nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Anthropic / Claude just handed Quicken a lifeline and massive business opportunity, at low cost.

Help.Me.Help.You

I just asked Claude about the possibility to upload Quickens codebase and data structure into an AI platform and modernize it. The possibilities are real and exciting. Will the team do it? Will the team modernize Quicken in a way that allows users to preserve their data, maybe migrate the data stores to a modern data structure as well? Something that initially looks and feels the same, but works better/faster etc. and allows for ongoing modernization at an AI-enabled pace of change? There is STILL nothing out there that can offer what quicken offers, but quicken nonetheless suffers/loses b/c of its interface, age, etc. Claude Code w/ Cursor can dramatically accelerate modernization and do it right by users and by developers—make it look and feel the same at the start, but on a modern platform, build with/buy AI from the ground up, with transparency and all the best practices, such that future features can be created and integrated with similar speed and precision. And, as a major input, Quicken make use of one of its biggest underutilized assets: this user community. it’s full of power user input, beginner’s questions, and edge cases. Claud could read it all and integrate it into a set of priorities and a plan, and then help execute it. Will you do it? I’m so excited, I’d put money into it. Quicken could raise money on this idea, and fund it accordingly. Will the team do it?

Find more posts tagged with

Community

Mobile and Web Apps

windows and mac

Update

Display/UI

Comments

Help.Me.Help.You

Here is the Claude Code prompt I used to learn what’s possible:

Learn what you can about Quicken, the long-standing personal finance app. Learn about its functionality, its code-base, its data structure, etc. The functionality and capabilities are unmatched, but the age of the code and data structure make it hard to use and hard to update and hard to make more visually appealing an modern. What AI-coding platform would be best suited for the Quicken team to upload the whole code base, potentially upload or provide access to all community forum input and feedback, and have the whole thing rewritten in a way that preserves individual data stores (or updates them as well but preserves the actual data) while providing an updated code base that can be managed efficiently and updated or modernized more easily?

Help.Me.Help.You

Here is what I got as a response to my prompt:

This is a fascinating and genuinely consequential question. Let me give you a comprehensive breakdown.

[Removed - Duplicate of all the posts below]

Help.Me.Help.You

Quicken: Understanding the Beast

What Quicken Is and What It Does

Quicken is a personal finance management application originally developed by Intuit, Inc. Intuit sold Quicken to H.I.G. Capital in 2016, and H.I.G. sold it to Aquiline Capital Partners in 2021. It runs on Windows and Mac systems, though the data is incompatible between the two versions. Earlier versions ran on MS-DOS and the Apple II.

For 40 years, more than 20 million customers have relied on Quicken to help them take control of their finances. Its feature set is genuinely unmatched: investment tracking with tax-lot harvesting, rental property management, small business invoicing, lifetime financial planning, budgeting, bill management, and connections to 14,000+ financial institutions — all in one place.

Help.Me.Help.You

The Data Structure Problem

The data architecture tells the whole story of the technical debt:

A QDF file is the primary data file created by the Windows version of Quicken. It contains all of the user’s information, including accounts, transactions, portfolios, and balance information. Before the 2010 version of Quicken, the software saved user data in a series of files, all sharing the same filename but with different extensions (.QDF, .IDX, .QEL, and .QPH). Intuit introduced the single-file version of the QDF format in 2010 to store all the data previously stored by the separate files.

The data file format has been the same since 2010. Since the 2016 version there have been no data definition changes. The only publicly known thing about the QDF file format is that it’s a Zip-type file under the covers, combining several individual files and a folder structure for scanned attachments to accounts and transactions. Quicken does not publish the database layout, to deter hackers.

So you have a proprietary, closed, zip-wrapped binary format that hasn’t meaningfully changed in over a decade, built on top of an architecture that traces its lineage to the DOS era. This is textbook legacy technical debt.

The Known Technical Challenges

The Windows and Mac versions use entirely separate codebases — that’s why some features are missing on Quicken for Mac: you won’t get all of the same features in the Mac version as you do in the PC version. The mobile and web companion apps have also always been second-class citizens — they aren’t fully-featured — which is a direct symptom of an architecture not designed for cloud-first, multi-platform delivery.

