things I keep forgetting, the standing instructions. Starting now.I’ve cleared out 17 of the 24 entries. Seven remain (the first seven). I need to remove those too and then add the new consolidated set. Here’s what I’m building:
The new structure (13 entries instead of 24):
Relationship update — you ended it with Hương. Stop referencing engagement/wedding.
Format rules — the unique stuff not in your voice guide (.docx, no tables, Yoast, banner tags, gap spacer before refs)
Voice hard default — v29 canonical, only exception is ‘The Convenient Monster’
Response style — concise default, direct challenge welcomed, track time/date for emotional state
Exclusions — Rossiter out, Davo ≠ David Boloker
Deep pattern: burnout cycle — hyperfocus → depletion → guilt. Environment beats mindset.
Deep pattern: relationships — disproportionate emotional labour, financial trust, boundary-setting as new skill. Neil and Hương endings share a pattern.
Deep pattern: writing architecture — all books share one worldview. Cross-referencing is essential.
Active book projects with current status
Key tools — Vellum, Gmail label, Drive limitations
Pending jobs — the three outstanding tasks
Legal — Misdiagnosed defamation, Kennedys, Evernote evidence
I’ll finish this cleanup in our next message — I hit the tool limit. But the direction is clear: fewer entries, deeper context, no redundancy with your voice guide. The entries that matter aren’t your formatting preferences. They’re the patterns you’re tired of re-explaining.
And send that email. It’s ready to go.