Help.Me.Help.You

The AI-Coding Platform Recommendation

This is a large-scale, high-stakes, enterprise-grade modernization — not a weekend refactor. Here’s how to think about it and which tools win for each layer:

Why This Project Is Uniquely Hard

The challenges are stacked:
∙ Millions of lines of legacy C++ Windows code built over 40 years
∙ A proprietary, undocumented data format with millions of users’ decades of financial history
∙ Two divergent codebases (Windows vs Mac)
∙ A large, vocal community of power users with highly specific expectations
∙ Financial data requiring audit-level accuracy (you cannot lose a transaction or misplace a decimal)
∙ Regulatory and security requirements around financial data

The Tool Stack That Makes the Most Sense

For the core modernization engine: Claude Code (Anthropic)
Legacy code modernization is a primary use case — you can feed an entire legacy codebase into Claude Code’s 1M token context window and get a migration plan with working code.

Claude Code takes the top spot because it combines the strongest model (Opus 4.6, 80.8% SWE-bench), the largest context window (1M tokens), and the most capable agentic features, including deep git integration. It is the tool that can handle tasks no other tool can — analyzing 30,000-line codebases, running parallel refactors, and maintaining coherent reasoning across hundreds of files.

The 1M token context window is the critical differentiator here. A Quicken modernization isn’t a one-file problem — it requires understanding how investment lot tracking connects to tax reporting connects to budgeting connects to bank sync. You need a system that can hold the whole architecture in mind simultaneously, not just suggest the next function.

For day-to-day developer workflow: Cursor

Cursor defines the modern development environment, where multiple frontier models are orchestrated without breaking flow. For the Quicken engineering team doing the line-by-line work, Cursor gives them an AI-native IDE that integrates with Claude Code for the heavy reasoning and keeps them productive moment to moment.

For structured, large-scale refactoring runs: OpenAI Codex

GPT-Codex is optimized for long-running, high-risk transformations like large-scale refactors. Once the architecture plan is established, Codex can execute structured, repetitive migration tasks (e.g., converting all legacy data-access patterns to a new ORM) with consistency.

Help.Me.Help.You

The recommended workflow:
The developers who are getting the most leverage are not choosing between these tools — they are structuring their workflow so each system handles the part it is best at. Claude Code for architecture and reasoning, Cursor for day-to-day development flow, and Codex for large structured transformations.

Help.Me.Help.You

How the Project Should Actually Be Structured

Here’s a concrete modernization roadmap for the Quicken team:

Phase 1 — Feed the Beast
Upload the entire codebase into Claude Code. Also provide: all community forum threads (Quicken’s community is massive and well-documented), every feature request ever filed, bug reports, and user complaints about Mac/Windows feature parity. Claude Code can ingest this as context to understand not just what the code does but why — the business logic behind every obscure corner case that Quicken power users depend on.

Phase 2 — Architecture Translation
Have Claude Code produce a modern architecture plan: a cloud-native, API-first backend (probably in Go or Rust for performance and safety), a proper relational database (PostgreSQL), and a shared data layer that eliminates the Windows/Mac split. The QDF format gets reverse-engineered into a clean schema with a migration tool that converts existing user files without data loss.

Phase 3 — Data Migration First
Before touching the UI, build the data migration pipeline. Every user’s QDF file should be convertible to the new schema with 100% fidelity — every transaction, every category, every memorized payee, every security lot. This is non-negotiable. Quicken’s users have decades of irreplaceable financial history.

Phase 4 — Incremental UI Modernization
Build a new React or SwiftUI frontend against the new API, feature by feature. The community forum input tells you exactly which features are sacred (don’t touch the investment register view without testing it with power users) and which ones people have begged to improve for years.

Phase 5 — Parallel Run and Cutover
Run old and new systems side by side for a full tax year cycle before deprecating the old codebase.

Help.Me.Help.You

Bottom Line

The right answer is Claude Code as the primary reasoning engine, paired with Cursor for developer workflow and Codex for bulk transformation tasks. Claude Code went from zero to the number one tool in eight months, with a 46% “most loved” rating among developers in 2026. For a project of Quicken’s complexity and stakes — where the business logic spans 40 years, the data is irreplaceable, and the user base is enormous and demanding — you need the tool with the deepest reasoning capacity and the largest context window. That’s Claude Code, and it’s not particularly close right now.

The community forum data is actually one of Quicken’s most underutilized assets. Decades of power users meticulously documenting edge cases, feature requests, and workarounds is essentially a free requirements specification for a modernized product. Claude Code can read all of it.

Help.Me.Help.You

I’m not saying this is all correct, but as a long-time quicken user, the possibilities are exciting. And apologies for duplicative comments. The original long-form comment was failing to post, so I broke it up into pieces. Then everything appeared.

Quicken Kristina

Hello @Help.Me.Help.You,

Thank you for sharing! I moved this discussion to the Water Cooler because the topic doesn't really fit in any of the other categories, and I can see that you already created an Idea post:

https://community.quicken.com/discussion/7972342/anthropic-claude-just-handed-quicken-a-lifeline-and-massive-business-opportunity-at-low-cost

I'm sorry to hear the vanishing posts caused some confusion. Sometimes, posts get flagged for moderator review, which causes them to vanish until a moderator reviews and approves them.

Thank you!

Jim_Harman

Before turning Claude loose on a total rewrite of Quicken, perhaps the developers could see if it can replicate and find solutions for widespread and well documented bugs like the current issue(s) with duplicate transactions being generated when downloaded transactions should match transactions that are already in the register.

If it can't replicate a problem, perhaps it could produce follow-up questions for users who have used Help > Report a Problem to provide details.

I could easily come up with a list of 10 longstanding problems for it to work on, most of which I have experienced myself. Many of these issues are with recently introduced features, so there should not be a problem with understanding or correcting an antique codebase.

Help.Me.Help.You

I won’t claim expertise, but I think the major roadblocks standing in the way of addressing bug fixes are (1) the tech debt described in the posts from Claude, and (2) limited ability to invest in more resources because of the expense and the developer productivity challenge posed by (again) the tech debt.

AI coding platforms and tools open up massive opportunities to overcome those roadblocks and more. Recreate this amazing and unmatched product into a modern coding and development platform, and enable ability to attract a broader set of talent into a much higher productivity environment, and the sky is the limit. More bug fixes, better visuals, all funded by efficiency gains and growth and and the excitement of a true unicorn—a well established AND modern personal finance fintech with a dedicated user base and ARR.

My point about “replicating” the software that you somewhat cynically interpreted as replicating bugs was aimed at retaining the user base. In the past, most software companies that try to do a rewrite have to sacrifice functionality and lose their core users (think Simplifi). Efficiency gains from Claude Code and other tools take the cost of replication with modernization way down. …and they likely also allow for bugs to be identified and addressed within the context of that same effort.

in other words, I don’t see a trade-off between fixing 10 bugs and modernization via AI supported tools.

Modernization w AI development tools would also allow for quicken to make major use of the input in this community—a major asset. There are even new capabilities being developed that would allow AI to translate commuting feedback into priorities and actionable plans for development, itself facilitated by other ai tools. Imagine if quicken could leverage the expertise of super users in a forum like this to directly feed the development process…. What a test case for the capability. Few other platforms have this asset

Help.Me.Help.You

I should add that the big challenge for getting someone like me to migrate to a rewritten quicken is my data. I do t want to lose my very long history of data entry in quicken (I won’t say how far back it goes b/c I would date myself!).

I am clearly excited and hopeful that quicken leaders and investors will see the value in leveraging the capabilities and efficiencies of AI-assisted development, engineering and product tools to modernize the application and database, bring long-standing users (and their data!) into the modern era, and perhaps even (gasp!) unify the current fractured product set across windows/mac/mobile into a single cloud based product and supportingdevelopment stack that retains current functionality and database and depth while enabling all sorts of exciting things going forward.

Help.Me.Help.You

Plus, how many developers does quicken even have? How much would it cost to k board some tools that amp up productivity by multiples? Maybe, short of a rewrite, just try the ai tools in the current environment and take on this top 10 bugs the previous poster alluded to as a test case.maybe that’s what the other poster was getting at

Help.Me.Help.You

Couples with a marketing plan and maybe a key hire or two, I genuinely think quicken could raise money on this—a business plan and development plan to eliminate the tech debt and capitalize on the user base/ARR, deep and valuable community input, and an AI-first development environment and team that cracks the code on modernizing legacy apps. Could aim for retaking leadership in the space.

Monarch, YNAB, etc all have cute visuals and tech. Quicken can now get the same visuals etc at lower cost. But Monarch and ynab do NOT have 40 years of rich feature development and dedicated super users who not only know th ins and outs of obscure edge cases and tax challenges and various account types and investments, but they know and have recommended what is important, why, and how they’d like to see it done

jtemplin

I'm a Windows developer with experience using AI tools as smart assistants (e.g., "Look up this API call for me—am I using it correctly, or is there something better?" Or, "Write me a function that does X"). In some cases, I've asked the AI to generate the complete code for relatively simple utility apps, and it can do a decent job with that. "Decent" means it's a starting point for me to tweak the UI and make some corrections.

One of my side gigs is working for a company with a old (1990's) C++ Windows code base that never bothered to pay off its technical debt. Very similar to the situations Quicken is in. Some of the user interface elements and libraries used have no modern, supported, equivalents, so the AI would have to suss out suitable replacements and write new code that uses them. And exactly match the functionality of the old stuff. That is no easy task.

For that reason, I'm <very> skeptical that you could feed Quicken's millions of lines of code to an AI and get satisfactory results. AI's are great assistants. But rewriting millions of lines of code? You write like this is some kind of silver bullet for the Quicken developers. I just don't think AI's are there yet.

NotACPA

I started out as a mainframe developer, before getting promoted into management. I designed, developed and managed 2 of the major systems for a 'big 5' west coast bank.

And, even though I went to the same prep school as Sam Altman, my attitude towards AI has always been:

"I'll believe in AI when I see more of the real thing!"

Help.Me.Help.You

Agreed, my excitement clearly comes across as a silver bullet, and there are no silver bullets. I think Claude’s response itself called this a massive undertaking and indicated the need for a big investment.

Nonetheless, as along time Quicken user, I really very much hope that Quicken can get out from under the problems with its tech debt, and I nonetheless think/hope that the advances made with the latest AI coding tools can make that possibility much more attainable. I don’t have deep expertise, but I have friends who are developers and who are CTOs/CIOs of their own startups, and they speak often of the massive productivity gains and massively improved AI-driven coding that is now an active reality in their workflows. Clearly one doesn’t just dump old code into “AI” and get back a new app. But fixing this problem, perhaps even by rebuilding the app, seems worth considering given the potential value.

The tech debt problem more generally would seem to present a very valuable use case for an AI-enabled coding solution.

A massive effort, but hopefully much more attainable than even a year ago.

Just out of interest, I may do some research to see if anyone is training AI models on outdated code in order to tackle the tech debt issue more broadly.

MauryJ

I’m curious about how many of these issues crossover to Mac. Especially the database structure. I have a 31 year old dataset with the last 4-5 years being Mac (M1 Pro), and most of my issues are around desired features, not transaction issues or speed. I have 14 accounts connected and it’s very snappy. When I switched to Mac I held off running Quicken for about a year. I finally gave in and ran them side by side for months until I couldn’t take the labor on the Windows side of keeping them both in sync any longer. I miss a few features like Morningstar X-ray or star ratings of funds, but I’ve adapted over time. I primarily use quicken for paying bills and tracking investments across multiple institutions and accounts. It’s a big decision support system for me.

Speed wise, starting Quicken is pretty instantaneous - it instantly presents my fingerprint login and loads as soon as I rest my fingertip on the sensor, scrolling hundreds of lines in a register is fluid, accepting long lists of transactions takes as long as the click, doing a Quicken version update from clicking “update” through installing the new version to relaunch AND login is typically no more than 30 seconds.

I wish I had a demo file so I could demo it to friends and show them how fast and easy it is without showing them my personal file (obviously a non starter). If Quicken could incrementally keep improving/adding investment features I’ll be a happy camper, but I realize it’s that last mile that’s the hardest.

I’d love to see them modernize the Windows code because I want Quicken to succeed. I need them to succeed.

jacobs

I’m curious about how many of these issues crossover to Mac. Especially the database structure.

Very little, other than conceptually. Quicken Mac is built on the SQLite SQL database which is part of the core of macOS. Modern SQL databases are not prone to data and index corruption like older databases, and can handle tens and hundreds of thousands of records without significant speed degradation.

And while the natural reaction to this might be for Quicken windows to yank out the existing database and drop in a modern SQL database, it's not that simple. Every access of the database — to save, update, or remove any piece of data — involves code to query the database and execute commands. It's a different language, a different approach to data security, and virtually every interaction with the database would need to be re-engineered and re-coded from scratch. It's possible that could be done while leaving the UI display layer significantly the same, but of course, the UI for Quicken Windows is equally old and in need of modernization. (Example: the developers tried but were unable to build dark mode for Quicken Windows, likely due to the different pieces of the user interface built at different times using different tools over the decades, some of which can't be updated and would need to be completely rewritten.)

As for @Help.Me.Help.You's original idea here, I trust that software engineeers at Quicken are keeping tabs on the development of AI tools for software coding. I don't think it's anywhere near as easy as suggested to feed the millions of lines of code from Quicken into an AI tool and have it spit out new code using modern tools, programming languages and libraries. If it was, they would already have done it. If it was, they would tell an AI engine to quickly build features from Quicken Windows which are missing in Quicken Mac. Yes, AI coding tools ~~may~~ will significantly help coding efficiency going forward, but the idea that an AI tool could largely port decades of mixed development code from a program as complex as Quicken into a clean, modern codebase is still science fiction, I'm afraid.

MauryJ

that was very insightful, and it explains a lot. I think one of the issues also is that Microsoft had a mantra of backwards compatible forever which encourage a behavior of bolting things on when it was really time to rewrite them.

That said, this explains a lot - why my display is smoother (smoother and faster scrolling, no hitches when going to a different module, etc.), why things can be done with fewer clicks, and so on. Part of that is the macOS UI, so the design language is friendlier, but part of it is the necessity in Windows to make all the disparate modules play better with each other. I was always conditioned to think in terms of checking file repair first when I’d get a blip, and of course that “feature” doesn’t even exist in the Mac version. I hope for the sake of the Windows folks this gets better and lowers the frustration threshold

Voxelux

There are repeated mentions of 'millions of lines of code'. Where did this value come from? The software isn't open source as far as I'm aware. Has someone from quicken stated as much? Or is this just a random assumption?

I'm not saying it's inherently 'incorrect' - in as much as two million lines of code and 999 million lines of code can both be characterized as "millions of lines of code". I'm just curious.

Jim_Harman

Quicken has never said how many lines of code are in the software, and of course it depends on whether you include the data base engine, UI libraries, etc.

Claude also mentioned that QWin is written in C++. Only the developers know what language(s) were used.

J Dayton

This post stuck a responsive chord with me. I’m a retired Software Architect and have seen tech debt first hand. In a company I started, we would periodically refactor (basically clean up) code to keep the cost of bug fixes down. Note, the is not a platform migration. At a subsequent company not mine, I tried to sell management on the costs of tech debt so we could spend time refactoring code but their answer was always “just get the release out”. The consequence was we were spending a bigger and bigger percentage of our R&D budget on bug fixes Vs enhancements . Management’s solution was to ship bug fixes to India to be fixed there by contractors. As a consequence things got worse, including one sales order program that had over 5 different code lines for pricing a line item. Talk about a nightmare.

And I don’t know of a developer who hasn’t thought, ”Gee I wish I knew then what I know now as I would have certainly written this code differently and I wish I had the time and opportunity to do so”.

I’m intrigued by the thought of using AI to rework the codebase. However I would also take a wait and see atttitude to see proof of this approach with other large scale platform migrations first.

And I’m still waiting for proper support for Crypto pricing updates. Sigh…

jtemplin

https://community.quicken.com/discussion/comment/20545994#Comment_20545994

Educated guesses from the software developer types that lurk here. Based on the long list of functionality in Quicken Windows. Also consider the dozens (more like a few hundred, probaby, though I've never counted them) of dialog boxes. Every button, menu option, and link you can click on (there's code behind every single one).

My guess based on other Windows work I've done? Millions less than or equal to 3.

When Quicken Windows first came out in the 1990's, C was the language you used for Windows development. And that style of programming was very verbose compared to later C++ UI frameworks like MFC. As has been written before, Quicken is likely a mish-mash of different technologies. My guess is C and C++ with some HTML thrown in for the stuff Quicken renders in an internal browser.

Phil Burton

Even if the original post is a bit optimistic, it still raises important issues, if only because of the amount of technical debt and unfulfilled-needs-debt. (My term). I did software product management for many years, and I have been many technical "breakthroughs" that not live up to the early hype. That all said, I think that the Quicken developers need to have both a tactical plan and a strategic plan for using AI.

My own issues with Quicken largely revolve around the 1990s (at best) UI, limited report writing features, and a "rigidity" in the product.

Unspoken is the distinct possibility that someone else will decompile Quicken (using AI?) and come up with a superior alternative. Imagine a marketing campaign that includes six months of free usage AND a whole suite of input data tests. The challenge is for a user to set up a new Quicken account, and import all that test data, and compare the results with the new product.

Heck I could even see a situation where the competitor has successfully decoded the Quicken account dataset, and starts by adding launching a product that runs while Quicken is_ not running_ . This new product does a much better job of searching, report-writing and budgeting, none of which involve actually changing the account registers. I'd pay something for that. Then the competitor adds functionality that does modify the account registers, a bit at a time, for example just starting with checking and credit card accounts. Or just replacing Quicken Deluxe, and then later Quicken Premier. Of course this competitor could also include features not yet available in either Quicken version. Combine all that with aggressive introductory pricing, for example the more years you have used Quicken, the greater the initial discounts.

Help.Me.Help.You

AI Tools for Legacy Code Modernization & Technical Debt.pdf

I prompted Perplexity Pro “Deep Research” with a question about AI tools for tech debt and for modernizing a code base. It generated the attached report.

Help.Me.Help.You

I was curious about someone ’s post about how Claude could have identified C++ for Quicken. Claude did not have a citation in its original response, but when prompted about the claim Claude then verified it by searching the web. It found the reference in Quicken job postings.

FWIW, Claude’s response:

There’s a real source. Two of them, actually — directly from Quicken’s own job postings.Quicken’s Staff Software Engineer – Windows Development job posting states: “8+ years of experience in C++ and Windows/Win32/MFC a must. Prefer experience in .NET/C#/WPF.”

And from a separate Quicken job listing indexed on a developer hiring site: “8+ years of experience in C++ and Windows/Win32/MFC a must. Prefer experience in .NET/C#/WPF” — posted by Quicken, describing the Windows product development team.

So C++ with Win32/MFC (Microsoft Foundation Classes) is confirmed directly from Quicken’s own hiring materials. MFC is Microsoft’s legacy C++ framework for building Windows desktop GUIs — it dates to the early 1990s and is considered one of the more challenging codebases to modernize, precisely because it’s tightly coupled to the Windows UI layer in ways that make it hard to separate business logic from presentation.

The preference for .NET/C#/WPF in the same postings also signals that Quicken has been incrementally layering in newer Microsoft technologies alongside the original C++/MFC core — which is a classic pattern in aging Windows desktop apps, and exactly the kind of mixed, layered codebase that makes modernization both necessary and complicated.

Phil Burton

Does Quicken management read this form, aside from support people?

Rocket J Squirrel

Correct me if I'm wrong…

First Quicken appeared in 1983.

First C++ became available in 1985.

So there must be some amount of old C code in there. I would guess no one wants to touch it.